Code Coverage |
||||||||||
Lines |
Functions and Methods |
Classes and Traits |
||||||||
| Total | |
87.10% |
108 / 124 |
|
81.82% |
9 / 11 |
CRAP | |
0.00% |
0 / 1 |
| Ai_Crawlers | |
87.10% |
108 / 124 |
|
81.82% |
9 / 11 |
25.24 | |
0.00% |
0 / 1 |
| init | |
0.00% |
0 / 1 |
|
0.00% |
0 / 1 |
2 | |||
| get_catalog | |
100.00% |
63 / 63 |
|
100.00% |
1 / 1 |
1 | |||
| is_blocked_by_default | |
100.00% |
2 / 2 |
|
100.00% |
1 / 1 |
2 | |||
| get_overrides | |
100.00% |
13 / 13 |
|
100.00% |
1 / 1 |
5 | |||
| is_blocked | |
100.00% |
4 / 4 |
|
100.00% |
1 / 1 |
2 | |||
| get_blocked_slugs | |
100.00% |
9 / 9 |
|
100.00% |
1 / 1 |
4 | |||
| search_engines_allowed | |
100.00% |
1 / 1 |
|
100.00% |
1 / 1 |
1 | |||
| is_crawl_restricted_subdomain | |
100.00% |
4 / 4 |
|
100.00% |
1 / 1 |
2 | |||
| has_static_robots_txt | |
100.00% |
1 / 1 |
|
100.00% |
1 / 1 |
1 | |||
| append_directives | |
100.00% |
11 / 11 |
|
100.00% |
1 / 1 |
3 | |||
| get_bootstrap_data | |
0.00% |
0 / 15 |
|
0.00% |
0 / 1 |
6 | |||
| 1 | <?php |
| 2 | /** |
| 3 | * AI crawler access controls. |
| 4 | * |
| 5 | * Lets a site owner allow or block individual AI crawlers (training and |
| 6 | * answer-engine bots) by appending per-user-agent `Disallow` rules to the |
| 7 | * WordPress-generated robots.txt. |
| 8 | * |
| 9 | * Persistence uses a STORE-INTENT model rather than a literal blocked list. The |
| 10 | * durable option `jetpack_seo_ai_crawler_overrides` holds only *deviations* from |
| 11 | * each bot's default policy (training crawlers blocked, answer engines allowed); |
| 12 | * a bot with no stored override falls back to its default. This keeps the stored |
| 13 | * map sparse and means newly added training crawlers are covered automatically |
| 14 | * without a migration. |
| 15 | * |
| 16 | * Caveat: the `robots_txt` filter only feeds WordPress's *virtual* robots.txt. |
| 17 | * A site serving a physical robots.txt file bypasses it entirely — a real |
| 18 | * limitation surfaced in the AI tab (see {@see self::has_static_robots_txt()}). |
| 19 | * |
| 20 | * @package automattic/jetpack-seo-package |
| 21 | */ |
| 22 | |
| 23 | namespace Automattic\Jetpack\SEO; |
| 24 | |
| 25 | /** |
| 26 | * Emits robots.txt directives that block opted-out AI crawlers. |
| 27 | */ |
| 28 | class Ai_Crawlers { |
| 29 | |
| 30 | /** |
| 31 | * Option holding the sparse map of per-crawler overrides |
| 32 | * (`slug => bool`, true = blocked). Only entries that deviate from the bot's |
| 33 | * default policy are stored. Mirrored in the plugin's settings endpoint |
| 34 | * whitelist (`jp_group => 'seo-tools'`). |
| 35 | * |
| 36 | * @var string |
| 37 | */ |
| 38 | const OPTION = 'jetpack_seo_ai_crawler_overrides'; |
| 39 | |
| 40 | /** |
| 41 | * Wire the robots.txt filter. |
| 42 | * |
| 43 | * @return void |
| 44 | */ |
| 45 | public static function init() { |
| 46 | add_filter( 'robots_txt', array( __CLASS__, 'append_directives' ), 10, 2 ); |
| 47 | } |
| 48 | |
| 49 | /** |
| 50 | * The known AI crawler catalog: slug => [ label, user_agent, type ]. |
| 51 | * |
| 52 | * `slug` is the stable key persisted in the override map and sent by the AI |
| 53 | * tab; `user_agent` is the token written to the `User-agent:` robots line; |
| 54 | * `label` is the human name shown in the UI; `type` is `answer` (fetches to |
| 55 | * cite in live AI answers, allowed by default) or `training` (collects to train |
| 56 | * models, blocked by default), which drives the AI tab's two sections and the |
| 57 | * per-type default. |
| 58 | * |
| 59 | * @return array<string, array<string, string>> |
| 60 | */ |
| 61 | public static function get_catalog() { |
| 62 | $catalog = array( |
| 63 | // Answer-engine crawlers fetch pages so AI assistants can cite them in |
| 64 | // live answers. Allowed by default — blocking them costs AI visibility. |
| 65 | 'oai-searchbot' => array( |
| 66 | 'label' => __( 'ChatGPT Search (OpenAI)', 'jetpack-seo' ), |
| 67 | 'user_agent' => 'OAI-SearchBot', |
| 68 | 'type' => 'answer', |
| 69 | ), |
| 70 | 'claude-searchbot' => array( |
| 71 | 'label' => __( 'Claude Search (Anthropic)', 'jetpack-seo' ), |
| 72 | 'user_agent' => 'Claude-SearchBot', |
| 73 | 'type' => 'answer', |
| 74 | ), |
| 75 | 'perplexitybot' => array( |
| 76 | 'label' => __( 'Perplexity', 'jetpack-seo' ), |
| 77 | 'user_agent' => 'PerplexityBot', |
| 78 | 'type' => 'answer', |
| 79 | ), |
| 80 | 'amzn-searchbot' => array( |
| 81 | 'label' => __( 'Amazon (Alexa)', 'jetpack-seo' ), |
| 82 | 'user_agent' => 'Amzn-SearchBot', |
| 83 | 'type' => 'answer', |
| 84 | ), |
| 85 | // Training crawlers collect content to train AI models. Blocked by |
| 86 | // default — blocking protects content with no AI-visibility downside. |
| 87 | 'gptbot' => array( |
| 88 | 'label' => __( 'ChatGPT (OpenAI)', 'jetpack-seo' ), |
| 89 | 'user_agent' => 'GPTBot', |
| 90 | 'type' => 'training', |
| 91 | ), |
| 92 | 'claudebot' => array( |
| 93 | 'label' => __( 'Claude (Anthropic)', 'jetpack-seo' ), |
| 94 | 'user_agent' => 'ClaudeBot', |
| 95 | 'type' => 'training', |
| 96 | ), |
| 97 | 'google-extended' => array( |
| 98 | 'label' => __( 'Google AI (Gemini)', 'jetpack-seo' ), |
| 99 | 'user_agent' => 'Google-Extended', |
| 100 | 'type' => 'training', |
| 101 | ), |
| 102 | 'applebot-extended' => array( |
| 103 | 'label' => __( 'Apple Intelligence', 'jetpack-seo' ), |
| 104 | 'user_agent' => 'Applebot-Extended', |
| 105 | 'type' => 'training', |
| 106 | ), |
| 107 | 'meta-externalagent' => array( |
| 108 | 'label' => __( 'Meta AI', 'jetpack-seo' ), |
| 109 | 'user_agent' => 'meta-externalagent', |
| 110 | 'type' => 'training', |
| 111 | ), |
| 112 | 'bytespider' => array( |
| 113 | 'label' => __( 'ByteDance', 'jetpack-seo' ), |
| 114 | 'user_agent' => 'Bytespider', |
| 115 | 'type' => 'training', |
| 116 | ), |
| 117 | 'ccbot' => array( |
| 118 | 'label' => __( 'Common Crawl', 'jetpack-seo' ), |
| 119 | 'user_agent' => 'CCBot', |
| 120 | 'type' => 'training', |
| 121 | ), |
| 122 | 'amazonbot' => array( |
| 123 | // Amazon's own docs: Amazonbot "may be used to train Amazon AI |
| 124 | // models" (Amzn-SearchBot is the answer-engine bot above). |
| 125 | 'label' => __( 'Amazon', 'jetpack-seo' ), |
| 126 | 'user_agent' => 'Amazonbot', |
| 127 | 'type' => 'training', |
| 128 | ), |
| 129 | ); |
| 130 | |
| 131 | return $catalog; |
| 132 | } |
| 133 | |
| 134 | /** |
| 135 | * Whether a catalog bot is blocked when the owner hasn't set an override for |
| 136 | * it: training crawlers are blocked by default, answer engines allowed. |
| 137 | * |
| 138 | * @param string $slug Catalog slug. |
| 139 | * @return bool |
| 140 | */ |
| 141 | private static function is_blocked_by_default( $slug ) { |
| 142 | $catalog = self::get_catalog(); |
| 143 | return isset( $catalog[ $slug ] ) && 'training' === $catalog[ $slug ]['type']; |
| 144 | } |
| 145 | |
| 146 | /** |
| 147 | * The resolved, sparse override map: `slug => bool` (true = blocked). |
| 148 | * |
| 149 | * Reads the stored option, drops unknown slugs, casts values to bool, and |
| 150 | * drops any entry that equals the bot's default policy — so the returned map |
| 151 | * contains only real deviations. A bot absent from this map falls back to its |
| 152 | * default in {@see self::is_blocked()}. |
| 153 | * |
| 154 | * @return array<string, bool> |
| 155 | */ |
| 156 | public static function get_overrides() { |
| 157 | $stored = get_option( self::OPTION, array() ); |
| 158 | if ( ! is_array( $stored ) ) { |
| 159 | return array(); |
| 160 | } |
| 161 | |
| 162 | $catalog = self::get_catalog(); |
| 163 | $overrides = array(); |
| 164 | foreach ( $stored as $slug => $blocked ) { |
| 165 | if ( ! isset( $catalog[ $slug ] ) ) { |
| 166 | continue; |
| 167 | } |
| 168 | $blocked = (bool) $blocked; |
| 169 | // Only keep entries that actually deviate from the default policy |
| 170 | // (training blocked, answer allowed) — resolved from the catalog we |
| 171 | // already hold rather than re-fetching it per slug. |
| 172 | if ( $blocked === ( 'training' === $catalog[ $slug ]['type'] ) ) { |
| 173 | continue; |
| 174 | } |
| 175 | $overrides[ $slug ] = $blocked; |
| 176 | } |
| 177 | |
| 178 | return $overrides; |
| 179 | } |
| 180 | |
| 181 | /** |
| 182 | * Whether a given crawler is currently blocked, applying an override when one |
| 183 | * exists and otherwise the bot's default policy. Unknown slugs return false. |
| 184 | * |
| 185 | * @param string $slug Catalog slug. |
| 186 | * @return bool |
| 187 | */ |
| 188 | public static function is_blocked( $slug ) { |
| 189 | $overrides = self::get_overrides(); |
| 190 | return array_key_exists( $slug, $overrides ) |
| 191 | ? $overrides[ $slug ] |
| 192 | : self::is_blocked_by_default( $slug ); |
| 193 | } |
| 194 | |
| 195 | /** |
| 196 | * The catalog slugs currently blocked (override or default). |
| 197 | * |
| 198 | * @return string[] |
| 199 | */ |
| 200 | public static function get_blocked_slugs() { |
| 201 | // Resolve the override map and catalog once, then fold in the per-type |
| 202 | // default — rather than calling is_blocked() per slug (which would re-read |
| 203 | // the option and rebuild the catalog on every robots.txt render). |
| 204 | $overrides = self::get_overrides(); |
| 205 | $blocked = array(); |
| 206 | foreach ( self::get_catalog() as $slug => $info ) { |
| 207 | $is_blocked = array_key_exists( $slug, $overrides ) |
| 208 | ? $overrides[ $slug ] |
| 209 | : ( 'training' === $info['type'] ); |
| 210 | if ( $is_blocked ) { |
| 211 | $blocked[] = $slug; |
| 212 | } |
| 213 | } |
| 214 | return $blocked; |
| 215 | } |
| 216 | |
| 217 | /** |
| 218 | * Whether the site allows search-engine (and therefore AI-crawler) indexing. |
| 219 | * |
| 220 | * When `blog_public` is off, WordPress disallows everything in robots.txt, so |
| 221 | * per-crawler controls are moot — the AI tab surfaces this instead of showing |
| 222 | * toggles that can't take effect. |
| 223 | * |
| 224 | * @return bool |
| 225 | */ |
| 226 | public static function search_engines_allowed() { |
| 227 | return (int) get_option( 'blog_public', 1 ) === 1; |
| 228 | } |
| 229 | |
| 230 | /** |
| 231 | * Whether the site is served from a WordPress.com staging subdomain |
| 232 | * (`*.wpcomstaging.com`) where the platform blocks all crawling regardless of |
| 233 | * the site's own settings — so AI-crawler controls can't take effect. |
| 234 | * |
| 235 | * @return bool |
| 236 | */ |
| 237 | public static function is_crawl_restricted_subdomain() { |
| 238 | $host = (string) wp_parse_url( home_url(), PHP_URL_HOST ); |
| 239 | $suffix = '.wpcomstaging.com'; |
| 240 | return strlen( $host ) > strlen( $suffix ) |
| 241 | && substr( $host, -strlen( $suffix ) ) === $suffix; |
| 242 | } |
| 243 | |
| 244 | /** |
| 245 | * Whether a physical `robots.txt` exists at the web root. The web server |
| 246 | * serves that file directly, bypassing WordPress's virtual robots.txt — so our |
| 247 | * `robots_txt` filter (and therefore these settings) never run. Common on |
| 248 | * local, sandbox, and staging sites. |
| 249 | * |
| 250 | * @return bool |
| 251 | */ |
| 252 | public static function has_static_robots_txt() { |
| 253 | return file_exists( ABSPATH . 'robots.txt' ); |
| 254 | } |
| 255 | |
| 256 | /** |
| 257 | * Append `Disallow: /` blocks for each blocked AI crawler to robots.txt. |
| 258 | * |
| 259 | * @param string $output The robots.txt content assembled so far. |
| 260 | * @param bool $public Whether the site is public (`blog_public`). When |
| 261 | * false WordPress already disallows everything, but the |
| 262 | * explicit per-bot blocks are still valid and harmless. |
| 263 | * @return string The robots.txt content with AI-crawler directives appended. |
| 264 | */ |
| 265 | public static function append_directives( $output, $public ) { |
| 266 | unset( $public ); |
| 267 | |
| 268 | $catalog = self::get_catalog(); |
| 269 | $blocked = self::get_blocked_slugs(); |
| 270 | if ( empty( $blocked ) ) { |
| 271 | return $output; |
| 272 | } |
| 273 | |
| 274 | $lines = array( '', '# AI crawlers blocked via Jetpack SEO.' ); |
| 275 | foreach ( $blocked as $slug ) { |
| 276 | $lines[] = 'User-agent: ' . $catalog[ $slug ]['user_agent']; |
| 277 | $lines[] = 'Disallow: /'; |
| 278 | $lines[] = ''; |
| 279 | } |
| 280 | |
| 281 | return $output . implode( "\n", $lines ) . "\n"; |
| 282 | } |
| 283 | |
| 284 | /** |
| 285 | * The AI tab's crawler bootstrap payload: the catalog (camelCased for JS), the |
| 286 | * current sparse override map, and the environment flags that decide whether |
| 287 | * the controls render or are replaced by an explanation. |
| 288 | * |
| 289 | * @return array |
| 290 | */ |
| 291 | public static function get_bootstrap_data() { |
| 292 | $catalog = array(); |
| 293 | foreach ( self::get_catalog() as $slug => $info ) { |
| 294 | $catalog[] = array( |
| 295 | 'slug' => $slug, |
| 296 | 'label' => $info['label'], |
| 297 | 'userAgent' => $info['user_agent'], |
| 298 | 'type' => $info['type'], |
| 299 | ); |
| 300 | } |
| 301 | |
| 302 | return array( |
| 303 | 'catalog' => $catalog, |
| 304 | 'overrides' => (object) self::get_overrides(), |
| 305 | 'searchEnginesVisible' => self::search_engines_allowed(), |
| 306 | 'restrictedSubdomain' => self::is_crawl_restricted_subdomain(), |
| 307 | 'staticRobotsTxt' => self::has_static_robots_txt(), |
| 308 | ); |
| 309 | } |
| 310 | } |