| 1 |
<?php |
| 2 |
/** |
| 3 |
* Content Brief Generator |
| 4 |
* |
| 5 |
* Handles AI-powered content brief generation with competitor analysis |
| 6 |
* |
| 7 |
* @package ThinkRank |
| 8 |
* @subpackage AI |
| 9 |
* @since 1.0.0 |
| 10 |
*/ |
| 11 |
|
| 12 |
namespace ThinkRank\AI; |
| 13 |
|
| 14 |
use ThinkRank\Core\Settings; |
| 15 |
use ThinkRank\AI\OpenAI_Client; |
| 16 |
use ThinkRank\AI\Claude_Client; |
| 17 |
use ThinkRank\AI\OpenRouter_Client; |
| 18 |
|
| 19 |
// Prevent direct access |
| 20 |
if (!defined('ABSPATH')) { |
| 21 |
exit; |
| 22 |
} |
| 23 |
|
| 24 |
/** |
| 25 |
* Content Brief Generator class |
| 26 |
*/ |
| 27 |
class Content_Brief_Generator { |
| 28 |
|
| 29 |
/** |
| 30 |
* AI request timeout (seconds) for brief generation. |
| 31 |
* |
| 32 |
* Content briefs request a very large completion (~0.9–0.95 of the model's |
| 33 |
* max tokens) from reasoning models, which routinely take 40–90s — far |
| 34 |
* longer than the AI clients' 30s default. Without this the HTTP call is |
| 35 |
* aborted with cURL error 28 and the brief never completes. PHP execution |
| 36 |
* time is covered by each client's raise_request_time_limit() (timeout+45). |
| 37 |
*/ |
| 38 |
private const AI_REQUEST_TIMEOUT = 120; |
| 39 |
|
| 40 |
/** |
| 41 |
* AI request timeout (seconds) for OpenAI specifically. |
| 42 |
* |
| 43 |
* OpenAI's reasoning models (GPT-5 / o-series) burn reasoning tokens before |
| 44 |
* emitting any content, and briefs request ~95% of the model's completion |
| 45 |
* limit — so the call frequently runs past the 120s the other providers |
| 46 |
* need. PHP execution time is covered by raise_request_time_limit() |
| 47 |
* (timeout+45); the web server's own read timeout still caps the maximum. |
| 48 |
*/ |
| 49 |
private const OPENAI_REQUEST_TIMEOUT = 300; |
| 50 |
|
| 51 |
/** |
| 52 |
* Minimum output-token budget for a content brief. |
| 53 |
* |
| 54 |
* A shorter length tier must never starve the structured JSON — plus any |
| 55 |
* reasoning/thinking tokens, which are drawn from the same budget — to the |
| 56 |
* point of truncating mid-response (the failure #165 fixed on Gemini). This |
| 57 |
* floor is only a safety net for small-ceiling models; it never exceeds the |
| 58 |
* model-aware base budget. See scale_tokens_for_length(). |
| 59 |
*/ |
| 60 |
private const MIN_BRIEF_TOKENS = 2000; |
| 61 |
|
| 62 |
/** |
| 63 |
* Content-length → budget multipliers, applied to the model-aware base. |
| 64 |
* |
| 65 |
* NOTE: provisional starting points (issue #287). They make Short/Medium/ |
| 66 |
* Long request measurably different budgets, but the exact figures should |
| 67 |
* be validated against recorded completion-token usage for a real brief on |
| 68 |
* each provider before being treated as final. Unknown lengths fall back to |
| 69 |
* the 'medium' tier (see scale_tokens_for_length()). |
| 70 |
*/ |
| 71 |
private const LENGTH_TOKEN_MULTIPLIERS = [ |
| 72 |
'short' => 0.6, |
| 73 |
'medium' => 0.8, |
| 74 |
'long' => 1.0, |
| 75 |
]; |
| 76 |
|
| 77 |
/** |
| 78 |
* Settings instance |
| 79 |
* |
| 80 |
* @var Settings |
| 81 |
*/ |
| 82 |
private Settings $settings; |
| 83 |
|
| 84 |
/** |
| 85 |
* AI client instance |
| 86 |
* |
| 87 |
* Null when the generator was built for storage-only work. |
| 88 |
* |
| 89 |
* @var OpenAI_Client|Claude_Client|null |
| 90 |
*/ |
| 91 |
private $ai_client; |
| 92 |
|
| 93 |
/** |
| 94 |
* Constructor |
| 95 |
* |
| 96 |
* @param Settings|null $settings Settings instance |
| 97 |
* @param OpenAI_Client|Claude_Client|null $ai_client AI client instance |
| 98 |
* @param bool $require_ai_client Whether a provider client is required. Pass |
| 99 |
* false for storage-only use (list/export/ |
| 100 |
* delete), which never calls a provider. |
| 101 |
*/ |
| 102 |
public function __construct(?Settings $settings = null, $ai_client = null, bool $require_ai_client = true) { |
| 103 |
$this->settings = $settings ?? Settings::instance(); |
| 104 |
|
| 105 |
if ($ai_client) { |
| 106 |
$this->ai_client = $ai_client; |
| 107 |
|
| 108 |
return; |
| 109 |
} |
| 110 |
|
| 111 |
// Read-only callers (listing, exporting and deleting saved briefs) only |
| 112 |
// touch the database and never reach a provider. Constructing a client |
| 113 |
// for them turns "no API key configured" — the default state of a fresh |
| 114 |
// install — into a hard failure, so let them opt out. |
| 115 |
if (!$require_ai_client) { |
| 116 |
return; |
| 117 |
} |
| 118 |
|
| 119 |
// Fallback to creating own client for backward compatibility |
| 120 |
$this->init_ai_client(); |
| 121 |
} |
| 122 |
|
| 123 |
/** |
| 124 |
* Initialize AI client based on available API keys |
| 125 |
* |
| 126 |
* @return void |
| 127 |
* |
| 128 |
* @throws \Exception On failure. |
| 129 |
*/ |
| 130 |
private function init_ai_client(): void { |
| 131 |
$provider = $this->settings->get('ai_provider', Settings::AI_PROVIDER_NONE); |
| 132 |
|
| 133 |
if ($provider === 'openai') { |
| 134 |
$api_key = $this->settings->get('openai_api_key'); |
| 135 |
if ($api_key) { |
| 136 |
$model = $this->settings->get('openai_model', Settings::DEFAULT_OPENAI_MODEL); |
| 137 |
$this->ai_client = new OpenAI_Client($api_key, $model, self::OPENAI_REQUEST_TIMEOUT); |
| 138 |
} |
| 139 |
} elseif ($provider === 'claude') { |
| 140 |
$api_key = $this->settings->get('claude_api_key'); |
| 141 |
if ($api_key) { |
| 142 |
$model = $this->settings->get('claude_model', Settings::DEFAULT_CLAUDE_MODEL); |
| 143 |
$this->ai_client = new Claude_Client($api_key, $model, self::AI_REQUEST_TIMEOUT); |
| 144 |
} |
| 145 |
} elseif ($provider === 'gemini') { |
| 146 |
$api_key = $this->settings->get('gemini_api_key'); |
| 147 |
if ($api_key) { |
| 148 |
$model = $this->settings->get('gemini_model', Settings::DEFAULT_GEMINI_MODEL); |
| 149 |
$this->ai_client = new Gemini_Client($api_key, $model, self::AI_REQUEST_TIMEOUT); |
| 150 |
} |
| 151 |
} elseif ($provider === 'openrouter') { |
| 152 |
$api_key = $this->settings->get('openrouter_api_key'); |
| 153 |
if ($api_key) { |
| 154 |
$model = $this->settings->get('openrouter_model', Settings::DEFAULT_OPENROUTER_MODEL); |
| 155 |
$this->ai_client = new OpenRouter_Client($api_key, $model, self::AI_REQUEST_TIMEOUT); |
| 156 |
} |
| 157 |
} elseif ($provider === 'openai_compatible') { |
| 158 |
// Same client as OpenAI, different host — and the key is optional, |
| 159 |
// so the URL and model id are what gate it (#721). The user's own |
| 160 |
// timeout applies: a local model writing a brief on CPU is slow, |
| 161 |
// and the setting exists for exactly that. |
| 162 |
$base_url = (string) $this->settings->get('openai_compatible_base_url', ''); |
| 163 |
$model = trim((string) $this->settings->get('openai_compatible_model', '')); |
| 164 |
if ('' !== $base_url && '' !== $model) { |
| 165 |
$this->ai_client = new OpenAI_Client( |
| 166 |
(string) $this->settings->get('openai_compatible_api_key', ''), |
| 167 |
$model, |
| 168 |
(int) $this->settings->get('openai_compatible_timeout', Settings::DEFAULT_OPENAI_COMPATIBLE_TIMEOUT), |
| 169 |
$base_url |
| 170 |
); |
| 171 |
$this->ai_client->set_json_mode((bool) $this->settings->get('openai_compatible_json_mode', false)); |
| 172 |
} |
| 173 |
} |
| 174 |
|
| 175 |
if (!$this->ai_client) { |
| 176 |
if ('openai_compatible' === $provider) { |
| 177 |
throw new \Exception('Please set the base URL and model id for your OpenAI-compatible endpoint in ThinkRank settings.'); |
| 178 |
} |
| 179 |
|
| 180 |
throw new \Exception('Please configure your AI provider API key in ThinkRank settings.'); |
| 181 |
} |
| 182 |
} |
| 183 |
|
| 184 |
/** |
| 185 |
* Get current AI model being used |
| 186 |
* |
| 187 |
* @return string Current model name |
| 188 |
*/ |
| 189 |
private function get_current_model(): string { |
| 190 |
// Try to get model from the actual AI client if available |
| 191 |
if ($this->ai_client && method_exists($this->ai_client, 'get_model')) { |
| 192 |
return $this->ai_client->get_model(); |
| 193 |
} |
| 194 |
|
| 195 |
// Fallback to settings |
| 196 |
$provider = $this->settings->get('ai_provider', Settings::AI_PROVIDER_NONE); |
| 197 |
if (Settings::AI_PROVIDER_NONE === $provider) { |
| 198 |
// No provider chosen, so there is no model to name. Reporting the |
| 199 |
// OpenAI default here would attribute output to a provider the site |
| 200 |
// never selected (#572). |
| 201 |
return ''; |
| 202 |
} |
| 203 |
if ($provider === 'claude') { |
| 204 |
return $this->settings->get('claude_model', Settings::DEFAULT_CLAUDE_MODEL); |
| 205 |
} elseif ($provider === 'gemini') { |
| 206 |
return $this->settings->get('gemini_model', Settings::DEFAULT_GEMINI_MODEL); |
| 207 |
} elseif ($provider === 'openrouter') { |
| 208 |
return $this->settings->get('openrouter_model', Settings::DEFAULT_OPENROUTER_MODEL); |
| 209 |
} elseif ($provider === 'openai_compatible') { |
| 210 |
return (string) $this->settings->get('openai_compatible_model', ''); |
| 211 |
} else { |
| 212 |
return $this->settings->get('openai_model', Settings::DEFAULT_OPENAI_MODEL); |
| 213 |
} |
| 214 |
} |
| 215 |
|
| 216 |
/** |
| 217 |
* Resolve the reasoning-effort level for a content-brief request. |
| 218 |
* |
| 219 |
* Without an explicit level, GPT-5 models run at their default (maximum) |
| 220 |
* reasoning effort against a ~95% completion budget — the slowest and |
| 221 |
* costliest configuration, where billed reasoning tokens (drawn from the |
| 222 |
* same budget) are spent before any visible output (issue #286). |
| 223 |
* |
| 224 |
* A brief is a structured planning task, so 'low' is a provisional middle |
| 225 |
* ground between 'minimal' and the model's default. |
| 226 |
* The level is filterable so a site can trade latency for more reasoning; |
| 227 |
* returning '' opts out entirely and lets the model use its default effort. |
| 228 |
* Only the GPT-5 family consumes this — o1/o3, gpt-4o and the non-OpenAI |
| 229 |
* clients ignore an unrecognised option key. |
| 230 |
* |
| 231 |
* @param string $model The resolved model ID (passed to the filter). |
| 232 |
* @param array $params The brief generation parameters (passed to the filter). |
| 233 |
* @return string One of 'minimal' | 'low' | 'medium' | 'high', or '' to opt out. |
| 234 |
*/ |
| 235 |
private function resolve_reasoning_effort(string $model, array $params): string { |
| 236 |
/** |
| 237 |
* Filter the reasoning-effort level used for content-brief generation. |
| 238 |
* |
| 239 |
* @param string $effort The default level ('low'). Return '' to opt out. |
| 240 |
* @param string $model The resolved model ID for this request. |
| 241 |
* @param array $params The brief generation parameters. |
| 242 |
*/ |
| 243 |
$effort = (string) apply_filters('thinkrank_content_brief_reasoning_effort', 'low', $model, $params); |
| 244 |
|
| 245 |
// Only values OpenAI accepts may reach the request body ('' opts out). |
| 246 |
// An unrecognised filter return (e.g. 'turbo') would otherwise be sent |
| 247 |
// verbatim and fail the whole brief with a 400, so degrade to the |
| 248 |
// documented default instead. |
| 249 |
$allowed = ['', 'minimal', 'low', 'medium', 'high']; |
| 250 |
return in_array($effort, $allowed, true) ? $effort : 'low'; |
| 251 |
} |
| 252 |
|
| 253 |
/** |
| 254 |
* Get current AI provider |
| 255 |
* |
| 256 |
* @return string Current provider name |
| 257 |
*/ |
| 258 |
private function get_current_provider(): string { |
| 259 |
return $this->settings->get('ai_provider', Settings::AI_PROVIDER_NONE); |
| 260 |
} |
| 261 |
|
| 262 |
/** |
| 263 |
* Extract token usage from AI response |
| 264 |
* |
| 265 |
* @param array $ai_response AI response data |
| 266 |
* @return int Number of tokens used |
| 267 |
*/ |
| 268 |
private function extract_token_usage(array $ai_response): int { |
| 269 |
$provider = $this->get_current_provider(); |
| 270 |
|
| 271 |
if ($provider === 'openai' || $provider === 'openrouter' || $provider === 'openai_compatible') { |
| 272 |
// OpenAI-compatible format: response['usage']['total_tokens'] |
| 273 |
return (int) ($ai_response['usage']['total_tokens'] ?? 0); |
| 274 |
} elseif ($provider === 'claude') { |
| 275 |
// Claude format: response['usage']['input_tokens'] + response['usage']['output_tokens'] |
| 276 |
$input_tokens = (int) ($ai_response['usage']['input_tokens'] ?? 0); |
| 277 |
$output_tokens = (int) ($ai_response['usage']['output_tokens'] ?? 0); |
| 278 |
return $input_tokens + $output_tokens; |
| 279 |
} elseif ($provider === 'gemini') { |
| 280 |
// Gemini format: response['usageMetadata']['totalTokenCount'] |
| 281 |
return (int) ($ai_response['usageMetadata']['totalTokenCount'] ?? 0); |
| 282 |
} |
| 283 |
|
| 284 |
// Fallback: return 0 if provider not recognized or no usage data |
| 285 |
return 0; |
| 286 |
} |
| 287 |
|
| 288 |
/** |
| 289 |
* Extract actual model used from AI response |
| 290 |
* |
| 291 |
* @param array $ai_response AI response data |
| 292 |
* @return string|null Actual model used or null if not found |
| 293 |
*/ |
| 294 |
private function extract_model_from_response(array $ai_response): ?string { |
| 295 |
// OpenAI format: response['model'] |
| 296 |
if (isset($ai_response['model'])) { |
| 297 |
return $ai_response['model']; |
| 298 |
} |
| 299 |
|
| 300 |
// Claude format: response['model'] |
| 301 |
if (isset($ai_response['model'])) { |
| 302 |
return $ai_response['model']; |
| 303 |
} |
| 304 |
|
| 305 |
// Gemini doesn't include model in response, fallback to client model |
| 306 |
return null; |
| 307 |
} |
| 308 |
|
| 309 |
/** |
| 310 |
* Generate content brief |
| 311 |
* |
| 312 |
* @param array $params Brief generation parameters |
| 313 |
* @return array Generated brief data |
| 314 |
* @throws \Exception If generation fails |
| 315 |
*/ |
| 316 |
public function generate_brief(array $params): array { |
| 317 |
// Validate required parameters |
| 318 |
$this->validate_brief_params($params); |
| 319 |
|
| 320 |
// Extract parameters |
| 321 |
$target_keywords = $params['target_keywords'] ?? []; |
| 322 |
$content_type = $params['content_type'] ?? 'blog_post'; |
| 323 |
$target_audience = $params['target_audience'] ?? 'general'; |
| 324 |
$content_length = $params['content_length'] ?? 'medium'; |
| 325 |
$tone = $params['tone'] ?? 'professional'; |
| 326 |
$competitor_urls = $params['competitor_urls'] ?? []; |
| 327 |
$additional_context = $params['additional_context'] ?? ''; |
| 328 |
// Write the brief in the site (or related post's) language rather than |
| 329 |
// defaulting to English on non-English sites (issue #234). |
| 330 |
$language = \ThinkRank\AI\Language_Resolver::resolve((int) ($params['post_id'] ?? 0)); |
| 331 |
|
| 332 |
// Analyze competitor URLs if provided |
| 333 |
$competitor_analysis = ''; |
| 334 |
if (!empty($competitor_urls)) { |
| 335 |
$competitor_analysis = $this->analyze_competitor_urls($competitor_urls); |
| 336 |
} |
| 337 |
|
| 338 |
// Build AI prompt using shared Prompt Builder |
| 339 |
$prompt_builder = $this->get_prompt_builder(); |
| 340 |
$prompt = $prompt_builder->build_content_brief_prompt( |
| 341 |
$target_keywords, |
| 342 |
$content_type, |
| 343 |
$target_audience, |
| 344 |
$content_length, |
| 345 |
$tone, |
| 346 |
$competitor_analysis, |
| 347 |
$additional_context, |
| 348 |
$this->get_current_provider(), |
| 349 |
$language |
| 350 |
); |
| 351 |
|
| 352 |
try { |
| 353 |
// Get the model-aware budget for a comprehensive brief, then scale |
| 354 |
// it to the requested content length so Short/Medium/Long actually |
| 355 |
// request different budgets (issue #287). Every client (OpenAI, |
| 356 |
// Claude, Gemini, OpenRouter) implements get_recommended_tokens(), |
| 357 |
// so there is no model-blind fallback. |
| 358 |
$base_tokens = (int) $this->ai_client->get_recommended_tokens('content_brief'); |
| 359 |
$max_tokens = $this->scale_tokens_for_length($base_tokens, $content_length); |
| 360 |
|
| 361 |
// Bound hidden reasoning on the GPT-5 family (issue #286). See |
| 362 |
// resolve_reasoning_effort(). Only the GPT-5 family reads this; |
| 363 |
// o1/o3, gpt-4o and the non-OpenAI clients ignore the option, and |
| 364 |
// an empty string opts out (model default effort). |
| 365 |
$reasoning_effort = $this->resolve_reasoning_effort($this->get_current_model(), $params); |
| 366 |
|
| 367 |
$completion_options = [ |
| 368 |
// For GPT‑5 family the client translates max_tokens to |
| 369 |
// max_completion_tokens internally. Temperature is intentionally |
| 370 |
// omitted: every client defaults it to 0.7, and reasoning models |
| 371 |
// reject it outright, so passing it here was misleading no-op. |
| 372 |
'max_tokens' => $max_tokens, |
| 373 |
// The brief is one JSON object. Only a compatible endpoint |
| 374 |
// with JSON mode on reads this; every other client ignores it. |
| 375 |
'json_object' => true, |
| 376 |
]; |
| 377 |
if ('' !== $reasoning_effort) { |
| 378 |
$completion_options['reasoning_effort'] = $reasoning_effort; |
| 379 |
} |
| 380 |
|
| 381 |
// Generate brief using AI |
| 382 |
$ai_response = $this->ai_client->generate_completion($prompt, $completion_options); |
| 383 |
|
| 384 |
// Detect a provider-side non-answer (refusal, policy block, or |
| 385 |
// truncation) BEFORE attempting text extraction. Otherwise a |
| 386 |
// refusal — which OpenAI returns as HTTP 200 with content=null — |
| 387 |
// slips past every isset() branch and gets serialized into the |
| 388 |
// brief body instead of being reported to the user. |
| 389 |
$this->guard_against_non_answer($ai_response, $max_tokens); |
| 390 |
|
| 391 |
// Extract text content from AI response |
| 392 |
$ai_text = ''; |
| 393 |
|
| 394 |
// Handle OpenAI response format |
| 395 |
if (isset($ai_response['choices'][0]['message']['content'])) { |
| 396 |
$content_field = $ai_response['choices'][0]['message']['content']; |
| 397 |
if (is_string($content_field)) { |
| 398 |
$ai_text = $content_field; |
| 399 |
} elseif (is_array($content_field)) { |
| 400 |
// Concatenate text parts from array-based content (Chat Completions multimodal) |
| 401 |
$parts = array_map(function($part) { |
| 402 |
if (is_array($part)) { |
| 403 |
return $part['text'] ?? ''; |
| 404 |
} |
| 405 |
return is_string($part) ? $part : ''; |
| 406 |
}, $content_field); |
| 407 |
$ai_text = trim(implode("\n", array_filter($parts))); |
| 408 |
} |
| 409 |
} |
| 410 |
// Handle Claude response format |
| 411 |
elseif (isset($ai_response['content'][0]['text'])) { |
| 412 |
$ai_text = $ai_response['content'][0]['text']; |
| 413 |
} |
| 414 |
// Handle Gemini response format |
| 415 |
elseif (isset($ai_response['candidates'][0]['content']['parts'][0]['text'])) { |
| 416 |
$ai_text = $ai_response['candidates'][0]['content']['parts'][0]['text']; |
| 417 |
} |
| 418 |
// Handle direct content field |
| 419 |
elseif (isset($ai_response['content']) && is_string($ai_response['content'])) { |
| 420 |
$ai_text = $ai_response['content']; |
| 421 |
} |
| 422 |
// Handle direct string response |
| 423 |
elseif (is_string($ai_response)) { |
| 424 |
$ai_text = $ai_response; |
| 425 |
} |
| 426 |
// No known provider shape matched and guard_against_non_answer() |
| 427 |
// found nothing it recognised. Never serialize the raw envelope |
| 428 |
// into the brief body — that turns a clear failure into a saved, |
| 429 |
// meaningless brief. Log the shape for diagnostics and fail. |
| 430 |
else { |
| 431 |
if (defined('WP_DEBUG') && WP_DEBUG) { |
| 432 |
$shape = is_array($ai_response) ? implode(', ', array_keys($ai_response)) : gettype($ai_response); |
| 433 |
// phpcs:ignore WordPress.PHP.DevelopmentFunctions.error_log_error_log -- Debug logging only when WP_DEBUG is enabled. |
| 434 |
error_log('[ThinkRank] Content brief: unrecognised AI response shape. Top-level keys: ' . $shape); |
| 435 |
} |
| 436 |
throw new \Exception('The AI returned a response in an unexpected format. Please try again.'); |
| 437 |
} |
| 438 |
|
| 439 |
// Ensure we have actual text content |
| 440 |
if (empty(trim($ai_text))) { |
| 441 |
throw new \Exception('AI response was empty or contained no text content.'); |
| 442 |
} |
| 443 |
|
| 444 |
// Extract token usage for analytics tracking |
| 445 |
$tokens_used = $this->extract_token_usage($ai_response); |
| 446 |
|
| 447 |
// Parse and structure the response |
| 448 |
$brief_data = $this->parse_ai_response($ai_text, $params); |
| 449 |
|
| 450 |
// Extract actual model from response before using it |
| 451 |
$actual_model = $this->extract_model_from_response($ai_response); |
| 452 |
|
| 453 |
// Add generation metadata (use actual model from response if available) |
| 454 |
$brief_data['generation_meta'] = [ |
| 455 |
'provider' => $this->get_current_provider(), |
| 456 |
'model' => $actual_model ?: $this->get_current_model(), |
| 457 |
'generated_at' => current_time('mysql'), |
| 458 |
'version' => '1.0' |
| 459 |
]; |
| 460 |
|
| 461 |
// Save brief to database |
| 462 |
$brief_id = $this->save_brief($brief_data); |
| 463 |
$brief_data['id'] = $brief_id; |
| 464 |
|
| 465 |
// Log AI usage for analytics tracking (including raw response and actual model used) |
| 466 |
$usage_id = $this->log_ai_usage(get_current_user_id(), 'Content Brief', $tokens_used, $brief_id, $ai_text, $actual_model); |
| 467 |
|
| 468 |
// Set raw response for immediate display |
| 469 |
$brief_data['raw_response'] = $ai_text; |
| 470 |
|
| 471 |
// Apply normalization for React compatibility |
| 472 |
$brief_data = $this->normalize_brief_data($brief_data); |
| 473 |
|
| 474 |
return $brief_data; |
| 475 |
|
| 476 |
} catch (\Exception $e) { |
| 477 |
// Provide more specific error messages |
| 478 |
$error_message = $e->getMessage(); |
| 479 |
|
| 480 |
// Messages we authored for the user (refusals, policy blocks, |
| 481 |
// token-limit truncation, unexpected shape) all start with "The AI " |
| 482 |
// and are already actionable. Pass them through verbatim instead of |
| 483 |
// flattening them via the substring matching below — e.g. so a |
| 484 |
// refusal is not rewritten into generic "empty content" advice. |
| 485 |
if (strpos($error_message, 'The AI ') === 0) { |
| 486 |
throw new \Exception(esc_html($error_message)); |
| 487 |
} |
| 488 |
|
| 489 |
if (strpos($error_message, 'API key') !== false) { |
| 490 |
throw new \Exception('API key configuration error. Please check your AI provider settings.'); |
| 491 |
} elseif (strpos($error_message, 'Invalid AI response format') !== false) { |
| 492 |
throw new \Exception('AI service returned an unexpected response format. Please try again.'); |
| 493 |
} elseif (strpos($error_message, 'empty') !== false) { |
| 494 |
throw new \Exception('AI service returned empty content. Please try again with different parameters.'); |
| 495 |
} else { |
| 496 |
throw new \Exception('Failed to generate content brief: ' . esc_html($error_message)); |
| 497 |
} |
| 498 |
} |
| 499 |
} |
| 500 |
|
| 501 |
/** |
| 502 |
* Scale the model-aware brief budget to the requested content length. |
| 503 |
* |
| 504 |
* get_recommended_tokens('content_brief') returns the budget for a full, |
| 505 |
* comprehensive (Long) brief, already capped at the model's completion |
| 506 |
* ceiling. Shorter tiers request proportionally less so that choosing Short |
| 507 |
* is genuinely faster and cheaper (issue #287), while every tier stays at or |
| 508 |
* below the base and at or above MIN_BRIEF_TOKENS so it cannot truncate. |
| 509 |
* |
| 510 |
* @param int $base_tokens Model-aware budget for a comprehensive brief. |
| 511 |
* @param string $content_length One of 'short' | 'medium' | 'long'. |
| 512 |
* @return int Scaled max_tokens, clamped to [floor, base_tokens]. |
| 513 |
*/ |
| 514 |
private function scale_tokens_for_length(int $base_tokens, string $content_length): int { |
| 515 |
// Unknown/missing length falls back to the medium tier — never to 0 or |
| 516 |
// to the raw ceiling. |
| 517 |
$multiplier = self::LENGTH_TOKEN_MULTIPLIERS[$content_length] |
| 518 |
?? self::LENGTH_TOKEN_MULTIPLIERS['medium']; |
| 519 |
|
| 520 |
$scaled = (int) round($base_tokens * $multiplier); |
| 521 |
|
| 522 |
// The floor can never exceed the base itself, so a model with a tiny |
| 523 |
// ceiling still yields a sane, in-range value. |
| 524 |
$floor = (int) min($base_tokens, self::MIN_BRIEF_TOKENS); |
| 525 |
|
| 526 |
return max($floor, min($scaled, $base_tokens)); |
| 527 |
} |
| 528 |
|
| 529 |
/** |
| 530 |
* Detect a provider-side non-answer and fail with the real reason. |
| 531 |
* |
| 532 |
* A refusal, content-policy block, or token-limit truncation is not a |
| 533 |
* usable brief. Each provider signals these differently, and none of the |
| 534 |
* signals set the content field the extraction chain looks for — so if we |
| 535 |
* don't catch them here they fall through to the "unexpected format" path |
| 536 |
* (or, historically, were serialized into the brief body). All messages |
| 537 |
* start with "The AI " so the outer catch passes them through unchanged. |
| 538 |
* |
| 539 |
* @param mixed $ai_response Raw response from the AI client. |
| 540 |
* @param int $requested_tokens The max_tokens this request asked for; 0 when unknown. |
| 541 |
* @throws \Exception If the response is a refusal, policy block, or truncation. |
| 542 |
*/ |
| 543 |
private function guard_against_non_answer($ai_response, int $requested_tokens = 0): void { |
| 544 |
if (!is_array($ai_response)) { |
| 545 |
return; |
| 546 |
} |
| 547 |
|
| 548 |
// --- OpenAI (Chat Completions) --- |
| 549 |
// A structured refusal is HTTP 200 with message.content=null and the |
| 550 |
// stated reason carried in message.refusal. finish_reason distinguishes |
| 551 |
// a policy block from a truncated completion. |
| 552 |
if (isset($ai_response['choices'][0]['message'])) { |
| 553 |
$message = $ai_response['choices'][0]['message']; |
| 554 |
$finish = (string) ($ai_response['choices'][0]['finish_reason'] ?? ''); |
| 555 |
|
| 556 |
if (!empty($message['refusal'])) { |
| 557 |
throw new \Exception(esc_html(sprintf( |
| 558 |
'The AI declined to generate this brief: %s', |
| 559 |
(string) $message['refusal'] |
| 560 |
))); |
| 561 |
} |
| 562 |
if ('content_filter' === $finish) { |
| 563 |
throw new \Exception('The AI blocked this request under its content policy. Try a different topic or less sensitive keywords.'); |
| 564 |
} |
| 565 |
// A self-hosted server can stop short of max_tokens because the |
| 566 |
// prompt and the answer together filled its context window |
| 567 |
// (Ollama loads models at 4096 by default). A bigger output budget |
| 568 |
// cannot fix that, so say what can. |
| 569 |
$completion_tokens = (int) ($ai_response['usage']['completion_tokens'] ?? 0); |
| 570 |
if ('length' === $finish && $requested_tokens > 0 && $completion_tokens > 0 && $completion_tokens < $requested_tokens) { |
| 571 |
throw new \Exception(esc_html(sprintf( |
| 572 |
'The AI stopped after %1$d tokens, short of the %2$d allowed, because the server ran out of context window before finishing the brief. Raise the context length on your AI server (for Ollama, set OLLAMA_CONTEXT_LENGTH to 16384 or more) and try again.', |
| 573 |
$completion_tokens, |
| 574 |
$requested_tokens |
| 575 |
))); |
| 576 |
} |
| 577 |
if ('length' === $finish) { |
| 578 |
throw new \Exception('The AI stopped at its output token limit before finishing the brief. Try fewer competitor URLs, or a model with a larger output limit. A shorter content length will not help: it asks for a smaller budget, not a smaller answer.'); |
| 579 |
} |
| 580 |
} |
| 581 |
|
| 582 |
// --- Claude (Messages) --- |
| 583 |
if (isset($ai_response['stop_reason'])) { |
| 584 |
$stop_reason = (string) $ai_response['stop_reason']; |
| 585 |
if ('refusal' === $stop_reason) { |
| 586 |
throw new \Exception('The AI declined to generate this brief for this topic. Try a different topic or less sensitive keywords.'); |
| 587 |
} |
| 588 |
if ('max_tokens' === $stop_reason) { |
| 589 |
throw new \Exception('The AI stopped at its output token limit before finishing the brief. Try fewer competitor URLs, or a model with a larger output limit. A shorter content length will not help: it asks for a smaller budget, not a smaller answer.'); |
| 590 |
} |
| 591 |
} |
| 592 |
|
| 593 |
// --- Gemini --- |
| 594 |
// A prompt rejected outright returns no candidate at all, only |
| 595 |
// promptFeedback.blockReason; a candidate can also finish on SAFETY or |
| 596 |
// PROHIBITED_CONTENT, or be truncated at MAX_TOKENS. |
| 597 |
$block_reason = (string) ($ai_response['promptFeedback']['blockReason'] ?? ''); |
| 598 |
if ('' !== $block_reason) { |
| 599 |
throw new \Exception(esc_html(sprintf( |
| 600 |
'The AI blocked this request under its content policy (%s). Try a different topic or less sensitive keywords.', |
| 601 |
$block_reason |
| 602 |
))); |
| 603 |
} |
| 604 |
$gemini_finish = (string) ($ai_response['candidates'][0]['finishReason'] ?? ''); |
| 605 |
if (in_array($gemini_finish, ['SAFETY', 'PROHIBITED_CONTENT'], true)) { |
| 606 |
throw new \Exception('The AI blocked this request under its content policy. Try a different topic or less sensitive keywords.'); |
| 607 |
} |
| 608 |
if ('MAX_TOKENS' === $gemini_finish) { |
| 609 |
throw new \Exception('The AI stopped at its output token limit before finishing the brief. Try fewer competitor URLs, or a model with a larger output limit. A shorter content length will not help: it asks for a smaller budget, not a smaller answer.'); |
| 610 |
} |
| 611 |
} |
| 612 |
|
| 613 |
/** |
| 614 |
* Validate brief generation parameters |
| 615 |
* |
| 616 |
* @param array $params Parameters to validate |
| 617 |
* @throws \Exception If validation fails |
| 618 |
*/ |
| 619 |
private function validate_brief_params(array $params): void { |
| 620 |
if (empty($params['target_keywords']) || !is_array($params['target_keywords'])) { |
| 621 |
throw new \Exception('Target keywords are required and must be an array.'); |
| 622 |
} |
| 623 |
|
| 624 |
$valid_content_types = ['blog_post', 'product_page', 'landing_page', 'tutorial']; |
| 625 |
if (!empty($params['content_type']) && !in_array($params['content_type'], $valid_content_types, true)) { |
| 626 |
throw new \Exception('Invalid content type specified.'); |
| 627 |
} |
| 628 |
|
| 629 |
$valid_lengths = ['short', 'medium', 'long']; |
| 630 |
if (!empty($params['content_length']) && !in_array($params['content_length'], $valid_lengths, true)) { |
| 631 |
throw new \Exception('Invalid content length specified.'); |
| 632 |
} |
| 633 |
|
| 634 |
$valid_tones = ['professional', 'casual', 'technical', 'friendly']; |
| 635 |
if (!empty($params['tone']) && !in_array($params['tone'], $valid_tones, true)) { |
| 636 |
throw new \Exception('Invalid tone specified.'); |
| 637 |
} |
| 638 |
} |
| 639 |
|
| 640 |
/** |
| 641 |
* Parse AI response into structured data |
| 642 |
* |
| 643 |
* @param string $ai_response Raw AI response |
| 644 |
* @param array $original_params Original generation parameters |
| 645 |
* @return array Structured brief data |
| 646 |
*/ |
| 647 |
private function parse_ai_response(string $ai_response, array $original_params): array { |
| 648 |
$json_data = $this->parse_json_response($ai_response); |
| 649 |
|
| 650 |
if (null === $json_data) { |
| 651 |
// JSON parsing failed - return error structure |
| 652 |
return $this->create_parsing_error_response($ai_response, $original_params); |
| 653 |
} |
| 654 |
|
| 655 |
return $this->structure_json_data($json_data, $original_params); |
| 656 |
} |
| 657 |
|
| 658 |
/** |
| 659 |
* Parse JSON response from AI |
| 660 |
* |
| 661 |
* @param string $ai_response Raw AI response |
| 662 |
* @return array|null Parsed JSON data or null if parsing fails |
| 663 |
*/ |
| 664 |
private function parse_json_response(string $ai_response): ?array { |
| 665 |
// Clean the response - remove any text before/after JSON |
| 666 |
$ai_response = trim($ai_response); |
| 667 |
|
| 668 |
// Handle markdown code blocks (```json ... ```) |
| 669 |
if (preg_match('/```(?:json)?\s*\n?(.*?)\n?```/s', $ai_response, $matches)) { |
| 670 |
$json_string = trim($matches[1]); |
| 671 |
} else { |
| 672 |
// Find JSON object boundaries |
| 673 |
$start = strpos($ai_response, '{'); |
| 674 |
$end = strrpos($ai_response, '}'); |
| 675 |
|
| 676 |
if (false === $start || false === $end || $start >= $end) { |
| 677 |
return null; |
| 678 |
} |
| 679 |
|
| 680 |
$json_string = substr($ai_response, $start, $end - $start + 1); |
| 681 |
} |
| 682 |
|
| 683 |
$json_data = json_decode($json_string, true); |
| 684 |
|
| 685 |
if (json_last_error() !== JSON_ERROR_NONE) { |
| 686 |
return null; |
| 687 |
} |
| 688 |
|
| 689 |
return $json_data; |
| 690 |
} |
| 691 |
|
| 692 |
/** |
| 693 |
* Structure JSON data into expected format |
| 694 |
* |
| 695 |
* @param array $json_data Parsed JSON data |
| 696 |
* @param array $original_params Original generation parameters |
| 697 |
* @return array Structured brief data |
| 698 |
*/ |
| 699 |
private function structure_json_data(array $json_data, array $original_params): array { |
| 700 |
return [ |
| 701 |
'title' => $json_data['title_suggestions'] ?? [], |
| 702 |
'meta_description' => $json_data['meta_descriptions'][0] ?? '', |
| 703 |
'meta_descriptions' => $json_data['meta_descriptions'] ?? [], |
| 704 |
'url_slugs' => $json_data['url_slugs'] ?? [], |
| 705 |
'outline' => self::strip_outline_level_labels($json_data['outline'] ?? []), |
| 706 |
'seo_recommendations' => [ |
| 707 |
'title_suggestions' => $json_data['title_suggestions'] ?? [], |
| 708 |
'meta_description' => $json_data['meta_descriptions'][0] ?? '', |
| 709 |
'meta_descriptions' => $json_data['meta_descriptions'] ?? [], |
| 710 |
'url_slugs' => $json_data['url_slugs'] ?? [], |
| 711 |
'focus_keyword_analysis' => $this->normalize_focus_keyword_analysis($json_data['focus_keyword_analysis'] ?? []), |
| 712 |
'internal_links' => $json_data['internal_linking'] ?? [], |
| 713 |
'related_keywords' => $json_data['related_keywords'] ?? [], |
| 714 |
'long_tail_keywords' => [] |
| 715 |
], |
| 716 |
'social_media' => $json_data['social_media'] ?? [ |
| 717 |
'open_graph' => ['title' => '', 'description' => ''], |
| 718 |
'twitter_card' => ['title' => '', 'description' => ''] |
| 719 |
], |
| 720 |
'schema_markup' => $json_data['schema_markup'] ?? [ |
| 721 |
'recommended_types' => [], |
| 722 |
'key_properties' => [], |
| 723 |
'faq_questions' => [] |
| 724 |
], |
| 725 |
'visual_content' => $json_data['visual_content'] ?? [ |
| 726 |
'image_recommendations' => [], |
| 727 |
'alt_text_suggestions' => [], |
| 728 |
'infographic_opportunities' => [] |
| 729 |
], |
| 730 |
'competitor_gaps' => $json_data['competitor_analysis']['content_gaps'] ?? [], |
| 731 |
'call_to_actions' => $json_data['call_to_actions'] ?? [], |
| 732 |
'writing_guidelines' => $json_data['writing_guidelines'] ?? [], |
| 733 |
'content_body' => self::strip_heading_level_labels((string) ($json_data['content_body'] ?? '')), |
| 734 |
'estimated_word_count' => $this->get_word_count_estimate($original_params['content_length'] ?? 'medium'), |
| 735 |
'raw_response' => '', // Will be retrieved from ai_usage table |
| 736 |
'generation_params' => $original_params, |
| 737 |
'parsing_status' => 'success', |
| 738 |
'created_at' => current_time('mysql') |
| 739 |
]; |
| 740 |
} |
| 741 |
|
| 742 |
/** |
| 743 |
* Remove a leading level label from a heading string. |
| 744 |
* |
| 745 |
* The prompt's own JSON example labelled outline headings with their level |
| 746 |
* (`"heading": "H1: Main Title"` next to a separate `"level": 1`), so the |
| 747 |
* model often carried the convention into the drafted article and Pro's |
| 748 |
* "Insert into post" wrote `<h2>H2: Real Heading</h2>` into published |
| 749 |
* content. The prompt no longer does that, but a prompt change never fully |
| 750 |
* binds a model — so the label is stripped here too (#410). |
| 751 |
* |
| 752 |
* Covers the label forms a model actually emits: `H2:`, `h3:`, `H2 -`, |
| 753 |
* `H4.`, `H2)` and the en/em dash variants, optionally wrapped in markdown |
| 754 |
* emphasis (`**H2:**`). The delimiter is anchored directly after the digit |
| 755 |
* so `H10:` — a plausible heading in a numbered list — is left alone, and |
| 756 |
* only a leading label is matched so body copy that mentions a level |
| 757 |
* survives. Trailing emphasis is consumed only when the same marker opened |
| 758 |
* the label, so `H2: *emphasised start*` keeps its asterisks. |
| 759 |
* |
| 760 |
* @since 2.0.1 |
| 761 |
* |
| 762 |
* @param string $heading Heading text. |
| 763 |
* @return string Heading without its level prefix. |
| 764 |
*/ |
| 765 |
public static function strip_level_label(string $heading): string { |
| 766 |
// En dash and em dash as raw UTF-8 bytes, so the pattern needs no /u |
| 767 |
// modifier and cannot blank a heading that is not valid UTF-8. |
| 768 |
$delimiter = '(?:[:.)\-]|\xe2\x80\x93|\xe2\x80\x94)'; |
| 769 |
$emphasis = '(\*{1,3}|_{1,3})'; |
| 770 |
|
| 771 |
$pattern = '/^\s*(?:' |
| 772 |
. $emphasis . '\s*[Hh][1-6]\s*' . $delimiter . '\s*\1' |
| 773 |
. '|[Hh][1-6]\s*' . $delimiter |
| 774 |
. ')\s*/'; |
| 775 |
|
| 776 |
return (string) preg_replace($pattern, '', $heading); |
| 777 |
} |
| 778 |
|
| 779 |
/** |
| 780 |
* Strip a level label from a heading's inner HTML. |
| 781 |
* |
| 782 |
* A model drafting publish-ready HTML often wraps the heading text in an |
| 783 |
* inline tag (`<h2><strong>H2: Real Heading</strong></h2>`). That pushes a |
| 784 |
* `<` in front of the label, so the leading run of inline opening tags is |
| 785 |
* set aside and re-attached around the cleaned text. |
| 786 |
* |
| 787 |
* @since 2.0.1 |
| 788 |
* |
| 789 |
* @param string $inner Heading inner HTML. |
| 790 |
* @return string Inner HTML without the level prefix. |
| 791 |
*/ |
| 792 |
private static function strip_inner_level_label(string $inner): string { |
| 793 |
$prefix = ''; |
| 794 |
|
| 795 |
if (preg_match('/^(\s*(?:<(?:strong|em|b|i|span|mark|code|u)\b[^>]*>\s*)+)(.*)$/is', $inner, $parts)) { |
| 796 |
$prefix = $parts[1]; |
| 797 |
$inner = $parts[2]; |
| 798 |
} |
| 799 |
|
| 800 |
return $prefix . self::strip_level_label($inner); |
| 801 |
} |
| 802 |
|
| 803 |
/** |
| 804 |
* Strip level labels from every heading in an outline. |
| 805 |
* |
| 806 |
* @since 2.0.1 |
| 807 |
* |
| 808 |
* @param mixed $outline Outline as returned by the model. |
| 809 |
* @return array Outline with clean headings. |
| 810 |
*/ |
| 811 |
public static function strip_outline_level_labels($outline): array { |
| 812 |
if (!is_array($outline)) { |
| 813 |
return []; |
| 814 |
} |
| 815 |
|
| 816 |
foreach ($outline as $index => $section) { |
| 817 |
if (is_array($section) && isset($section['heading']) && is_string($section['heading'])) { |
| 818 |
$outline[$index]['heading'] = self::strip_level_label($section['heading']); |
| 819 |
} elseif (is_string($section)) { |
| 820 |
$outline[$index] = self::strip_level_label($section); |
| 821 |
} |
| 822 |
} |
| 823 |
|
| 824 |
return $outline; |
| 825 |
} |
| 826 |
|
| 827 |
/** |
| 828 |
* Strip level labels from the heading text inside drafted HTML. |
| 829 |
* |
| 830 |
* This is the path that reaches published post content, so it is the one |
| 831 |
* that matters most. Only the text directly inside an <h1>-<h6> is touched. |
| 832 |
* |
| 833 |
* @since 2.0.1 |
| 834 |
* |
| 835 |
* @param string $html Drafted article body. |
| 836 |
* @return string Body with clean headings. |
| 837 |
*/ |
| 838 |
public static function strip_heading_level_labels(string $html): string { |
| 839 |
if ('' === $html || false === stripos($html, '<h')) { |
| 840 |
return $html; |
| 841 |
} |
| 842 |
|
| 843 |
return (string) preg_replace_callback( |
| 844 |
'/(<h([1-6])\b[^>]*>)(.*?)(<\/h\2>)/is', |
| 845 |
static function (array $parts): string { |
| 846 |
return $parts[1] . self::strip_inner_level_label($parts[3]) . $parts[4]; |
| 847 |
}, |
| 848 |
$html |
| 849 |
); |
| 850 |
} |
| 851 |
|
| 852 |
/** |
| 853 |
* Create error response when JSON parsing fails |
| 854 |
* |
| 855 |
* @param string $ai_response Raw AI response |
| 856 |
* @param array $original_params Original generation parameters |
| 857 |
* @return array Error response structure |
| 858 |
*/ |
| 859 |
private function create_parsing_error_response(string $ai_response, array $original_params): array { |
| 860 |
return [ |
| 861 |
'title' => ['Error: Unable to parse AI response'], |
| 862 |
'meta_description' => 'AI response could not be parsed as valid JSON.', |
| 863 |
'meta_descriptions' => ['AI response could not be parsed as valid JSON.'], |
| 864 |
'url_slugs' => ['error-parsing-response'], |
| 865 |
'outline' => [], |
| 866 |
'seo_recommendations' => [ |
| 867 |
'title_suggestions' => ['Error: Unable to parse AI response'], |
| 868 |
'meta_description' => 'AI response could not be parsed as valid JSON.', |
| 869 |
'meta_descriptions' => ['AI response could not be parsed as valid JSON.'], |
| 870 |
'url_slugs' => ['error-parsing-response'], |
| 871 |
'focus_keyword_analysis' => [ |
| 872 |
'primary_placement' => [], |
| 873 |
'secondary_integration' => [], |
| 874 |
'density_guidelines' => [] |
| 875 |
], |
| 876 |
'internal_links' => [], |
| 877 |
'related_keywords' => [], |
| 878 |
'long_tail_keywords' => [] |
| 879 |
], |
| 880 |
'social_media' => [ |
| 881 |
'open_graph' => ['title' => 'Error', 'description' => 'Parsing failed'], |
| 882 |
'twitter_card' => ['title' => 'Error', 'description' => 'Parsing failed'] |
| 883 |
], |
| 884 |
'schema_markup' => [ |
| 885 |
'recommended_types' => [], |
| 886 |
'key_properties' => [], |
| 887 |
'faq_questions' => [] |
| 888 |
], |
| 889 |
'visual_content' => [ |
| 890 |
'image_recommendations' => [], |
| 891 |
'alt_text_suggestions' => [], |
| 892 |
'infographic_opportunities' => [] |
| 893 |
], |
| 894 |
'competitor_gaps' => [], |
| 895 |
'call_to_actions' => [], |
| 896 |
'writing_guidelines' => [], |
| 897 |
'content_body' => '', |
| 898 |
'estimated_word_count' => $this->get_word_count_estimate($original_params['content_length'] ?? 'medium'), |
| 899 |
'raw_response' => $ai_response, // Store raw response in error case |
| 900 |
'generation_params' => $original_params, |
| 901 |
'parsing_status' => 'failed', |
| 902 |
'created_at' => current_time('mysql') |
| 903 |
]; |
| 904 |
} |
| 905 |
|
| 906 |
/** |
| 907 |
* Normalize focus keyword analysis to ensure proper array structure |
| 908 |
* |
| 909 |
* @param array $focus_keyword_analysis Raw focus keyword analysis data |
| 910 |
* @return array Normalized focus keyword analysis |
| 911 |
*/ |
| 912 |
private function normalize_focus_keyword_analysis(array $focus_keyword_analysis): array { |
| 913 |
$normalized = [ |
| 914 |
'primary_placement' => [], |
| 915 |
'secondary_integration' => [], |
| 916 |
'density_guidelines' => [] |
| 917 |
]; |
| 918 |
|
| 919 |
// Normalize primary_placement |
| 920 |
if (isset($focus_keyword_analysis['primary_placement'])) { |
| 921 |
if (is_array($focus_keyword_analysis['primary_placement'])) { |
| 922 |
$normalized['primary_placement'] = $focus_keyword_analysis['primary_placement']; |
| 923 |
} elseif (is_string($focus_keyword_analysis['primary_placement'])) { |
| 924 |
// Convert string to array by splitting on common delimiters |
| 925 |
$normalized['primary_placement'] = array_filter(array_map('trim', preg_split('/[,;]/', $focus_keyword_analysis['primary_placement']))); |
| 926 |
} |
| 927 |
} |
| 928 |
|
| 929 |
// Normalize secondary_integration - this is the problematic field |
| 930 |
if (isset($focus_keyword_analysis['secondary_integration'])) { |
| 931 |
if (is_array($focus_keyword_analysis['secondary_integration'])) { |
| 932 |
$normalized['secondary_integration'] = $focus_keyword_analysis['secondary_integration']; |
| 933 |
} elseif (is_string($focus_keyword_analysis['secondary_integration'])) { |
| 934 |
// Convert string to array - split by sentences or use as single item |
| 935 |
$text = trim($focus_keyword_analysis['secondary_integration']); |
| 936 |
if (!empty($text)) { |
| 937 |
// Split by sentences if it contains periods, otherwise use as single item |
| 938 |
if (strpos($text, '.') !== false) { |
| 939 |
$sentences = array_filter(array_map('trim', explode('.', $text))); |
| 940 |
$normalized['secondary_integration'] = array_map(function($sentence) { |
| 941 |
return $sentence . (substr($sentence, -1) !== '.' ? '.' : ''); |
| 942 |
}, $sentences); |
| 943 |
} else { |
| 944 |
$normalized['secondary_integration'] = [$text]; |
| 945 |
} |
| 946 |
} |
| 947 |
} |
| 948 |
} |
| 949 |
|
| 950 |
// Normalize density_guidelines |
| 951 |
if (isset($focus_keyword_analysis['density_guidelines'])) { |
| 952 |
if (is_array($focus_keyword_analysis['density_guidelines'])) { |
| 953 |
$normalized['density_guidelines'] = $focus_keyword_analysis['density_guidelines']; |
| 954 |
} elseif (is_string($focus_keyword_analysis['density_guidelines'])) { |
| 955 |
// Convert string to array by splitting on common delimiters |
| 956 |
$guidelines = array_filter(array_map('trim', preg_split('/[,;]/', $focus_keyword_analysis['density_guidelines']))); |
| 957 |
$normalized['density_guidelines'] = $guidelines ?: [$focus_keyword_analysis['density_guidelines']]; |
| 958 |
} |
| 959 |
} |
| 960 |
|
| 961 |
return $normalized; |
| 962 |
} |
| 963 |
private function analyze_competitor_urls(array $urls): string { |
| 964 |
$analysis_results = []; |
| 965 |
$failed_urls = []; |
| 966 |
|
| 967 |
// Limit to first 3 URLs to prevent timeout |
| 968 |
$urls = array_slice($urls, 0, 3); |
| 969 |
|
| 970 |
foreach ($urls as $url) { |
| 971 |
$url = trim($url); |
| 972 |
if (empty($url) || !filter_var($url, FILTER_VALIDATE_URL)) { |
| 973 |
$failed_urls[] = $url . " (invalid URL)"; |
| 974 |
continue; |
| 975 |
} |
| 976 |
|
| 977 |
// SSRF guard: only fetch public http/https hosts. Blocks loopback, |
| 978 |
// link-local (cloud metadata), private and reserved ranges before any |
| 979 |
// request is made. |
| 980 |
if (!$this->is_safe_public_url($url)) { |
| 981 |
$failed_urls[] = $url . " (blocked: non-public host)"; |
| 982 |
continue; |
| 983 |
} |
| 984 |
|
| 985 |
$content_data = $this->scrape_competitor_content($url); |
| 986 |
if ($content_data) { |
| 987 |
$analysis_results[] = $this->format_competitor_analysis($url, $content_data); |
| 988 |
} else { |
| 989 |
$failed_urls[] = $url . " (failed to scrape)"; |
| 990 |
} |
| 991 |
} |
| 992 |
|
| 993 |
$result = ""; |
| 994 |
|
| 995 |
if (!empty($analysis_results)) { |
| 996 |
$result .= implode("\n\n", $analysis_results); |
| 997 |
} |
| 998 |
|
| 999 |
if (!empty($failed_urls)) { |
| 1000 |
$result .= "\n\nNote: The following URLs could not be analyzed:\n"; |
| 1001 |
$result .= "- " . implode("\n- ", $failed_urls); |
| 1002 |
} |
| 1003 |
|
| 1004 |
if (empty($analysis_results)) { |
| 1005 |
return "No competitor URLs could be successfully analyzed. Please ensure URLs are accessible and valid."; |
| 1006 |
} |
| 1007 |
|
| 1008 |
return $result; |
| 1009 |
} |
| 1010 |
|
| 1011 |
/** |
| 1012 |
* Whether a competitor URL is safe to fetch: an http/https URL whose host |
| 1013 |
* resolves only to public IP addresses. |
| 1014 |
* |
| 1015 |
* Prevents SSRF — a low-privilege user could otherwise point competitor |
| 1016 |
* scraping at loopback, link-local (e.g. 169.254.169.254 cloud metadata), |
| 1017 |
* CGNAT, private or reserved addresses to probe internal services. The |
| 1018 |
* block list lives in {@see \ThinkRank\Core\Url_Safety} so this and the |
| 1019 |
* schema importer can never drift apart. |
| 1020 |
* |
| 1021 |
* @param string $url URL to validate. |
| 1022 |
* @return bool True when safe to fetch. |
| 1023 |
*/ |
| 1024 |
private function is_safe_public_url(string $url): bool { |
| 1025 |
return \ThinkRank\Core\Url_Safety::is_safe_public_url($url); |
| 1026 |
} |
| 1027 |
|
| 1028 |
/** |
| 1029 |
* Scrape content from a competitor URL |
| 1030 |
* |
| 1031 |
* @param string $url The URL to scrape |
| 1032 |
* @return array|null Content data or null if failed |
| 1033 |
*/ |
| 1034 |
private function scrape_competitor_content(string $url): ?array { |
| 1035 |
// Url_Safety::safe_remote_get() follows redirects manually and re-checks |
| 1036 |
// the resolved host on every hop, so a target that redirects to (or |
| 1037 |
// rebinds onto) an internal address after the pre-flight check is |
| 1038 |
// refused rather than fetched. |
| 1039 |
$response = \ThinkRank\Core\Url_Safety::safe_remote_get($url, [ |
| 1040 |
'timeout' => 8, // Reduced from 15 to 8 seconds |
| 1041 |
'user-agent' => 'Mozilla/5.0 (compatible; ThinkRank SEO Bot)', |
| 1042 |
'headers' => [ |
| 1043 |
'Accept' => 'text/html,application/xhtml+xml,application/xml;q=0.9,*/*;q=0.8', |
| 1044 |
'Accept-Language' => 'en-US,en;q=0.5', |
| 1045 |
] |
| 1046 |
]); |
| 1047 |
|
| 1048 |
if (is_wp_error($response)) { |
| 1049 |
return null; |
| 1050 |
} |
| 1051 |
|
| 1052 |
$status_code = wp_remote_retrieve_response_code($response); |
| 1053 |
if ($status_code !== 200) { |
| 1054 |
return null; |
| 1055 |
} |
| 1056 |
|
| 1057 |
$html = wp_remote_retrieve_body($response); |
| 1058 |
if (empty($html)) { |
| 1059 |
return null; |
| 1060 |
} |
| 1061 |
|
| 1062 |
return $this->parse_html_content($html, $url); |
| 1063 |
} |
| 1064 |
|
| 1065 |
/** |
| 1066 |
* Parse HTML content and extract key SEO elements |
| 1067 |
* |
| 1068 |
* @param string $html HTML content |
| 1069 |
* @param string $url Original URL for context |
| 1070 |
* @return array Parsed content data |
| 1071 |
*/ |
| 1072 |
private function parse_html_content(string $html, string $url): array { |
| 1073 |
// Create DOMDocument to parse HTML |
| 1074 |
$dom = new \DOMDocument(); |
| 1075 |
|
| 1076 |
// Suppress warnings for malformed HTML |
| 1077 |
libxml_use_internal_errors(true); |
| 1078 |
$dom->loadHTML('<?xml encoding="UTF-8">' . $html); |
| 1079 |
libxml_clear_errors(); |
| 1080 |
|
| 1081 |
$xpath = new \DOMXPath($dom); |
| 1082 |
|
| 1083 |
// Extract title |
| 1084 |
$title_nodes = $xpath->query('//title'); |
| 1085 |
$title = $title_nodes->length > 0 ? trim($title_nodes->item(0)->textContent) : ''; |
| 1086 |
|
| 1087 |
// Extract meta description |
| 1088 |
$meta_desc_nodes = $xpath->query('//meta[@name="description"]/@content'); |
| 1089 |
$meta_description = $meta_desc_nodes->length > 0 ? trim($meta_desc_nodes->item(0)->textContent) : ''; |
| 1090 |
|
| 1091 |
// Extract headings (H1-H6) |
| 1092 |
$headings = []; |
| 1093 |
for ($i = 1; $i <= 6; $i++) { |
| 1094 |
$heading_nodes = $xpath->query("//h{$i}"); |
| 1095 |
foreach ($heading_nodes as $node) { |
| 1096 |
$text = trim($node->textContent); |
| 1097 |
if (!empty($text)) { |
| 1098 |
$headings["h{$i}"][] = $text; |
| 1099 |
} |
| 1100 |
} |
| 1101 |
} |
| 1102 |
|
| 1103 |
// Extract body text and calculate word count |
| 1104 |
$body_nodes = $xpath->query('//body'); |
| 1105 |
$body_text = ''; |
| 1106 |
if ($body_nodes->length > 0) { |
| 1107 |
$body_text = $this->extract_clean_text($body_nodes->item(0)); |
| 1108 |
} |
| 1109 |
|
| 1110 |
$word_count = str_word_count($body_text); |
| 1111 |
|
| 1112 |
// Extract meta keywords if present |
| 1113 |
$meta_keywords_nodes = $xpath->query('//meta[@name="keywords"]/@content'); |
| 1114 |
$meta_keywords = $meta_keywords_nodes->length > 0 ? trim($meta_keywords_nodes->item(0)->textContent) : ''; |
| 1115 |
|
| 1116 |
// Extract internal links count |
| 1117 |
$internal_links = $xpath->query('//a[starts-with(@href, "/") or contains(@href, "' . wp_parse_url($url, PHP_URL_HOST) . '")]'); |
| 1118 |
$internal_link_count = $internal_links->length; |
| 1119 |
|
| 1120 |
// Extract external links count |
| 1121 |
$external_links = $xpath->query('//a[starts-with(@href, "http") and not(contains(@href, "' . wp_parse_url($url, PHP_URL_HOST) . '"))]'); |
| 1122 |
$external_link_count = $external_links->length; |
| 1123 |
|
| 1124 |
// Extract images count and alt text analysis |
| 1125 |
$images = $xpath->query('//img'); |
| 1126 |
$image_count = $images->length; |
| 1127 |
$images_with_alt = $xpath->query('//img[@alt and @alt!=""]'); |
| 1128 |
$images_with_alt_count = $images_with_alt->length; |
| 1129 |
|
| 1130 |
// Extract schema markup |
| 1131 |
$schema_scripts = $xpath->query('//script[@type="application/ld+json"]'); |
| 1132 |
$has_schema = $schema_scripts->length > 0; |
| 1133 |
|
| 1134 |
// Extract last modified date if available |
| 1135 |
$last_modified_nodes = $xpath->query('//meta[@name="last-modified"]/@content | //meta[@property="article:modified_time"]/@content'); |
| 1136 |
$last_modified = $last_modified_nodes->length > 0 ? $last_modified_nodes->item(0)->textContent : ''; |
| 1137 |
|
| 1138 |
// Calculate readability metrics |
| 1139 |
$readability_score = $this->calculate_readability_score($body_text); |
| 1140 |
|
| 1141 |
// Extract keyword density for target keywords (if provided) |
| 1142 |
$keyword_density = $this->analyze_keyword_density($body_text, $title); |
| 1143 |
|
| 1144 |
// Detect content freshness indicators |
| 1145 |
$freshness_indicators = $this->detect_freshness_indicators($html, $body_text); |
| 1146 |
|
| 1147 |
return [ |
| 1148 |
'url' => $url, |
| 1149 |
'title' => $title, |
| 1150 |
'meta_description' => $meta_description, |
| 1151 |
'meta_keywords' => $meta_keywords, |
| 1152 |
'headings' => $headings, |
| 1153 |
'word_count' => $word_count, |
| 1154 |
'internal_links' => $internal_link_count, |
| 1155 |
'external_links' => $external_link_count, |
| 1156 |
'images' => [ |
| 1157 |
'total' => $image_count, |
| 1158 |
'with_alt' => $images_with_alt_count, |
| 1159 |
'alt_ratio' => $image_count > 0 ? round(($images_with_alt_count / $image_count) * 100, 1) : 0 |
| 1160 |
], |
| 1161 |
'seo' => [ |
| 1162 |
'has_schema' => $has_schema, |
| 1163 |
'title_length' => strlen($title), |
| 1164 |
'meta_desc_length' => strlen($meta_description), |
| 1165 |
'title_score' => $this->score_title_seo($title), |
| 1166 |
'meta_desc_score' => $this->score_meta_description($meta_description) |
| 1167 |
], |
| 1168 |
'content_quality' => [ |
| 1169 |
'readability_score' => $readability_score, |
| 1170 |
'keyword_density' => $keyword_density, |
| 1171 |
'freshness_indicators' => $freshness_indicators, |
| 1172 |
'content_depth' => $this->assess_content_depth($headings, $word_count) |
| 1173 |
], |
| 1174 |
'last_modified' => $last_modified, |
| 1175 |
'content_preview' => substr($body_text, 0, 500) . '...', |
| 1176 |
'analysis_timestamp' => current_time('mysql') |
| 1177 |
]; |
| 1178 |
} |
| 1179 |
|
| 1180 |
/** |
| 1181 |
* Extract clean text from DOM node, removing scripts and styles |
| 1182 |
* |
| 1183 |
* @param \DOMNode $node DOM node to extract text from |
| 1184 |
* @return string Clean text content |
| 1185 |
*/ |
| 1186 |
private function extract_clean_text(\DOMNode $node): string { |
| 1187 |
// Remove script and style elements |
| 1188 |
$xpath = new \DOMXPath($node->ownerDocument); |
| 1189 |
$scripts = $xpath->query('.//script | .//style', $node); |
| 1190 |
|
| 1191 |
foreach ($scripts as $script) { |
| 1192 |
$script->parentNode->removeChild($script); |
| 1193 |
} |
| 1194 |
|
| 1195 |
// Get text content and clean it up |
| 1196 |
$text = $node->textContent; |
| 1197 |
|
| 1198 |
// Remove extra whitespace and normalize |
| 1199 |
$text = preg_replace('/\s+/', ' ', $text); |
| 1200 |
$text = trim($text); |
| 1201 |
|
| 1202 |
return $text; |
| 1203 |
} |
| 1204 |
|
| 1205 |
/** |
| 1206 |
* Format competitor analysis for AI prompt |
| 1207 |
* |
| 1208 |
* @param string $url Competitor URL |
| 1209 |
* @param array $content_data Parsed content data |
| 1210 |
* @return string Formatted analysis |
| 1211 |
*/ |
| 1212 |
private function format_competitor_analysis(string $url, array $content_data): string { |
| 1213 |
$analysis = "=== COMPETITOR ANALYSIS ===\n"; |
| 1214 |
$analysis .= "URL: {$url}\n"; |
| 1215 |
$analysis .= "Title: {$content_data['title']} (Length: {$content_data['seo']['title_length']} chars, Score: {$content_data['seo']['title_score']['grade']})\n"; |
| 1216 |
|
| 1217 |
if (!empty($content_data['meta_description'])) { |
| 1218 |
$analysis .= "Meta Description: {$content_data['meta_description']} (Length: {$content_data['seo']['meta_desc_length']} chars, Score: {$content_data['seo']['meta_desc_score']['grade']})\n"; |
| 1219 |
} |
| 1220 |
|
| 1221 |
$analysis .= "\nCONTENT METRICS:\n"; |
| 1222 |
$analysis .= "- Word Count: {$content_data['word_count']} words\n"; |
| 1223 |
$analysis .= "- Content Depth: {$content_data['content_quality']['content_depth']['level']} (Score: {$content_data['content_quality']['content_depth']['score']}/100)\n"; |
| 1224 |
$analysis .= "- Readability: {$content_data['content_quality']['readability_score']['level']} (Score: {$content_data['content_quality']['readability_score']['score']}/100)\n"; |
| 1225 |
$analysis .= "- Internal Links: {$content_data['internal_links']}\n"; |
| 1226 |
$analysis .= "- External Links: {$content_data['external_links']}\n"; |
| 1227 |
$analysis .= "- Images: {$content_data['images']['total']} total, {$content_data['images']['with_alt']} with alt text ({$content_data['images']['alt_ratio']}%)\n"; |
| 1228 |
|
| 1229 |
// Add heading structure |
| 1230 |
if (!empty($content_data['headings'])) { |
| 1231 |
$analysis .= "\nCONTENT STRUCTURE:\n"; |
| 1232 |
foreach ($content_data['headings'] as $level => $headings) { |
| 1233 |
$analysis .= "- " . strtoupper($level) . " ({count}): " . implode(', ', array_slice($headings, 0, 3)); |
| 1234 |
if (count($headings) > 3) { |
| 1235 |
$analysis .= "... (+" . (count($headings) - 3) . " more)"; |
| 1236 |
} |
| 1237 |
$analysis .= "\n"; |
| 1238 |
} |
| 1239 |
} |
| 1240 |
|
| 1241 |
// Add SEO features |
| 1242 |
$analysis .= "\nSEO FEATURES:\n"; |
| 1243 |
$analysis .= "- Schema Markup: " . ($content_data['seo']['has_schema'] ? 'Yes' : 'No') . "\n"; |
| 1244 |
if (!empty($content_data['meta_keywords'])) { |
| 1245 |
$analysis .= "- Meta Keywords: {$content_data['meta_keywords']}\n"; |
| 1246 |
} |
| 1247 |
|
| 1248 |
// Add content quality insights |
| 1249 |
if (!empty($content_data['content_quality']['keyword_density']['top_keywords'])) { |
| 1250 |
$analysis .= "\nTOP KEYWORDS:\n"; |
| 1251 |
foreach (array_slice($content_data['content_quality']['keyword_density']['top_keywords'], 0, 5) as $kw) { |
| 1252 |
$analysis .= "- {$kw['keyword']}: {$kw['count']} times ({$kw['density']}%)\n"; |
| 1253 |
} |
| 1254 |
} |
| 1255 |
|
| 1256 |
// Add freshness indicators |
| 1257 |
if (!empty($content_data['content_quality']['freshness_indicators'])) { |
| 1258 |
$analysis .= "\nCONTENT FRESHNESS:\n"; |
| 1259 |
foreach ($content_data['content_quality']['freshness_indicators'] as $indicator) { |
| 1260 |
$analysis .= "- {$indicator}\n"; |
| 1261 |
} |
| 1262 |
} |
| 1263 |
|
| 1264 |
$analysis .= "\n" . str_repeat("=", 50) . "\n"; |
| 1265 |
|
| 1266 |
return $analysis; |
| 1267 |
} |
| 1268 |
|
| 1269 |
/** |
| 1270 |
* Get word count estimate based on content length |
| 1271 |
* |
| 1272 |
* @param string $content_length Content length setting |
| 1273 |
* @return int Estimated word count |
| 1274 |
*/ |
| 1275 |
private function get_word_count_estimate(string $content_length): int { |
| 1276 |
$estimates = [ |
| 1277 |
'short' => 650, |
| 1278 |
'medium' => 1250, |
| 1279 |
'long' => 2500 |
| 1280 |
]; |
| 1281 |
|
| 1282 |
return $estimates[$content_length] ?? 1250; |
| 1283 |
} |
| 1284 |
private function save_brief(array $brief_data): int { |
| 1285 |
global $wpdb; |
| 1286 |
|
| 1287 |
$table_name = $wpdb->prefix . 'thinkrank_content_briefs'; |
| 1288 |
|
| 1289 |
// Prepare data for insertion |
| 1290 |
$insert_data = [ |
| 1291 |
'user_id' => get_current_user_id(), |
| 1292 |
'title' => $brief_data['title'][0] ?? 'Untitled Brief', |
| 1293 |
'target_keywords' => wp_json_encode($brief_data['generation_params']['target_keywords'] ?? []), |
| 1294 |
'content_type' => $brief_data['generation_params']['content_type'] ?? 'blog_post', |
| 1295 |
'brief_data' => wp_json_encode($brief_data), |
| 1296 |
'created_at' => current_time('mysql'), |
| 1297 |
'updated_at' => current_time('mysql') |
| 1298 |
]; |
| 1299 |
|
| 1300 |
$insert_format = [ |
| 1301 |
'%d', // user_id |
| 1302 |
'%s', // title |
| 1303 |
'%s', // target_keywords |
| 1304 |
'%s', // content_type |
| 1305 |
'%s', // brief_data |
| 1306 |
'%s', // created_at |
| 1307 |
'%s' // updated_at |
| 1308 |
]; |
| 1309 |
|
| 1310 |
// phpcs:ignore WordPress.DB.DirectDatabaseQuery.DirectQuery,WordPress.DB.DirectDatabaseQuery.NoCaching, PluginCheck.Security.DirectDB.UnescapedDBParameter -- Content brief storage requires direct database access |
| 1311 |
$result = $wpdb->insert($table_name, $insert_data, $insert_format); |
| 1312 |
|
| 1313 |
if (false === $result) { |
| 1314 |
throw new \Exception('Failed to save content brief to database.'); |
| 1315 |
} |
| 1316 |
|
| 1317 |
/** |
| 1318 |
* Fires after a content brief is persisted. |
| 1319 |
* |
| 1320 |
* Analytics listens to drop its cached overview so the brief counts |
| 1321 |
* on the Usages page are not stale for a TTL. |
| 1322 |
* |
| 1323 |
* @since 2.2.1 |
| 1324 |
* |
| 1325 |
* @param int $brief_id Row id of the stored brief. |
| 1326 |
*/ |
| 1327 |
do_action('thinkrank_content_brief_created', (int) $wpdb->insert_id); |
| 1328 |
|
| 1329 |
return $wpdb->insert_id; |
| 1330 |
} |
| 1331 |
|
| 1332 |
/** |
| 1333 |
* Normalize brief data for React compatibility |
| 1334 |
* |
| 1335 |
* @param array $brief_data Brief data to normalize |
| 1336 |
* @return array Normalized brief data |
| 1337 |
*/ |
| 1338 |
private function normalize_brief_data(array $brief_data): array { |
| 1339 |
// Normalize focus_keyword_analysis |
| 1340 |
if (isset($brief_data['seo_recommendations']['focus_keyword_analysis'])) { |
| 1341 |
$brief_data['seo_recommendations']['focus_keyword_analysis'] = |
| 1342 |
$this->normalize_focus_keyword_analysis($brief_data['seo_recommendations']['focus_keyword_analysis']); |
| 1343 |
} |
| 1344 |
|
| 1345 |
// Normalize call_to_actions (convert objects to strings) |
| 1346 |
if (isset($brief_data['call_to_actions']) && is_array($brief_data['call_to_actions'])) { |
| 1347 |
$brief_data['call_to_actions'] = array_map(function($cta) { |
| 1348 |
if (is_array($cta) && isset($cta['text'])) { |
| 1349 |
return $cta['text'] . (isset($cta['placement']) ? ' (' . $cta['placement'] . ')' : ''); |
| 1350 |
} |
| 1351 |
return is_string($cta) ? $cta : ''; |
| 1352 |
}, $brief_data['call_to_actions']); |
| 1353 |
} |
| 1354 |
|
| 1355 |
// Normalize visual content image_recommendations (convert objects to strings) |
| 1356 |
if (isset($brief_data['visual_content']['image_recommendations']) && is_array($brief_data['visual_content']['image_recommendations'])) { |
| 1357 |
$brief_data['visual_content']['image_recommendations'] = array_map(function($rec) { |
| 1358 |
if (is_array($rec)) { |
| 1359 |
$text = ''; |
| 1360 |
if (isset($rec['type'])) { $text .= $rec['type'] . ': '; |
| 1361 |
} |
| 1362 |
if (isset($rec['description'])) { $text .= $rec['description']; |
| 1363 |
} |
| 1364 |
if (isset($rec['alt_text'])) { $text .= ' (Alt: ' . $rec['alt_text'] . ')'; |
| 1365 |
} |
| 1366 |
return $text ?: 'Image recommendation'; |
| 1367 |
} |
| 1368 |
return is_string($rec) ? $rec : 'Image recommendation'; |
| 1369 |
}, $brief_data['visual_content']['image_recommendations']); |
| 1370 |
} |
| 1371 |
|
| 1372 |
return $this->sanitize_brief_output($brief_data); |
| 1373 |
} |
| 1374 |
|
| 1375 |
/** |
| 1376 |
* Strip untrusted markup out of brief fields before they leave the server. |
| 1377 |
* |
| 1378 |
* Brief content crosses a trust boundary: it is assembled by an external AI |
| 1379 |
* provider from prompts that can include text fetched from competitor URLs. |
| 1380 |
* It was previously copied out of the decoded JSON verbatim and rendered in |
| 1381 |
* the admin SPA through dangerouslySetInnerHTML, so a malicious or |
| 1382 |
* prompt-injected response could execute script in the admin origin (#365). |
| 1383 |
* |
| 1384 |
* Runs on the read path as well as generation, so briefs stored before this |
| 1385 |
* fix are sanitized when they are loaded. |
| 1386 |
* |
| 1387 |
* @since 1.32.0 |
| 1388 |
* |
| 1389 |
* @param array $brief_data Brief data to sanitize. |
| 1390 |
* @return array Sanitized brief data. |
| 1391 |
*/ |
| 1392 |
private function sanitize_brief_output(array $brief_data): array { |
| 1393 |
foreach ($brief_data as $key => $value) { |
| 1394 |
// The raw provider response is debug output shown as plain text, and |
| 1395 |
// the generation params are our own values — leave both intact. |
| 1396 |
if ('raw_response' === $key || 'generation_params' === $key) { |
| 1397 |
continue; |
| 1398 |
} |
| 1399 |
|
| 1400 |
if ('content_body' === $key && is_string($value)) { |
| 1401 |
// Deliberately HTML: it is the drafted article and is rendered as |
| 1402 |
// markup. wp_kses_post() keeps normal post formatting while |
| 1403 |
// dropping script/style/iframe, event-handler attributes and |
| 1404 |
// javascript: URLs. |
| 1405 |
$brief_data[$key] = wp_kses_post($value); |
| 1406 |
continue; |
| 1407 |
} |
| 1408 |
|
| 1409 |
if (is_array($value)) { |
| 1410 |
$brief_data[$key] = $this->sanitize_brief_output($value); |
| 1411 |
} elseif (is_string($value)) { |
| 1412 |
// Every other field is plain text (headings, keywords, guidance). |
| 1413 |
// Markdown emphasis markers are preserved; HTML tags are not. |
| 1414 |
$brief_data[$key] = wp_strip_all_tags($value); |
| 1415 |
} |
| 1416 |
} |
| 1417 |
|
| 1418 |
return $brief_data; |
| 1419 |
} |
| 1420 |
|
| 1421 |
/** |
| 1422 |
* Get saved briefs for current user |
| 1423 |
* |
| 1424 |
* @param int $limit Number of briefs to retrieve |
| 1425 |
* @param int $offset Offset for pagination |
| 1426 |
* @return array Array of saved briefs |
| 1427 |
*/ |
| 1428 |
public function get_user_briefs(int $limit = 10, int $offset = 0): array { |
| 1429 |
global $wpdb; |
| 1430 |
|
| 1431 |
// Get table name and escape it properly (table names cannot be parameterized) |
| 1432 |
$table_name = esc_sql($wpdb->prefix . 'thinkrank_content_briefs'); |
| 1433 |
$user_id = get_current_user_id(); |
| 1434 |
|
| 1435 |
// phpcs:ignore WordPress.DB.DirectDatabaseQuery.DirectQuery,WordPress.DB.DirectDatabaseQuery.NoCaching, PluginCheck.Security.DirectDB.UnescapedDBParameter -- Content brief retrieval requires direct database access |
| 1436 |
$results = $wpdb->get_results( |
| 1437 |
$wpdb->prepare( |
| 1438 |
// phpcs:ignore WordPress.DB.PreparedSQL.InterpolatedNotPrepared, PluginCheck.Security.DirectDB.UnescapedDBParameter -- Table name is properly escaped using esc_sql() |
| 1439 |
"SELECT * FROM `{$table_name}` WHERE user_id = %d ORDER BY created_at DESC LIMIT %d OFFSET %d", |
| 1440 |
$user_id, |
| 1441 |
$limit, |
| 1442 |
$offset |
| 1443 |
), |
| 1444 |
ARRAY_A |
| 1445 |
); |
| 1446 |
|
| 1447 |
// $wpdb->get_results() returns null on a DB error; this method's return |
| 1448 |
// type is : array, so normalize before iterating/returning. |
| 1449 |
if (!is_array($results)) { |
| 1450 |
return []; |
| 1451 |
} |
| 1452 |
|
| 1453 |
// Decode JSON data and normalize for React compatibility |
| 1454 |
foreach ($results as &$brief) { |
| 1455 |
$brief = $this->hydrate_brief_row($brief); |
| 1456 |
} |
| 1457 |
unset($brief); |
| 1458 |
|
| 1459 |
return $results; |
| 1460 |
} |
| 1461 |
|
| 1462 |
/** |
| 1463 |
* Get a single saved brief by id, scoped to the current user. |
| 1464 |
* |
| 1465 |
* @param int $brief_id Brief ID. |
| 1466 |
* @return array|null Hydrated brief, or null if it doesn't exist or does not |
| 1467 |
* belong to the current user. |
| 1468 |
*/ |
| 1469 |
public function get_brief(int $brief_id): ?array { |
| 1470 |
global $wpdb; |
| 1471 |
|
| 1472 |
// Table names cannot be parameterized; escape it. |
| 1473 |
$table_name = esc_sql($wpdb->prefix . 'thinkrank_content_briefs'); |
| 1474 |
$user_id = get_current_user_id(); |
| 1475 |
|
| 1476 |
// phpcs:ignore WordPress.DB.DirectDatabaseQuery.DirectQuery,WordPress.DB.DirectDatabaseQuery.NoCaching, PluginCheck.Security.DirectDB.UnescapedDBParameter -- Content brief retrieval requires direct database access |
| 1477 |
$brief = $wpdb->get_row( |
| 1478 |
$wpdb->prepare( |
| 1479 |
// phpcs:ignore WordPress.DB.PreparedSQL.InterpolatedNotPrepared, PluginCheck.Security.DirectDB.UnescapedDBParameter -- Table name is properly escaped using esc_sql() |
| 1480 |
"SELECT * FROM `{$table_name}` WHERE id = %d AND user_id = %d LIMIT 1", |
| 1481 |
$brief_id, |
| 1482 |
$user_id |
| 1483 |
), |
| 1484 |
ARRAY_A |
| 1485 |
); |
| 1486 |
|
| 1487 |
if (!$brief) { |
| 1488 |
return null; |
| 1489 |
} |
| 1490 |
|
| 1491 |
return $this->hydrate_brief_row($brief); |
| 1492 |
} |
| 1493 |
|
| 1494 |
/** |
| 1495 |
* Decode + normalize a raw content-brief DB row for API/React consumption. |
| 1496 |
* |
| 1497 |
* @param array $brief Raw database row. |
| 1498 |
* @return array Hydrated brief. |
| 1499 |
*/ |
| 1500 |
private function hydrate_brief_row(array $brief): array { |
| 1501 |
$brief['target_keywords'] = json_decode($brief['target_keywords'], true); |
| 1502 |
$brief['brief_data'] = json_decode($brief['brief_data'], true); |
| 1503 |
|
| 1504 |
// Cast: the row comes from $wpdb, which returns every column as a |
| 1505 |
// string, and both helpers declare an int parameter. |
| 1506 |
$brief_id = (int) $brief['id']; |
| 1507 |
|
| 1508 |
// Retrieve raw response from ai_usage table |
| 1509 |
$brief['brief_data']['raw_response'] = $this->get_raw_response_for_brief($brief_id); |
| 1510 |
|
| 1511 |
// Update model with actual model used (if available in ai_usage table) |
| 1512 |
$actual_model = $this->get_actual_model_for_brief($brief_id); |
| 1513 |
if ($actual_model && isset($brief['brief_data']['generation_meta'])) { |
| 1514 |
$brief['brief_data']['generation_meta']['model'] = $actual_model; |
| 1515 |
} |
| 1516 |
|
| 1517 |
// Apply normalization to existing briefs to ensure React compatibility |
| 1518 |
$brief['brief_data'] = $this->normalize_brief_data($brief['brief_data']); |
| 1519 |
|
| 1520 |
return $brief; |
| 1521 |
} |
| 1522 |
|
| 1523 |
/** |
| 1524 |
* Delete brief |
| 1525 |
* |
| 1526 |
* @param int $brief_id Brief ID to delete |
| 1527 |
* @return bool Success status |
| 1528 |
*/ |
| 1529 |
public function delete_brief(int $brief_id): bool { |
| 1530 |
global $wpdb; |
| 1531 |
|
| 1532 |
$table_name = $wpdb->prefix . 'thinkrank_content_briefs'; |
| 1533 |
$user_id = get_current_user_id(); |
| 1534 |
|
| 1535 |
// phpcs:ignore WordPress.DB.DirectDatabaseQuery.DirectQuery,WordPress.DB.DirectDatabaseQuery.NoCaching, PluginCheck.Security.DirectDB.UnescapedDBParameter -- Content brief deletion requires direct database access |
| 1536 |
$result = $wpdb->delete( |
| 1537 |
$table_name, |
| 1538 |
[ |
| 1539 |
'id' => $brief_id, |
| 1540 |
'user_id' => $user_id |
| 1541 |
], |
| 1542 |
['%d', '%d'] |
| 1543 |
); |
| 1544 |
|
| 1545 |
return $result !== false; |
| 1546 |
} |
| 1547 |
|
| 1548 |
/** |
| 1549 |
* Calculate readability score using Flesch Reading Ease |
| 1550 |
* |
| 1551 |
* @param string $text Text to analyze |
| 1552 |
* @return array Readability metrics |
| 1553 |
*/ |
| 1554 |
private function calculate_readability_score(string $text): array { |
| 1555 |
if (empty($text)) { |
| 1556 |
return ['score' => 0, 'level' => 'Unknown', 'grade' => 'N/A']; |
| 1557 |
} |
| 1558 |
|
| 1559 |
// Count sentences (approximate) |
| 1560 |
$sentences = preg_split('/[.!?]+/', $text); |
| 1561 |
$sentence_count = count(array_filter($sentences, function($s) { return trim($s) !== ''; |
| 1562 |
})); |
| 1563 |
|
| 1564 |
// Count words |
| 1565 |
$word_count = str_word_count($text); |
| 1566 |
|
| 1567 |
// Count syllables (approximate) |
| 1568 |
$syllable_count = $this->count_syllables($text); |
| 1569 |
|
| 1570 |
if ($sentence_count === 0 || $word_count === 0) { |
| 1571 |
return ['score' => 0, 'level' => 'Unknown', 'grade' => 'N/A']; |
| 1572 |
} |
| 1573 |
|
| 1574 |
// Flesch Reading Ease formula |
| 1575 |
$avg_sentence_length = $word_count / $sentence_count; |
| 1576 |
$avg_syllables_per_word = $syllable_count / $word_count; |
| 1577 |
|
| 1578 |
$flesch_score = 206.835 - (1.015 * $avg_sentence_length) - (84.6 * $avg_syllables_per_word); |
| 1579 |
$flesch_score = max(0, min(100, $flesch_score)); // Clamp between 0-100 |
| 1580 |
|
| 1581 |
// Determine reading level |
| 1582 |
if ($flesch_score >= 90) { |
| 1583 |
$level = 'Very Easy'; |
| 1584 |
$grade = '5th grade'; |
| 1585 |
} elseif ($flesch_score >= 80) { |
| 1586 |
$level = 'Easy'; |
| 1587 |
$grade = '6th grade'; |
| 1588 |
} elseif ($flesch_score >= 70) { |
| 1589 |
$level = 'Fairly Easy'; |
| 1590 |
$grade = '7th grade'; |
| 1591 |
} elseif ($flesch_score >= 60) { |
| 1592 |
$level = 'Standard'; |
| 1593 |
$grade = '8th-9th grade'; |
| 1594 |
} elseif ($flesch_score >= 50) { |
| 1595 |
$level = 'Fairly Difficult'; |
| 1596 |
$grade = '10th-12th grade'; |
| 1597 |
} elseif ($flesch_score >= 30) { |
| 1598 |
$level = 'Difficult'; |
| 1599 |
$grade = 'College level'; |
| 1600 |
} else { |
| 1601 |
$level = 'Very Difficult'; |
| 1602 |
$grade = 'Graduate level'; |
| 1603 |
} |
| 1604 |
|
| 1605 |
return [ |
| 1606 |
'score' => round($flesch_score, 1), |
| 1607 |
'level' => $level, |
| 1608 |
'grade' => $grade |
| 1609 |
]; |
| 1610 |
} |
| 1611 |
|
| 1612 |
/** |
| 1613 |
* Count syllables in text (approximate) |
| 1614 |
* |
| 1615 |
* @param string $text Text to analyze |
| 1616 |
* @return int Syllable count |
| 1617 |
*/ |
| 1618 |
private function count_syllables(string $text): int { |
| 1619 |
$words = str_word_count(strtolower($text), 1); |
| 1620 |
$syllable_count = 0; |
| 1621 |
|
| 1622 |
foreach ($words as $word) { |
| 1623 |
$syllable_count += $this->count_word_syllables($word); |
| 1624 |
} |
| 1625 |
|
| 1626 |
return max(1, $syllable_count); // At least 1 syllable |
| 1627 |
} |
| 1628 |
|
| 1629 |
/** |
| 1630 |
* Count syllables in a single word |
| 1631 |
* |
| 1632 |
* @param string $word Word to analyze |
| 1633 |
* @return int Syllable count |
| 1634 |
*/ |
| 1635 |
private function count_word_syllables(string $word): int { |
| 1636 |
$word = strtolower($word); |
| 1637 |
$vowels = 'aeiouy'; |
| 1638 |
$syllable_count = 0; |
| 1639 |
$previous_was_vowel = false; |
| 1640 |
|
| 1641 |
for ($i = 0, $len = strlen($word); $i < $len; $i++) { |
| 1642 |
$is_vowel = strpos($vowels, $word[$i]) !== false; |
| 1643 |
if ($is_vowel && !$previous_was_vowel) { |
| 1644 |
$syllable_count++; |
| 1645 |
} |
| 1646 |
$previous_was_vowel = $is_vowel; |
| 1647 |
} |
| 1648 |
|
| 1649 |
// Handle silent 'e' |
| 1650 |
if (substr($word, -1) === 'e' && $syllable_count > 1) { |
| 1651 |
$syllable_count--; |
| 1652 |
} |
| 1653 |
|
| 1654 |
return max(1, $syllable_count); |
| 1655 |
} |
| 1656 |
|
| 1657 |
/** |
| 1658 |
* Analyze keyword density in content |
| 1659 |
* |
| 1660 |
* @param string $text Content text |
| 1661 |
* @param string $title Page title |
| 1662 |
* @return array Keyword analysis |
| 1663 |
*/ |
| 1664 |
private function analyze_keyword_density(string $text, string $title): array { |
| 1665 |
$combined_text = strtolower($title . ' ' . $text); |
| 1666 |
$words = str_word_count($combined_text, 1); |
| 1667 |
$total_words = count($words); |
| 1668 |
|
| 1669 |
if ($total_words === 0) { |
| 1670 |
return ['top_keywords' => [], 'total_words' => 0]; |
| 1671 |
} |
| 1672 |
|
| 1673 |
// Count word frequency |
| 1674 |
$word_counts = array_count_values($words); |
| 1675 |
|
| 1676 |
// Filter out common stop words |
| 1677 |
$stop_words = ['the', 'a', 'an', 'and', 'or', 'but', 'in', 'on', 'at', 'to', 'for', 'of', 'with', 'by', 'is', 'are', 'was', 'were', 'be', 'been', 'have', 'has', 'had', 'do', 'does', 'did', 'will', 'would', 'could', 'should', 'may', 'might', 'must', 'can', 'this', 'that', 'these', 'those', 'i', 'you', 'he', 'she', 'it', 'we', 'they', 'me', 'him', 'her', 'us', 'them']; |
| 1678 |
|
| 1679 |
foreach ($stop_words as $stop_word) { |
| 1680 |
unset($word_counts[$stop_word]); |
| 1681 |
} |
| 1682 |
|
| 1683 |
// Filter out single characters and numbers |
| 1684 |
$word_counts = array_filter($word_counts, function($count, $word) { |
| 1685 |
return strlen($word) > 2 && !is_numeric($word) && $count > 1; |
| 1686 |
}, ARRAY_FILTER_USE_BOTH); |
| 1687 |
|
| 1688 |
// Sort by frequency |
| 1689 |
arsort($word_counts); |
| 1690 |
|
| 1691 |
// Calculate density and format results |
| 1692 |
$top_keywords = []; |
| 1693 |
foreach (array_slice($word_counts, 0, 10, true) as $word => $count) { |
| 1694 |
$density = round(($count / $total_words) * 100, 2); |
| 1695 |
$top_keywords[] = [ |
| 1696 |
'keyword' => $word, |
| 1697 |
'count' => $count, |
| 1698 |
'density' => $density |
| 1699 |
]; |
| 1700 |
} |
| 1701 |
|
| 1702 |
return [ |
| 1703 |
'top_keywords' => $top_keywords, |
| 1704 |
'total_words' => $total_words |
| 1705 |
]; |
| 1706 |
} |
| 1707 |
|
| 1708 |
/** |
| 1709 |
* Detect content freshness indicators |
| 1710 |
* |
| 1711 |
* @param string $html Full HTML content |
| 1712 |
* @param string $text Body text |
| 1713 |
* @return array Freshness indicators |
| 1714 |
*/ |
| 1715 |
private function detect_freshness_indicators(string $html, string $text): array { |
| 1716 |
$indicators = []; |
| 1717 |
|
| 1718 |
// Check for date patterns in content |
| 1719 |
if (preg_match('/\b(updated|revised|modified|published).*?(\d{4}|\d{1,2}\/\d{1,2}\/\d{2,4})/i', $text)) { |
| 1720 |
$indicators[] = 'Contains recent update dates'; |
| 1721 |
} |
| 1722 |
|
| 1723 |
// Check for current year references |
| 1724 |
$current_year = gmdate('Y'); |
| 1725 |
if (strpos($text, $current_year) !== false) { |
| 1726 |
$indicators[] = "References current year ({$current_year})"; |
| 1727 |
} |
| 1728 |
|
| 1729 |
// Check for "latest", "new", "recent" keywords |
| 1730 |
if (preg_match('/\b(latest|newest|recent|updated|current|modern|today)\b/i', $text)) { |
| 1731 |
$indicators[] = 'Uses freshness keywords'; |
| 1732 |
} |
| 1733 |
|
| 1734 |
// Check for structured data with dates |
| 1735 |
if (preg_match('/"dateModified"|"datePublished"/i', $html)) { |
| 1736 |
$indicators[] = 'Has structured date metadata'; |
| 1737 |
} |
| 1738 |
|
| 1739 |
return $indicators; |
| 1740 |
} |
| 1741 |
|
| 1742 |
/** |
| 1743 |
* Score title for SEO effectiveness |
| 1744 |
* |
| 1745 |
* @param string $title Page title |
| 1746 |
* @return array Title scoring |
| 1747 |
*/ |
| 1748 |
private function score_title_seo(string $title): array { |
| 1749 |
$score = 0; |
| 1750 |
$max_score = 100; |
| 1751 |
$feedback = []; |
| 1752 |
|
| 1753 |
// Length check (optimal: 50-60 characters) |
| 1754 |
$length = strlen($title); |
| 1755 |
if ($length >= 50 && $length <= 60) { |
| 1756 |
$score += 25; |
| 1757 |
$feedback[] = 'Good length (50-60 chars)'; |
| 1758 |
} elseif ($length >= 40 && $length <= 70) { |
| 1759 |
$score += 15; |
| 1760 |
$feedback[] = 'Acceptable length'; |
| 1761 |
} else { |
| 1762 |
$feedback[] = $length < 40 ? 'Too short (under 40 chars)' : 'Too long (over 70 chars)'; |
| 1763 |
} |
| 1764 |
|
| 1765 |
// Word count (optimal: 5-9 words) |
| 1766 |
$word_count = str_word_count($title); |
| 1767 |
if ($word_count >= 5 && $word_count <= 9) { |
| 1768 |
$score += 20; |
| 1769 |
$feedback[] = 'Good word count'; |
| 1770 |
} elseif ($word_count >= 3 && $word_count <= 12) { |
| 1771 |
$score += 10; |
| 1772 |
$feedback[] = 'Acceptable word count'; |
| 1773 |
} else { |
| 1774 |
$feedback[] = $word_count < 3 ? 'Too few words' : 'Too many words'; |
| 1775 |
} |
| 1776 |
|
| 1777 |
// Check for power words |
| 1778 |
$power_words = ['ultimate', 'complete', 'guide', 'best', 'top', 'essential', 'proven', 'expert', 'advanced', 'beginner']; |
| 1779 |
$has_power_words = false; |
| 1780 |
foreach ($power_words as $power_word) { |
| 1781 |
if (stripos($title, $power_word) !== false) { |
| 1782 |
$has_power_words = true; |
| 1783 |
break; |
| 1784 |
} |
| 1785 |
} |
| 1786 |
if ($has_power_words) { |
| 1787 |
$score += 15; |
| 1788 |
$feedback[] = 'Contains power words'; |
| 1789 |
} |
| 1790 |
|
| 1791 |
// Check for numbers |
| 1792 |
if (preg_match('/\d+/', $title)) { |
| 1793 |
$score += 10; |
| 1794 |
$feedback[] = 'Contains numbers'; |
| 1795 |
} |
| 1796 |
|
| 1797 |
// Check for emotional triggers |
| 1798 |
$emotional_words = ['amazing', 'incredible', 'shocking', 'secret', 'revealed', 'proven', 'guaranteed']; |
| 1799 |
$has_emotional_words = false; |
| 1800 |
foreach ($emotional_words as $emotional_word) { |
| 1801 |
if (stripos($title, $emotional_word) !== false) { |
| 1802 |
$has_emotional_words = true; |
| 1803 |
break; |
| 1804 |
} |
| 1805 |
} |
| 1806 |
if ($has_emotional_words) { |
| 1807 |
$score += 10; |
| 1808 |
$feedback[] = 'Contains emotional triggers'; |
| 1809 |
} |
| 1810 |
|
| 1811 |
// Uniqueness check (avoid generic titles) |
| 1812 |
$generic_patterns = ['untitled', 'new page', 'home', 'welcome']; |
| 1813 |
$is_generic = false; |
| 1814 |
foreach ($generic_patterns as $pattern) { |
| 1815 |
if (stripos($title, $pattern) !== false) { |
| 1816 |
$is_generic = true; |
| 1817 |
break; |
| 1818 |
} |
| 1819 |
} |
| 1820 |
if (!$is_generic) { |
| 1821 |
$score += 20; |
| 1822 |
$feedback[] = 'Appears unique'; |
| 1823 |
} else { |
| 1824 |
$feedback[] = 'Appears generic'; |
| 1825 |
} |
| 1826 |
|
| 1827 |
return [ |
| 1828 |
'score' => min($score, $max_score), |
| 1829 |
'max_score' => $max_score, |
| 1830 |
'grade' => $this->get_grade_from_score($score), |
| 1831 |
'feedback' => $feedback |
| 1832 |
]; |
| 1833 |
} |
| 1834 |
|
| 1835 |
/** |
| 1836 |
* Score meta description for SEO effectiveness |
| 1837 |
* |
| 1838 |
* @param string $meta_desc Meta description |
| 1839 |
* @return array Meta description scoring |
| 1840 |
*/ |
| 1841 |
private function score_meta_description(string $meta_desc): array { |
| 1842 |
$score = 0; |
| 1843 |
$max_score = 100; |
| 1844 |
$feedback = []; |
| 1845 |
|
| 1846 |
if (empty($meta_desc)) { |
| 1847 |
return [ |
| 1848 |
'score' => 0, |
| 1849 |
'max_score' => $max_score, |
| 1850 |
'grade' => 'F', |
| 1851 |
'feedback' => ['No meta description found'] |
| 1852 |
]; |
| 1853 |
} |
| 1854 |
|
| 1855 |
// Length check (optimal: 150-160 characters) |
| 1856 |
$length = strlen($meta_desc); |
| 1857 |
if ($length >= 150 && $length <= 160) { |
| 1858 |
$score += 30; |
| 1859 |
$feedback[] = 'Optimal length (150-160 chars)'; |
| 1860 |
} elseif ($length >= 120 && $length <= 170) { |
| 1861 |
$score += 20; |
| 1862 |
$feedback[] = 'Good length'; |
| 1863 |
} elseif ($length >= 100 && $length <= 180) { |
| 1864 |
$score += 10; |
| 1865 |
$feedback[] = 'Acceptable length'; |
| 1866 |
} else { |
| 1867 |
$feedback[] = $length < 100 ? 'Too short (under 100 chars)' : 'Too long (over 180 chars)'; |
| 1868 |
} |
| 1869 |
|
| 1870 |
// Check for call-to-action |
| 1871 |
$cta_words = ['learn', 'discover', 'find out', 'get', 'download', 'try', 'start', 'join', 'sign up', 'contact', 'buy', 'shop']; |
| 1872 |
$has_cta = false; |
| 1873 |
foreach ($cta_words as $cta_word) { |
| 1874 |
if (stripos($meta_desc, $cta_word) !== false) { |
| 1875 |
$has_cta = true; |
| 1876 |
break; |
| 1877 |
} |
| 1878 |
} |
| 1879 |
if ($has_cta) { |
| 1880 |
$score += 20; |
| 1881 |
$feedback[] = 'Contains call-to-action'; |
| 1882 |
} |
| 1883 |
|
| 1884 |
// Check for unique selling proposition |
| 1885 |
$usp_words = ['best', 'top', 'leading', 'expert', 'professional', 'trusted', 'proven', 'award-winning']; |
| 1886 |
$has_usp = false; |
| 1887 |
foreach ($usp_words as $usp_word) { |
| 1888 |
if (stripos($meta_desc, $usp_word) !== false) { |
| 1889 |
$has_usp = true; |
| 1890 |
break; |
| 1891 |
} |
| 1892 |
} |
| 1893 |
if ($has_usp) { |
| 1894 |
$score += 15; |
| 1895 |
$feedback[] = 'Contains unique selling proposition'; |
| 1896 |
} |
| 1897 |
|
| 1898 |
// Check for benefits/value proposition |
| 1899 |
$benefit_words = ['save', 'improve', 'increase', 'boost', 'enhance', 'optimize', 'maximize', 'reduce', 'eliminate']; |
| 1900 |
$has_benefits = false; |
| 1901 |
foreach ($benefit_words as $benefit_word) { |
| 1902 |
if (stripos($meta_desc, $benefit_word) !== false) { |
| 1903 |
$has_benefits = true; |
| 1904 |
break; |
| 1905 |
} |
| 1906 |
} |
| 1907 |
if ($has_benefits) { |
| 1908 |
$score += 15; |
| 1909 |
$feedback[] = 'Highlights benefits'; |
| 1910 |
} |
| 1911 |
|
| 1912 |
// Readability check |
| 1913 |
$sentences = preg_split('/[.!?]+/', $meta_desc); |
| 1914 |
$sentence_count = count(array_filter($sentences, function($s) { return trim($s) !== ''; |
| 1915 |
})); |
| 1916 |
if ($sentence_count >= 1 && $sentence_count <= 3) { |
| 1917 |
$score += 20; |
| 1918 |
$feedback[] = 'Good sentence structure'; |
| 1919 |
} else { |
| 1920 |
$feedback[] = $sentence_count === 0 ? 'No clear sentences' : 'Too many sentences'; |
| 1921 |
} |
| 1922 |
|
| 1923 |
return [ |
| 1924 |
'score' => min($score, $max_score), |
| 1925 |
'max_score' => $max_score, |
| 1926 |
'grade' => $this->get_grade_from_score($score), |
| 1927 |
'feedback' => $feedback |
| 1928 |
]; |
| 1929 |
} |
| 1930 |
|
| 1931 |
/** |
| 1932 |
* Assess content depth based on structure and length |
| 1933 |
* |
| 1934 |
* @param array $headings Heading structure |
| 1935 |
* @param int $word_count Word count |
| 1936 |
* @return array Content depth assessment |
| 1937 |
*/ |
| 1938 |
private function assess_content_depth(array $headings, int $word_count): array { |
| 1939 |
$depth_score = 0; |
| 1940 |
$max_score = 100; |
| 1941 |
|
| 1942 |
// Word count scoring (more words = more depth) |
| 1943 |
if ($word_count >= 2000) { |
| 1944 |
$depth_score += 40; |
| 1945 |
} elseif ($word_count >= 1000) { |
| 1946 |
$depth_score += 30; |
| 1947 |
} elseif ($word_count >= 500) { |
| 1948 |
$depth_score += 20; |
| 1949 |
} elseif ($word_count >= 300) { |
| 1950 |
$depth_score += 10; |
| 1951 |
} |
| 1952 |
|
| 1953 |
// Heading structure scoring |
| 1954 |
$total_headings = 0; |
| 1955 |
$heading_levels = 0; |
| 1956 |
foreach ($headings as $level => $level_headings) { |
| 1957 |
$total_headings += count($level_headings); |
| 1958 |
$heading_levels++; |
| 1959 |
} |
| 1960 |
|
| 1961 |
if ($total_headings >= 10) { |
| 1962 |
$depth_score += 25; |
| 1963 |
} elseif ($total_headings >= 5) { |
| 1964 |
$depth_score += 15; |
| 1965 |
} elseif ($total_headings >= 3) { |
| 1966 |
$depth_score += 10; |
| 1967 |
} |
| 1968 |
|
| 1969 |
// Heading hierarchy scoring |
| 1970 |
if ($heading_levels >= 3) { |
| 1971 |
$depth_score += 20; |
| 1972 |
} elseif ($heading_levels >= 2) { |
| 1973 |
$depth_score += 15; |
| 1974 |
} |
| 1975 |
|
| 1976 |
// Content structure bonus |
| 1977 |
if (isset($headings['h1']) && isset($headings['h2'])) { |
| 1978 |
$depth_score += 15; |
| 1979 |
} |
| 1980 |
|
| 1981 |
// Determine depth level |
| 1982 |
if ($depth_score >= 80) { |
| 1983 |
$level = 'Comprehensive'; |
| 1984 |
} elseif ($depth_score >= 60) { |
| 1985 |
$level = 'Detailed'; |
| 1986 |
} elseif ($depth_score >= 40) { |
| 1987 |
$level = 'Moderate'; |
| 1988 |
} elseif ($depth_score >= 20) { |
| 1989 |
$level = 'Basic'; |
| 1990 |
} else { |
| 1991 |
$level = 'Shallow'; |
| 1992 |
} |
| 1993 |
|
| 1994 |
return [ |
| 1995 |
'score' => min($depth_score, $max_score), |
| 1996 |
'level' => $level, |
| 1997 |
'word_count' => $word_count, |
| 1998 |
'total_headings' => $total_headings, |
| 1999 |
'heading_levels' => $heading_levels |
| 2000 |
]; |
| 2001 |
} |
| 2002 |
|
| 2003 |
/** |
| 2004 |
* Convert numeric score to letter grade |
| 2005 |
* |
| 2006 |
* @param int $score Numeric score |
| 2007 |
* @return string Letter grade |
| 2008 |
*/ |
| 2009 |
private function get_grade_from_score(int $score): string { |
| 2010 |
if ($score >= 90) { return 'A'; |
| 2011 |
} |
| 2012 |
if ($score >= 80) { return 'B'; |
| 2013 |
} |
| 2014 |
if ($score >= 70) { return 'C'; |
| 2015 |
} |
| 2016 |
if ($score >= 60) { return 'D'; |
| 2017 |
} |
| 2018 |
return 'F'; |
| 2019 |
} |
| 2020 |
|
| 2021 |
/** |
| 2022 |
* Log AI usage for analytics |
| 2023 |
* |
| 2024 |
* @param int $user_id User ID |
| 2025 |
* @param string $action Action performed |
| 2026 |
* @param int $tokens_used Tokens consumed |
| 2027 |
* @param int|null $post_id Related post/brief ID |
| 2028 |
* @param string|null $raw_response Raw AI response for debugging |
| 2029 |
* @param string|null $actual_model Actual model used (from response) |
| 2030 |
* @return int Usage record ID |
| 2031 |
*/ |
| 2032 |
private function log_ai_usage(int $user_id, string $action, int $tokens_used, ?int $post_id = null, ?string $raw_response = null, ?string $actual_model = null): int { |
| 2033 |
global $wpdb; |
| 2034 |
|
| 2035 |
$table_name = $wpdb->prefix . 'thinkrank_ai_usage'; |
| 2036 |
|
| 2037 |
$metadata = []; |
| 2038 |
if ($raw_response) { |
| 2039 |
$metadata['raw_response'] = $raw_response; |
| 2040 |
} |
| 2041 |
if ($actual_model) { |
| 2042 |
$metadata['actual_model'] = $actual_model; |
| 2043 |
} |
| 2044 |
|
| 2045 |
// phpcs:ignore WordPress.DB.DirectDatabaseQuery.DirectQuery,WordPress.DB.DirectDatabaseQuery.NoCaching, PluginCheck.Security.DirectDB.UnescapedDBParameter -- AI usage logging requires direct database access |
| 2046 |
$wpdb->insert( |
| 2047 |
$table_name, |
| 2048 |
[ |
| 2049 |
'user_id' => $user_id, |
| 2050 |
'action' => $action, |
| 2051 |
'tokens_used' => $tokens_used, |
| 2052 |
'provider' => $this->settings->get('ai_provider', Settings::AI_PROVIDER_NONE), |
| 2053 |
'post_id' => $post_id, |
| 2054 |
'metadata' => !empty($metadata) ? wp_json_encode($metadata) : null, |
| 2055 |
'created_at' => current_time('mysql'), |
| 2056 |
], |
| 2057 |
['%d', '%s', '%d', '%s', '%d', '%s', '%s'] |
| 2058 |
); |
| 2059 |
|
| 2060 |
/** |
| 2061 |
* Fires after an AI usage row is recorded. |
| 2062 |
* |
| 2063 |
* @since 2.2.1 |
| 2064 |
* |
| 2065 |
* @param int $user_id User the usage was recorded against. |
| 2066 |
*/ |
| 2067 |
do_action('thinkrank_ai_usage_logged', $user_id); |
| 2068 |
|
| 2069 |
return $wpdb->insert_id; |
| 2070 |
} |
| 2071 |
|
| 2072 |
/** |
| 2073 |
* Get raw AI response for a brief from ai_usage table |
| 2074 |
* |
| 2075 |
* @param int $brief_id Brief ID |
| 2076 |
* @return string Raw AI response or empty string if not found |
| 2077 |
*/ |
| 2078 |
private function get_raw_response_for_brief(int $brief_id): string { |
| 2079 |
global $wpdb; |
| 2080 |
|
| 2081 |
$table_name = $wpdb->prefix . 'thinkrank_ai_usage'; |
| 2082 |
|
| 2083 |
// phpcs:ignore WordPress.DB.DirectDatabaseQuery.DirectQuery,WordPress.DB.DirectDatabaseQuery.NoCaching, PluginCheck.Security.DirectDB.UnescapedDBParameter -- AI usage retrieval requires direct database access |
| 2084 |
$result = $wpdb->get_var( |
| 2085 |
// phpcs:ignore WordPress.DB.PreparedSQL.InterpolatedNotPrepared, PluginCheck.Security.DirectDB.UnescapedDBParameter -- Table name is validated with WordPress prefix |
| 2086 |
$wpdb->prepare( |
| 2087 |
// phpcs:ignore WordPress.DB.PreparedSQL.InterpolatedNotPrepared, PluginCheck.Security.DirectDB.UnescapedDBParameter -- Table name is validated with WordPress prefix |
| 2088 |
"SELECT metadata FROM `{$table_name}` WHERE post_id = %d AND action = 'content_brief' ORDER BY created_at DESC LIMIT 1", |
| 2089 |
$brief_id |
| 2090 |
) |
| 2091 |
); |
| 2092 |
|
| 2093 |
if ($result) { |
| 2094 |
$metadata = json_decode($result, true); |
| 2095 |
return $metadata['raw_response'] ?? ''; |
| 2096 |
} |
| 2097 |
|
| 2098 |
return ''; |
| 2099 |
} |
| 2100 |
|
| 2101 |
/** |
| 2102 |
* Get actual model used for a brief from ai_usage table |
| 2103 |
* |
| 2104 |
* @param int $brief_id Brief ID |
| 2105 |
* @return string|null Actual model used or null if not found |
| 2106 |
*/ |
| 2107 |
private function get_actual_model_for_brief(int $brief_id): ?string { |
| 2108 |
global $wpdb; |
| 2109 |
|
| 2110 |
$table_name = $wpdb->prefix . 'thinkrank_ai_usage'; |
| 2111 |
|
| 2112 |
// phpcs:ignore WordPress.DB.DirectDatabaseQuery.DirectQuery,WordPress.DB.DirectDatabaseQuery.NoCaching, PluginCheck.Security.DirectDB.UnescapedDBParameter -- AI usage retrieval requires direct database access |
| 2113 |
$result = $wpdb->get_var( |
| 2114 |
// phpcs:ignore WordPress.DB.PreparedSQL.InterpolatedNotPrepared, PluginCheck.Security.DirectDB.UnescapedDBParameter -- Table name is validated with WordPress prefix |
| 2115 |
$wpdb->prepare( |
| 2116 |
// phpcs:ignore WordPress.DB.PreparedSQL.InterpolatedNotPrepared, PluginCheck.Security.DirectDB.UnescapedDBParameter -- Table name is validated with WordPress prefix |
| 2117 |
"SELECT metadata FROM `{$table_name}` WHERE post_id = %d AND action = 'content_brief' ORDER BY created_at DESC LIMIT 1", |
| 2118 |
$brief_id |
| 2119 |
) |
| 2120 |
); |
| 2121 |
|
| 2122 |
if ($result) { |
| 2123 |
$metadata = json_decode($result, true); |
| 2124 |
return $metadata['actual_model'] ?? null; |
| 2125 |
} |
| 2126 |
|
| 2127 |
return null; |
| 2128 |
} |
| 2129 |
|
| 2130 |
/** |
| 2131 |
* Get Prompt Builder instance |
| 2132 |
* |
| 2133 |
* @since 1.0.0 |
| 2134 |
* |
| 2135 |
* @return \ThinkRank\AI\Prompt_Builder Prompt Builder instance |
| 2136 |
*/ |
| 2137 |
private function get_prompt_builder(): \ThinkRank\AI\Prompt_Builder { |
| 2138 |
if (!class_exists('ThinkRank\\AI\\Prompt_Builder')) { |
| 2139 |
require_once THINKRANK_PLUGIN_DIR . 'includes/ai/class-prompt-builder.php'; |
| 2140 |
} |
| 2141 |
return new \ThinkRank\AI\Prompt_Builder(); |
| 2142 |
} |
| 2143 |
} |
| 2144 |
|
