What the evidence says about six AI search claims

Use this table to review a proposal line by line. The findings reflect the documents available on 13 November 2026. A statement about a named feature applies to that feature.

ClaimPrimary sourceWhat it establishesWhat it does not establish
Google needs llms.txt for AI Overviews or AI Mode.Google Search Central: generative AI optimization guide, updated 10 July 2026.Google says Search ignores llms.txt as a special file for visibility and rankings, including its generative features.How an assistant outside Google Search handles the file. A file can be fetched without being treated as a special signal.
Special Schema.org markup gets a page into Google’s AI answers.Google Search Central: AI features and your website; structured data guidelines.Google requires no special markup. Accurate markup for supported types can make a page eligible for rich results in Search.That a validator pass will produce a rich result or that markup will earn a link in an AI answer.
ChatGPT Search needs llms.txt.OpenAI: Overview of crawlers.The guide describes OAI-SearchBot for search, a separate GPTBot setting for training and published IP ranges. It does not list llms.txt as a search requirement.What every OpenAI product may fetch or whether a particular accessible page will be cited.
A Perplexity documentation index proves Perplexity Search uses my file.Perplexity: Perplexity Crawlers.The guide links to Perplexity’s own documentation index and recommends allowing PerplexityBot for search. It distinguishes user-triggered fetches.That publishing your own file changes how Perplexity Search picks sources or citations.
A Lighthouse finding means Google Search uses llms.txt.Chrome for Developers: Lighthouse llms.txt audit, updated 5 May 2026.The audit checks the file in Lighthouse’s Agentic Browsing category. The llms.txt v2 proposal describes an overview and links for compatible tools.That passing the audit changes visibility in Google Search.
A log entry for llms.txt proves the file earned a citation.Ahrefs: study of llms.txt requests on sites in its analytics sample, May 2026 observation window.The study observes requests to a path. A verified request can show which client fetched a response and when.Whether the client parsed the file, used it to choose a source, or caused a change in citations.

Google: Optimizing for generative AI features · Google: AI features and your website · Google: General structured data guidelines · OpenAI: Overview of crawlers · Perplexity: Perplexity Crawlers · llms.txt: Format proposal, v2 · Chrome for Developers: Lighthouse llms.txt audit · Ahrefs: Study of llms.txt requests in May 2026

Which file or markup solves which problem?

ToolJobWhen to work on it
robots.txtPublish crawl rules for crawlers that follow them.The rule for the intended search crawler and public path conflicts with the access you want. Use authentication for private content.
XML sitemapList preferred URLs for discovery by supporting search engines.The sitemap is missing important published URLs or lists redirects and outdated addresses. Discovery does not guarantee indexing.
Schema.org structured dataDescribe facts about the page and its entities in a machine-readable vocabulary.A supported type fits the page and the visible facts can be kept in sync. JSON-LD is one way to encode the data.
llms.txtOffer an optional Markdown reading guide with descriptions and links, as proposed in llms.txt v2.A named tool or documentation workflow uses it, and someone can keep it aligned with the public pages.

Use the AI-ready website checklist to place these tasks alongside navigation, content and language checks.

Google: Robots.txt introduction and guide · Google: Learn about sitemaps · Google: General structured data guidelines · llms.txt: Format proposal, v2

How to accept or reject a proposed implementation

Ask for four things in the ticket: the target platform and feature, the dated documentation or observed failure, the exact change, and the evidence you will accept when it is done. Split a bundled quote into separate tasks. “Improve AI visibility” describes a goal; it does not tell you whether the work is complete.

Example for a service site: a public service page is inaccessible to the intended search crawler, while the agency proposal offers an llms.txt file. Prioritize the access fix. Accept it when the intended URL returns the expected public text and the crawler can be verified through the operator’s tool or appropriate logs. If the same proposal offers structured data, ask which supported feature it targets and which visible facts it represents.

For a documentation team, an optional llms.txt task can have a different acceptance check: a named tool opens the guide and reaches the current reference for the same question, in the right language. Test with and without the guide, keep the question and tool settings fixed, and record what changed. That tests the workflow you named, not public search visibility.

For a Google supporting link, check that the page is indexed and eligible for a snippet. Check the property’s Search generative AI participation setting in Search Console as well. Meeting these conditions does not guarantee a link.

To reproduce a failed fetch, follow the one-URL crawler access check before changing discovery files.

Google: AI features and your website · Google Search Console: Search generative AI control

Questions to settle before you ship

Can I keep the llms.txt I already have?

Yes, if it serves a known workflow and the public links are maintained. Check for broken links, outdated claims and the wrong language versions. If no one relies on it, deferring work is reasonable. Google’s guidance gives no Search visibility reason to keep or remove it.

Should I remove existing structured data if Google does not require it for AI search?

Keep markup that is accurate and has an identifiable purpose, such as describing an article for a supported Search feature. Fix markup that contradicts visible content. A rich result validator tests eligibility and formatting for its supported features; it cannot tell you whether an AI answer will cite the page.

A citation appeared after we added the file. Is that evidence of impact?

Record it as a dated observation. Compare repeated runs with the same questions, platform, search mode and language; track page edits and other changes in parallel. A single before-and-after observation cannot isolate the effect of the file, even if a verified log shows that it was fetched.

Google: Optimizing for generative AI features · Google: General structured data guidelines

Start with one change you can verify

Choose one public URL and the platform you care about. Fix the first documented problem. If the page is already accessible and there is no named consumer for an optional file, spend the time making the page more useful to the reader.

For help reproducing a technical issue and defining the retest, bring the URL and evidence to our technical website maintenance service.

Sources & further reading

BeeCited provides website and GEO services. We checked the sources on 13 November 2026. This is a review of public documentation and published observations. AI assisted the editorial, factual and language checks.

  1. Google: Optimizing for generative AI features —
  2. Google: AI features and your website —
  3. Google: General structured data guidelines —
  4. OpenAI: Overview of crawlers —
  5. Perplexity: Perplexity Crawlers —
  6. llms.txt: Format proposal, v2 —
  7. Chrome for Developers: Lighthouse llms.txt audit —
  8. Ahrefs: Study of llms.txt requests in May 2026 —
  9. Google: Robots.txt introduction and guide —
  10. Google: Learn about sitemaps —
  11. Google Search Console: Search generative AI control —

You may also need