A "we're invisible in AI search" request can arrive as a feeling rather than a question. This guide turns that feeling into a diagnostic you can actually run: a bounded set of known pages, a stated audience, a claim you can test, and a handoff someone can act on. It is written as an investigation frame, not a delivery plan.
Work through the questions here on paper or in the private diagnostic worksheet, which keeps your answers on your own device. Nothing on this site verifies your results, certifies anything, or reports a score.
Why the frame matters before the fix
Google states that its SEO best practices remain relevant for AI features such as AI Overviews and AI Mode, and that there are no additional requirements or special optimizations to appear in them (Google Search Central, AI features). That single sentence removes most of the imagined work. It also raises the bar on the diagnostic: if there is no separate AI checklist, the useful question is not "what AI trick is missing" but "which ordinary Search condition is failing, and for which pages and audience."
It also means a claim like "our pages are hidden from AI answers" should be treated as a hypothesis. Google notes that AI Overviews and AI Mode may use a query fan-out technique and may use different models, so the set of responses and links they show will vary (same source). Variance is a property of the system, not necessarily a defect in the site — which is exactly why the frame has to fix its terms in advance.
The seven fields of the frame
The worksheet asks for seven items. Each exists to stop a different failure.
1. Known pages
Name specific URLs, not "the blog." For eligibility as a supporting link in AI Overviews or AI Mode, Google requires that a page be indexed and eligible to be shown in Search with a snippet, meeting the Search technical requirements, with no additional technical requirements (Google Search Central). A URL list is what you check those conditions against. Beware unsupported inference here: indexing and serving is not guaranteed even when requirements are met, per the same page.
2. Target audience
Who is asking the question, and in what words? Google's people-first guidance asks whether you have an existing or intended audience that would find the content useful if they came directly to you (Creating helpful, reliable, people-first content). Write the audience as a reader with a task, not a persona with a demographic profile. You are not allowed to invent an audience and call it research; state it as an assumption to test.
3. Claimed issue
Reduce the complaint to one falsifiable sentence: "Our pricing page does not appear as a supporting link for comparison-style queries." Vague claims ("low AI visibility") cannot produce an interpretable result. If the claim cannot be written in one sentence, the diagnostic is not ready.
4. Existing evidence
List what you already have: internal link paths, whether important text is in textual form, robots.txt lines, snippet controls. Google's SEO fundamentals list in the AI features guide includes allowing crawling in robots.txt and by any CDN or hosting infrastructure, making content findable through internal links, and ensuring important content is available in textual form (Google Search Central). Mark each item observed, unobserved, or assumed. Do not upgrade an assumption because it seems obvious.
5. Comparison
What are you comparing against — a competitor page, a past observation, a different query phrasing? Comparison is where diagnostics drift into assertion, so record the basis. Google states that AI Mode and AI Overviews may use different models and techniques and that the links they show will vary (same). A single differing observation is not a trend.
6. Next test
One small test with a readable outcome. Examples of testable primitives: whether a URL is indexed and snippet-eligible; whether a robots.txt directive blocks the crawler you care about; whether page text is available as text rather than only in an image. Keep the test about a documented condition, not about "getting picked." The follow-up guide, write a small test, covers designing that step.
7. Handoff
Who acts on the result, and what exactly do they need? A handoff that says "improve content" is not a handoff. See make an advisory handoff useful.
Access first, content second
One framing error is collapsing two different questions into one. Google describes robots.txt as a file that tells crawlers which URLs they can access, used mainly to avoid overloading a site, and states it is not a mechanism for keeping a page out of Google (Introduction to robots.txt). A page disallowed in robots.txt can still be indexed if linked elsewhere, with its URL and possibly anchor text appearing in results (same). So a robots.txt line is an access fact, not an indexing verdict.
For ChatGPT search, OpenAI documents separate robots.txt tags: OAI-SearchBot for surfacing sites in search results in ChatGPT's search features, and GPTBot for content that may be used in training, each setting independent of the others (Overview of OpenAI Crawlers). OpenAI states sites opted out of OAI-SearchBot will not be shown in ChatGPT search answers though they can still appear as navigational links, and that it can take about 24 hours from a robots.txt update for its systems to adjust (same). It also notes ChatGPT-User handles user-initiated actions, is not used for automatic crawling, and that robots.txt rules may not apply to it (same). These are vendor statements about their own user agents only; other crawlers may differ.
If the claimed issue is an access problem, content work will not move it. If the claim is a content problem, robots.txt work will not fix it either. The access-versus-content guide splits these explicitly.
Keep the diagnostic honest
Google's helpful-content guidance asks who created content, how it was created, and why (Creating helpful content). It also states that fabricating creator profiles — AI-generated headshots, made-up names, or false credentials — is a form of deception and a signal of a low-quality page (same). That is why this publication carries no invented expert biography. Your own diagnostic notes should follow the same rule: record the source of every claim, and where you are inferring rather than quoting, say so.
Google also describes its ranking systems as designed to prioritize helpful, reliable information created to benefit people rather than content created to manipulate rankings, and notes that trust is the most important aspect of E-E-A-T, while stating E-E-A-T is not itself a specific ranking factor (same). Read that as framing guidance, not a scoring rubric: there is no number to maximize.
Open your frame
Take the seven fields to the diagnostic brief worksheet. It is private to your browser, holds no account, and produces no report — it is a place to think in a fixed order so the next test is worth running.