Google AI Overviews and AI Mode are the generative answer features in Google Search: a summary at the top of the results page (AI Overviews) and a conversational search tab (AI Mode), both with clickable links to the pages that support the answer. Google’s own guide to these features, published in May 2026 and updated in July 2026, gives founders a short answer: they run on Google’s core ranking systems, so good SEO is the way in.
How do AI Overviews and AI Mode choose which pages to cite?
Google describes two steps. First comes grounding (retrieval-augmented generation): Google’s ranking systems pull relevant, current pages from the Search index. Then the model reads those pages and writes an answer with links to the sources.
AI Mode adds query fan-out. The model turns one question into several related searches and fetches results for each. Google’s example: “how to fix a lawn that’s full of weeds” can fan out to searches about the best herbicides, removing weeds without chemicals, and preventing weeds. Your page can be cited for a sub-question the searcher never typed.
The takeaway is plain: the pool of pages comes from the normal index. If a page does not rank-qualify there, it never reaches the model.
What are the eligibility rules?
Google lists two requirements, plus a warning that meeting them guarantees nothing:
- The page is indexed and eligible to show in Search with a snippet, so it meets the Search technical requirements.
- Your site is included in Search generative AI features in Search Console.
Take the snippet rule seriously. A nosnippet robots directive or a tight max-snippet value on a page removes or limits what Google can show from it, and that affects the generative features as well. Check your templates before anything else.
What does Google say you can ignore?
This list saves the most time. Google’s guide says none of these help, and some are policy risks:
| Tactic | Google’s position |
|---|---|
llms.txt, AI text files, Markdown copies of pages |
Google Search does not use them. They neither help nor hurt. See llms.txt. |
| “Chunking” content into tiny pieces | Not needed. There is no ideal page length. |
| Writing special versions for AI, chasing long-tail wording | Not needed. Google’s systems understand synonyms and meaning. |
| Manufactured “mentions” on other sites | Ranking and spam systems make them ineffective. |
| Special schema for AI | None exists. Structured data stays useful for rich results. |
| Mass pages built for every fan-out query | Violates the scaled content abuse policy. |
Google also calls “AEO” and “GEO” SEO under another name and warns against third-party tools that claim access to Google’s internal metrics.
How do you improve your chances, step by step?
- Confirm eligibility. Open the Page indexing report and URL Inspection in Google Search Console. Every page you want cited must be indexed. Search your templates for
nosnippetandmax-snippet:0. - Keep the page crawlable and readable. Content must load without a login, and the main text must exist in the rendered HTML (see JavaScript SEO). Use clear headings and short paragraphs. Google says semantic HTML helps where practical but perfect code is not required.
- Publish content nobody else can. Google asks for non-commodity content: first-hand experience, your own data, your own photos, a clear point of view. A summary that any model could write is the page least likely to be cited. Build this into E-E-A-T signals such as named authors and real examples.
- Answer the question early. Put a direct answer near the top of the page, then the detail. The method page on answer-first content shows the format.
- Cover the follow-up questions. Fan-out means the next three questions a reader asks matter. Add them as sections on the same page instead of 30 thin pages.
- Add good images and video where they explain something, and follow Google’s image and video guidance.
- Feed Google your business data. Shops keep Merchant Center feeds accurate and local businesses keep their Business Profile complete, because Google points to both for these features.
- Use AI writing tools with care. AI-assisted text must meet the spam policies, including scaled content abuse.
For a founder-level walk through the basics behind all of this, see the free digital marketing module of the Fe/male Switch startup school.
How do you measure AI Overviews and AI Mode visibility?
Google announced the Generative AI performance reports in Search Console on 3 June 2026 and rolled them out to all websites worldwide by 31 August 2026. There is a report for Search and one for Discover. They show:
- Impressions: how many times your URLs appeared in generative AI features.
- Pages: which URLs appeared.
- Countries and devices (devices for Search only).
- Dates, with hourly, daily, weekly and monthly granularity.
The same data also sits inside the overall performance report. Google’s announcement lists impressions, not clicks, so treat the report as a visibility measure and watch organic sessions in your analytics for the traffic side. Track a fixed set of 20 to 50 questions your customers ask and note monthly whether your pages appear; AI visibility tracking covers the tools.
What mistakes should you avoid?
- Buying “AI Overview ranking” packages that promise a guaranteed spot. Google says no third party has access to its ranking or AI systems.
- Blocking snippets site-wide by accident through a plugin or template, then wondering why nothing is cited.
- Rewriting every page for AI. Google says it is unnecessary. Spend the hours on original content instead.
- Publishing near-duplicate pages per fan-out query. That is scaled content abuse.
- Confusing Google-Extended with Search. That token controls Gemini training and grounding outside Search and does not affect AI Overviews. See AI crawler access.
- Expecting clicks to match impressions. An AI answer can satisfy the searcher. Make your page worth the click with data, tools or depth the summary cannot hold.
Where can you learn more?
Read Google’s generative AI guide in full: it takes about ten minutes. Next steps on SEO100: understand the wider term in generative engine optimization and set up the Search Console reports.