How to appear in Google AI Overviews

The mistake that hides the useful page

The common mistake is building a page for an AI answer before making it competitive in search. You add a glossary, an FAQ block and perhaps an llms.txt file, while the page sits below the results Google uses to form the answer. After the AI-focused edits, the page often contains broad claims, generic questions and markup, but still does not explain who it serves or why it belongs in the result. That costs you twice: the page is harder to find, and the work aimed at AI Overviews has no clear route into the result.

Start with the less novel answer: Google AI Overviews cite pages already ranking in the top results. CiteGraph's citation scans, described in our methodology, show this pattern in the pages selected for AI answers. The work is less about a special submission and more about getting a useful page into the set Google already trusts for the query.

A few terms will recur. Search eligibility means Google can crawl, render and index the material. Retrieval means the page is a plausible result for the query. A citation is the final selection of that page in the Overview. Better schema may support the first two stages, but it cannot replace them or guarantee a citation.

Google AI Overviews inherit the search result

The practical consequence is that you should study the query's existing results before changing the page. Look at the pages Google already returns for the buyer's wording. If the result set is made up of comparisons, roundups or product pages, a general homepage is usually the wrong format. If it contains detailed guides, your page needs to answer the same job with clearer evidence and a more direct structure.

Our scan data shows the same pattern. Listicles and roundups account for 67.5% of all recorded citations, from 30,585 citations, while competitor product pages account for 24.4%, from 11,052 citations. Together they take 91.9% of citations. The product's own site accounts for 1.5%, from 663 citations, so publishing a page on your own domain is not enough by itself. The page has to fit the type of source the answer already uses.

That does not mean you should copy a roundup or turn every page into a comparison. It means page type is part of search intent. A product page can explain what you sell, while a comparison page answers a buyer choosing between options. Google AI Overviews can cite either, when the page matches the task behind the query.

Start with the checks that affect search access

The first group of checks is familiar SEO work, and it deserves priority here. In CiteGraph's fifteen readiness checks across 121 audited sites, 116 sites passed the check for AI crawlers allowed in robots.txt, while five failed. Google says its common crawlers are used to find information for search indexes and other products, and that they always obey robots.txt rules when crawling automatically. The Google crawler documentation also separates Google Search crawling from Google Extended, which does not affect inclusion in Google Search or act as a ranking signal.

Rendering is another direct concern. Content was present without JavaScript on 92 of the 121 sites, failed on 18 and warned on nine. A sitemap was published on 100 sites. Canonical and Open Graph tags passed on 92 sites, with 27 warnings, and title and description clarity passed on 92, with 27 warnings. These checks describe access and clarity, not citation likelihood, but failures can make the page harder to understand, select or fetch.

The same applies to page structure. A clear H1 passed on 101 sites and failed on 18. The homepage answered a plain request on 119 sites and failed on two. Those are not AI tricks. They are signs that the page gives search systems a stable subject and a usable answer. Fix them before adding another machine-readable layer.

Here is the short checklist for this stage:

• Allow the relevant Google crawler in robots.txt.
• Keep the important answer in HTML that is available without JavaScript.
• Give the page one clear H1 that states its subject.
• Use a title and description that say what the page is about.
• Set canonical and Open Graph tags consistently.
• Publish a sitemap and make the target page reachable through the site.
• Check that the page answers the plain version of the search request.

Make the target page answerable

Once access is sound, write for the question that could trigger the Overview. The page should state what the thing is, who it is for, what it costs and what it is being compared against when those facts apply. This is the plain statement check in the readiness audit. It passed on 20 of 121 sites, failed on 43 and warned on 56, making it the weakest result in the sample alongside comparison coverage and on-page questions.

A before and after makes the change concrete. Before, the opening says, “We help teams work better together.” The price is behind a sales conversation, and the features sit under headings such as “Powerful collaboration” and “Built for modern work.” After, it says, “Project tracker for five-person agencies, $29 per seat per month, compared with Trello and Asana.” The page then uses headings for tasks, reporting, integrations and limits, so the opening claim has visible support.

This is where “AI Overviews optimization” as a mindset becomes unhelpful. It encourages edits aimed at an imagined AI preference rather than the query. Use the search result as the brief. Put the answer near the top, cover the follow-up questions a buyer would need, and link to the detailed evidence instead of making the reader hunt for it.

Questions answered on the page passed on 56 sites, failed on none and warned on 63. A warning means the audit found an issue worth inspecting, not that the page is barred from citation. Add useful questions only when they belong to the page. A long FAQ with thin answers is extra furniture, not extra authority.

Use schema and metadata as support

Structured data helps describe the page, entity or product. Organization schema passed on 72 of 121 sites and failed on 47. Product or SoftwareApplication schema passed on 60 and failed on 59. Those gaps are worth fixing when the markup accurately reflects visible page content, after the work on search intent, page quality and access.

The same rule applies to price. Price was visible or a pricing page was linked on 90 sites, with 29 warnings. If a buyer needs price to choose between products, make that information easy to find and link to the relevant page. If the query does not involve buying or comparing cost, adding a price block to satisfy an audit is unlikely to improve the answer.

Canonical tags, Open Graph tags, titles and descriptions are more useful as hygiene than as a promise of inclusion. They reduce ambiguity about the page and its presentation. Treat metadata as a way to keep the page legible while the core work earns its place in search.

Do not mistake a readiness pass for a citation

Several checks are useful diagnostics but weak targets for this particular job. llms.txt passed on 66 of 121 sites and warned on 55, yet the audit does not show that adding it makes a page rank in the results used by AI Overviews. The site's own pages getting cited passed on only five sites, failed on 11 and warned on five. That is evidence about observed citation coverage, not a setting that produces citations.

Comparison and alternatives pages passed on 54 sites, failed on 21 and warned on 46. This check matters when people compare products, because the scan data shows that competitor product pages and listicles dominate recorded citations. It does not mean every business needs an alternatives page. Build one when the query and the result set call for it, then make it fair, specific and useful enough to rank.

The audit's worst results are a plain statement of what it is, for whom, at what price, against what, followed by comparison or alternatives pages and questions answered on the page. Those are content decisions. By contrast, crawler access, rendering, titles, H1s, canonicals and sitemaps are the technical base that lets those decisions be seen. Both groups matter, but they solve different problems.

This is also why adding every available schema type is the wrong response. Organization markup, product markup, FAQ markup and llms.txt can describe a site, while the page still needs to rank for the underlying query.

Our data has a limit. We measure samples, not every search result or every Overview, so we can print what we found in the 121 audited sites and recorded citation scans, but we do not know what we have not measured.

What to do this week

• Check crawler access for the target URL in robots.txt.
• Rewrite the page opening so it states what the product is, who it is for and what it costs when price matters.
• Run the changed URL through the free AI visibility checker and record whether it enters the relevant results and earns a citation.