Google, Microsoft, and OpenAI have backed work on a common way for websites to reject certain artificial intelligence uses. Their support gives the effort unusual weight across search, cloud computing, and generative AI markets. The goal sounds simple, but it demands careful technical and legal choices.<\/p>\n\n\n\n
The initiative focuses on machine-readable preferences that automated systems can discover and interpret consistently. Publishers could distinguish AI training from search indexing or other automated processing. A shared approach could replace the growing collection of company-specific controls.<\/p>\n\n\n\n
The Internet Engineering Task Force provides an important forum for this work. Its AI Preferences working group develops technical methods for expressing preferences about AI-related content use. Participants commonly refer to the effort as AIPREF.<\/p>\n\n\n\n
The IETF creates voluntary internet standards through open discussion, technical drafts, testing, and consensus. It does not pass laws or directly punish companies. Successful IETF standards still influence how browsers, servers, networks, and online services communicate.<\/p>\n\n\n\n
Google, Microsoft, and OpenAI bring extensive experience with crawlers, models, search products, and cloud infrastructure. Their involvement could improve compatibility across large platforms. However, participation does not automatically guarantee support for every final requirement.<\/p>\n\n\n\n
Website operators have long used robots.txt files to manage automated crawler access. The Robots Exclusion Protocol became an official IETF standard through RFC 9309. It tells compliant crawlers which website paths they may access.<\/p>\n\n\n\n
However, robots.txt originally addressed crawling behavior rather than detailed content rights. Blocking a crawler can also prevent useful services, including search discovery. It rarely explains whether an operator permits training, summarization, retrieval, or commercial licensing.<\/p>\n\n\n\n
AI developers have consequently introduced separate crawler names and instructions. Publishers must research each company, update rules, and monitor changing documentation. That fragmented process creates unnecessary work and increases configuration errors.<\/p>\n\n\n\n
Google offers the Google-Extended product token for managing certain uses involving Gemini systems. Google says the token does not affect a website’s inclusion or ranking within Google Search. This separation allows publishers to retain search visibility while limiting specified generative AI uses.<\/p>\n\n\n\n
OpenAI also identifies different crawlers for different purposes. Publishers can block GPTBot if they want to restrict potential model-training access. OpenAI separately identifies OAI-SearchBot for surfacing websites within its search products.<\/p>\n\n\n\n
Microsoft operates search, cloud, assistant, and model services with their own technical requirements. A common vocabulary could reduce the need for provider-specific interpretations. The proposed standard therefore addresses coordination, not merely another crawler label.<\/p>\n\n\n\n
The phrase \u201cAI use\u201d covers several activities that affect publishers differently. Developers train foundation models on large collections of text, images, audio, and code. Other systems retrieve current pages when answering individual questions.<\/p>\n\n\n\n
Search engines may index content before generating summaries or links. Assistants may receive a page directly from a user. Archiving, evaluation, safety testing, and model fine-tuning create further categories.<\/p>\n\n\n\n
A useful standard must express these differences without becoming impossible to implement. Broad labels offer simplicity but may hide meaningful distinctions. Highly detailed labels provide control but can confuse website owners and software developers.<\/p>\n\n\n\n
Accordingly, AIPREF discussions concern both vocabulary and delivery mechanisms. A preference might appear through an HTTP header, website file, or page-level metadata. Implementers must also decide how preferences apply to embedded material and linked resources.<\/p>\n\n\n\n
A technical signal communicates a preference, but it cannot physically stop a determined scraper. Robots.txt has always depended upon voluntary compliance. AI preference signals will probably face the same limitation unless laws or contracts require observance.<\/p>\n\n\n\n
The system also cannot automatically remove information collected before an operator published an opt-out. Developers may retain older datasets, backups, or derived model parameters. Retroactive deletion creates technical questions that a simple web signal cannot solve.<\/p>\n\n\n\n
Still, a standard can create clear evidence about a publisher’s stated choices. Responsible developers can automate compliance across millions of websites. Auditors could also compare declared preferences with documented collection practices.<\/p>\n\n\n\n
Copyright laws differ across countries, and courts continue examining AI training disputes. Some developers argue that training can qualify for legal exceptions. Authors, artists, publishers, and software companies often challenge that position.<\/p>\n\n\n\n
The European Union recognizes rights reservations for certain text and data mining activities. Machine-readable signals can help rights holders communicate those reservations. However, technical standards cannot determine whether a particular use violates copyright law.<\/p>\n\n\n\n
United States courts evaluate issues including fair use, market harm, copying, and access. Outcomes can depend on specific datasets and product behavior. An opt-out signal may support legal arguments, but it does not settle them.<\/p>\n\n\n\n
Licensing remains another option for publishers and AI companies. A standardized preference could direct developers toward licensing information or approved uses. That approach might support negotiations without blocking every automated service.<\/p>\n\n\n\n
Large media companies can negotiate contracts and build custom enforcement systems. Independent creators, local newsrooms, and small businesses usually lack those resources. A simple standard could give them comparable controls at a lower cost.<\/p>\n\n\n\n
Content management platforms could add preference settings to ordinary publishing screens. Hosting providers might apply rules across entire domains. Such tools would let users choose options without manually editing server files.<\/p>\n\n\n\n
Clear controls could also prevent accidental overblocking. A publisher might welcome traditional search indexing while rejecting model training. Another operator might permit research uses but require permission for commercial development.<\/p>\n\n\n\n
A common standard would reduce compliance work for model developers. Instead of interpreting many custom notices, collection systems could process one documented vocabulary. Smaller AI companies would benefit because they cannot maintain extensive policy teams.<\/p>\n\n\n\n
Consistent signals could improve dataset records and internal governance. Developers could store each resource’s preference alongside its source and collection date. They could then review permissions before training, evaluation, or retrieval.<\/p>\n\n\n\n
Nevertheless, companies must handle changes carefully. A website owner may revise preferences after a crawler visits. Systems need rules for caching, versioning, verification, and conflicting instructions.<\/p>\n\n\n\n
The standard must identify who can set a valid preference. Domain owners, individual authors, platform operators, and customers may claim different rights. Shared hosting and social networks make that question particularly difficult.<\/p>\n\n\n\n
Bad actors could also publish false instructions for material they do not control. Authentication methods may reduce abuse, although complex verification could burden smaller websites. Designers must balance reliability with straightforward deployment.<\/p>\n\n\n\n
International adoption presents another challenge. The standard needs neutral language that works across legal systems and business models. It must also remain useful as AI products and training methods change.<\/p>\n\n\n\n
Support from Google, Microsoft, and OpenAI makes broad implementation more plausible. Yet an IETF effort requires continued drafting, public review, testing, and consensus. Competing priorities could still influence the final design.<\/p>\n\n\n\n
Publishers should therefore distinguish a developing specification from an established universal rule. They must continue using documented crawler controls and applicable legal tools. Developers should likewise follow current instructions while monitoring the shared effort.<\/p>\n\n\n\n
A successful standard would not end disputes about training data or copyright. It could still make preferences clearer, cheaper, and easier to process. That practical improvement explains why cooperation among major rivals matters.<\/p>\n","protected":false},"excerpt":{"rendered":"
Google, Microsoft, and OpenAI have backed work on a common way for websites to reject certain artificial intelligence uses. Their support gives the effort unusual weight across search, cloud computing, and generative AI markets. The goal sounds simple, but it demands careful technical and legal choices. The initiative focuses on machine-readable preferences that automated systems […]<\/p>\n","protected":false},"author":2,"featured_media":10962,"comment_status":"closed","ping_status":"closed","sticky":false,"template":"","format":"standard","meta":{"apple_news_api_created_at":"2026-11-09T02:58:15Z","apple_news_api_id":"11afd379-28b8-4517-a97c-99a3dafa82f7","apple_news_api_modified_at":"2026-11-09T02:58:15Z","apple_news_api_revision":"AAAAAAAAAAD\/\/\/\/\/\/\/\/\/\/w==","apple_news_api_share_url":"https:\/\/apple.news\/AEa_TeSi4RRepfJmj2vqC9w","apple_news_cover_media_provider":"image","apple_news_coverimage":0,"apple_news_coverimage_caption":"","apple_news_cover_video_id":0,"apple_news_cover_video_url":"","apple_news_cover_embedwebvideo_url":"","apple_news_is_hidden":"","apple_news_is_paid":"","apple_news_is_preview":"","apple_news_is_sponsored":"","apple_news_maturity_rating":"","apple_news_metadata":"\"\"","apple_news_pullquote":"","apple_news_pullquote_position":"","apple_news_slug":"","apple_news_sections":[],"apple_news_suppress_video_url":false,"apple_news_use_image_component":false,"_jetpack_newsletter_access":"","_jetpack_dont_email_post_to_subs":false,"_jetpack_newsletter_tier_id":0,"_jetpack_memberships_contains_paywalled_content":false,"_jetpack_hide_featured_image":false,"_jetpack_featured_image_duplicate_dismissed":0,"_jetpack_feature_clip_id":0,"_jetpack_memberships_contains_paid_content":false,"footnotes":"","jetpack_post_was_ever_published":false,"_ppma_block_editor_authors":""},"categories":[1],"tags":[],"ppma_author":[356],"class_list":["post-10961","post","type-post","status-publish","format-standard","has-post-thumbnail","hentry","category-news"],"apple_news_notices":[],"yoast_head":"\n