Make your website readable by AI: access, content and evidence

Check access, useful text, crawl settings and factual consistency. A practical website-readability checklist that distinguishes indexing, retrieval, AI citations and actual enquiries.

Short answer

Make the information buyers need available on accessible pages with clear text, accurate facts and useful links. Inspect crawl settings for the specific service you want to reach. Then keep page access, indexing, AI citations and visits as separate evidence. Adding an AI-labelled file or schema does not demonstrate any of those outcomes.

Do you need an AI-specific file?

A file such as llms.txt can act as a curated description of selected site content. Before investing in one, identify a documented consumer and the problem it solves. Another copy of your service description also creates another place where information can become outdated.

Google's guidance for AI features in Search says no new machine-readable file, AI text file or special schema is needed for those features. A supporting page must meet the relevant Search conditions, including indexing and snippet eligibility. This concerns Google's features; it is not a statement that every assistant processes websites identically.

Keep optional experiments separate from required repairs. If buyers cannot read your service page or its contact path is broken, an additional text file does not resolve that defect.

Separate search access from model training

Decide access by documented user agent and purpose, not by a blanket label of “AI bot”. OpenAI's official crawler documentation distinguishes OAI-SearchBot for search from GPTBot for content that may be used in foundation-model training. Their robots.txt settings are independent.

The same documentation identifies ChatGPT-User with certain user-initiated actions, rather than automatic crawling or the Search inclusion control. Do not infer search eligibility from one ChatGPT-User request. Do not treat robots.txt as access control for private documents: those need the appropriate authentication and authorisation.

Perplexity's documentation of its search crawlers also distinguishes search from foundation-model training. Read the current documentation for each service before changing a rule. Check hosting or security-layer restrictions alongside the site's own settings.

Keep visible facts and structured data aligned

Start with the facts a buyer needs: what the business provides, who the offer suits, where it operates, the limitations and the next step. Present them in text, not only inside images or an interactive element that has not been tested.

Use structured data when it accurately describes the visible page. Check organisation details, dates and links against the content a visitor can read. A technically valid object with an invented review, stronger service promise or outdated price is not a reliable description.

Choose markup for the page's actual type and supported use. Do not add unrelated types merely to increase the number of schema objects. The purpose of a content check is accuracy and accessibility, not a claim that markup forces an assistant to cite you.

Separate five kinds of evidence

  • Access: the intended page responds and its important text can be read.
  • Indexing: a search service reports the page in its index under the inspected conditions.
  • Retrieval: a request appears in logs. Verify its origin before labelling it a particular crawler.
  • Citation: a recorded answer references your page for a specific question and test context.
  • Visit or enquiry: your measurement or receiving system records the corresponding event, with its attribution limits.

A successful fetch is not a citation. A citation is not a click, and a click is not a qualified enquiry. Preserve those distinctions when deciding whether technical work achieved its purpose.

Run a checklist on pages buyers actually need

Choose a representative service page, decision guide and contact path. Record the date and exact URLs, then inspect:

  1. The response status and intended canonical URL.
  2. The visible title, main heading, offer description and supporting evidence.
  3. Links that connect the page to useful explanations in the same language.
  4. The applicable indexing, snippet and crawler controls.
  5. Whether structured data agrees with the visible content.
  6. Whether the next action reaches the expected form, resource or booking page.

Test the delivered page after a change, not just the source file. An inaccessible production route is not repaired because a local template compiles. Keep a before-and-after record of the specific defect and check.

Avoid fixes that make the report look better than the site

Do not publish keyword variants that repeat the same answer. Do not change modification dates without meaningful revisions. Do not copy confidential documents into public summaries to make them easier for a crawler to read.

When information differs between the page, metadata and a business listing, identify the source of truth and the person allowed to correct it. Record third-party pages you cannot change rather than treating them as completed repairs.

Prioritise the pages closest to a useful decision

Use business relevance and observed defects to choose a small correction batch. A clear existing service page can be more useful than another general definition of AI. Define who reviews the facts and what evidence will establish that each correction is delivered.

Then use the AI visibility audit method to observe questions, engines, dates, languages, test countries and cited sources under documented conditions. Keep those observations separate from the technical acceptance checklist.

Scope the work before buying a package

A proposal should identify the pages, technical access, factual review, correction scope and post-publication checks. Separate implementation from monitoring. Do not compare proposals only by the number of pages or schema types they promise to add.

The SEO and AI search visibility service sets out this approach. A 30-minute project conversation establishes priorities and fit; a detailed audit or implementation is defined separately in a proposal.

Frequently asked questions

Is llms.txt required for AI search visibility?

Google says no new AI text file or special schema is required for its AI search features. That is a Google-specific statement, not a universal rule for every product. Treat any optional machine-readable summary as an additional document to maintain, not a substitute for accessible pages or proof of citation.

Are OAI-SearchBot and GPTBot the same thing?

No. OpenAI documents OAI-SearchBot for search and GPTBot for content that may be used in foundation-model training. Their robots.txt controls are independent. ChatGPT-User serves certain user-initiated actions and is not the control used to determine Search inclusion.

Does valid structured data guarantee an AI citation?

No. Use appropriate markup that agrees with the visible page. Google states that there is no special schema needed for its AI search features, and meeting its requirements does not guarantee crawling, indexing or serving.

How do we prove that the site is readable?

Check the actual response, visible text, navigation and access controls on the intended pages. Then separately inspect indexing where available, documented crawler requests, observed citations and measured visits. Passing one check does not prove the next outcome.

Which work should come first?

Fix evidence-based blockers on pages buyers need: inaccessible content, missing useful text, broken links, inconsistent facts or an unintended crawl restriction. Choose the next change from a dated inspection rather than a generic requirement to add an AI-specific file.

LYVIA

LYVIA Team

AI automation and SEO/GEO visibility

LYVIA builds custom tools and works on SEO and AI search visibility for established small and mid-sized businesses.