Home / Resources / Optimizing for AI Overviews
GEO · Google AI OverviewsHow to optimize for Google AI Overviews
Google AI Overviews answer many searches at the top of the results page, often before a single blue link. This is a step-by-step guide to being one of the sources they cite: what Google has actually said about eligibility, how its query fan-out changes content planning, how to structure pages, and how to measure the result without fooling yourself.
Key takeaway: There is no special markup or trick for Google AI Overviews. A page is eligible to appear as a supporting link if it is indexed and can be shown in Search with a snippet, and the work that earns citations is ordinary but rigorous: answer the question and its related sub-questions directly, add original value, make the entity behind the content clear, and measure appearances query by query.
In this guide
- What AI Overviews are and how they choose sources
- What Google has actually said
- Step 1: Confirm your pages are eligible
- Step 2: Find the queries that trigger AI Overviews
- Step 3: Map the question and its fan-out
- Step 4: Write answer-first, in chunks
- Step 5: Add what only you can add
- Step 6: Make the entity and the page easy to trust
- Step 7: Page experience, media and freshness
- Step 8: Measure honestly
- Common mistakes
- A 30-60-90 day plan
- Should you opt out?
What AI Overviews are and how they choose sources
AI Overviews are AI-generated summaries that Google shows above the normal results for some searches. Each one synthesises information from several sources and displays links to the pages that support it. They are not a single-page excerpt like a classic featured snippet: the system writes a new answer, then attaches supporting links.
Google's documentation describes a technique called query fan-out for AI Overviews and AI Mode: instead of running only the words the user typed, the system issues several related searches across subtopics and data sources, then combines what it finds. That matters for content strategy, because your page may be retrieved for a sub-question that the user never typed.
No one outside Google knows the exact selection weights, so treat anything claiming a precise formula as speculation. What is observable is that overviews tend to appear on informational and multi-part questions more than on pure navigational queries, and that they lean heavily on pages that already perform well in Search. Ranking well is a filter, not a guarantee of a citation.
What Google has actually said
Google's Search Central guidance for AI features makes a few points that remove a lot of guesswork:
- No additional requirements. To appear as a supporting link in AI Overviews or AI Mode, a page needs to be indexed and eligible to be shown in Google Search with a snippet. There is no special markup, file or opt-in.
- Standard SEO best practice applies. The same fundamentals that help you rank (crawlable pages, helpful content made for people, good page experience) are the foundation.
- Existing controls apply. Settings such as
noindex,nosnippet,max-snippetanddata-nosnippetaffect whether and how your content can be shown. TheGoogle-Extendedtoken controls whether content is used for training and grounding of Gemini, and does not affect Search or AI Overview eligibility. - Structured data is not required, but it should match the visible content if you use it.
The practical consequence: llms.txt, "AI Overview schema" and similar add-ons are not part of Google's requirements. Spend the effort on the items below instead. Google updates this guidance, so confirm against the current Search Central documentation before acting on any single detail.
Step 1: Confirm your pages are eligible
Eligibility failures are the cheapest to fix and the most embarrassing to miss. Check each target page against this list:
| Check | What good looks like | How to verify |
|---|---|---|
| Indexable | Returns 200, no noindex, canonical points to itself | URL Inspection in Search Console |
| Crawlable | Not blocked in robots.txt; Googlebot can fetch it | robots.txt report, URL Inspection live test |
| Snippet allowed | No nosnippet or restrictive max-snippet on pages you want cited | View source; check headers and meta robots |
| Content in text | Key answers exist in the rendered HTML, not only in images or after interaction | URL Inspection rendered HTML; view with JavaScript off |
| Page experience | Fast, stable, mobile-friendly | PageSpeed Insights, Core Web Vitals report |
Two controls deserve care. A nosnippet or very low max-snippet value removes your text from snippets and from AI features, so audit templates and CMS defaults, especially after a redesign. If you need to hide part of a page, data-nosnippet on that element is more surgical than blocking the whole page:
<!-- Removes your text from snippets and AI features. Avoid on pages you want cited: -->
<meta name="robots" content="nosnippet">
<!-- Hide only a specific block from snippets: -->
<div data-nosnippet>Internal pricing notes</div>
Step 2: Find the queries that trigger AI Overviews
Do not optimise for AI Overviews in the abstract. Build a working list of queries where they appear and where you should be present:
- Start from your Search Console queries. Export queries with impressions for your key pages and sort by informational intent: "what is", "how to", "best way to", comparisons and multi-part questions.
- Check each one by hand in a clean browser session. Note whether an overview appears, which sources it cites and whether you are among them. Repeat from a second location if you serve multiple regions, because results vary by place, time and user.
- Record it in a sheet: query, overview present (yes/no), you cited (yes/no), competitors cited, date. This becomes your tracking baseline.
Expect volatility. Overviews appear and disappear for the same query over days, so you are looking for patterns across dozens of queries, not a single screenshot.
Step 3: Map the question and its fan-out
Because the system expands a question into related sub-searches, a page that only answers the head question can miss out. Take each priority query and list the sub-questions a thorough answer would need. For "how to choose payroll software for a restaurant group", the fan-out might include: what features matter for tipped staff, how multi-location payroll works, what it typically costs, how integrations with scheduling and point-of-sale tools work, and what compliance issues apply.
Then decide: one comprehensive page with a clear section for each sub-question, or a hub page linking to dedicated pages. Either works if each sub-question gets a direct, self-contained answer. Link them together with descriptive anchor text so the relationship is obvious, as covered in our guide to internal linking.
Step 4: Write answer-first, in chunks
AI systems retrieve and quote passages, not whole pages. Make every section a self-contained unit that makes sense when lifted out of context.
- Phrase headings as the question a buyer would ask.
- Put the direct answer in the first one or two sentences under the heading, then add detail, evidence and caveats.
- Use the right shape for the answer: a numbered list for steps, a table for comparisons, a short definition for "what is" questions.
- Avoid pronoun-heavy, context-dependent sentences such as "it also does this". Name the subject.
| Buried (weak) | Answer-first (strong) |
|---|---|
| "Pricing is something many of our customers ask about, and there are a lot of factors involved depending on what they need..." | "A typical project costs between X and Y, depending on three factors: scope, number of locations and integrations. Here is how each changes the price." |
Use real numbers and ranges you can stand behind. Vague promises give an engine nothing to quote. Our guide to chunk-formatted content goes deeper on structure.
Step 5: Add what only you can add
Google's own advice for AI features emphasises unique, non-commodity content. If your page says what ten others already say, an engine has no reason to pick it. Ways to add genuine value:
- Original data: your own benchmarks, survey results, price ranges or anonymised client patterns, with method and date stated.
- First-hand experience: specific examples, photos, process detail and lessons that a copy could not fake.
- Named expertise: an author with visible credentials, and an editor or reviewer where accuracy matters.
- Honest limits: say what you do not know or what depends on context. It builds trust with readers and avoids overclaiming.
AI-assisted drafting is fine if people edit and fact-check it and the result genuinely helps readers more than what already exists. Unreviewed, mass-produced pages are the pattern to avoid. See does AI-generated content hurt SEO?
Step 6: Make the entity and the page easy to trust
- Clear entity signals. Keep your organisation's name, address, description and profiles consistent everywhere, with a clear About page and author pages. See entity SEO.
- Structured data that matches the page. Organization, Article, Product or LocalBusiness markup helps systems parse what a page is about. It is not required for AI Overviews and does not guarantee anything. Never mark up content that is not visible. FAQPage markup rarely produces a visual result in Google now, so add it for machine readability, not for a rich result. See schema markup.
- Visible trust signals. Accurate contact details, clear ownership, sources cited for claims and updated dates. These matter most for health, finance and legal topics. See E-E-A-T.
Step 7: Page experience, media and freshness
- Page experience. Fast loading, stable layout and a usable mobile experience. These are table stakes rather than a differentiator, but a slow or broken page loses ground.
- Go beyond text where it helps. High-quality images and video with descriptive alt text, captions or transcripts support understanding and give readers more reasons to click.
- Freshness where it matters. For queries where currency matters (prices, regulations, tools, statistics), update the page with real changes and show the updated date. Do not change dates without changing content.
Step 8: Measure honestly
Measurement for AI features is imperfect, so use three sources and look for agreement between them:
- Search Console. Google reports performance for Search including its AI features within overall results, and does not necessarily break overview appearances out on their own. Check Google's current documentation for how AI features are counted, and watch impressions and clicks for the informational queries on your list.
- Your manual query log from Step 2, re-checked every two to four weeks.
- A prompt panel across other AI assistants, because the same fixes tend to help beyond Google. See how to track AI search visibility, or start with the free AI Rank Tracker.
Judge success by qualified traffic, brand searches and leads, not only citation counts. Some overview citations drive fewer clicks than a classic ranking, but they still build visibility and consideration.
Common mistakes
- Searching for an "AI Overview hack" instead of fixing indexing, content quality and entity clarity.
- Blocking snippets site-wide through a template default, then wondering why nothing is cited.
- Answering only the head query and ignoring the sub-questions an overview needs.
- Publishing large volumes of thin AI-written pages.
- Marking up content in schema that is not visible on the page.
- Treating one screenshot of an overview as proof of success or failure.
- Changing "last updated" dates without changing the content.
A 30-60-90 day plan
| Window | Focus | Output |
|---|---|---|
| Days 1-30 | Eligibility and baseline | Indexing and snippet audit fixed; query log of 30 to 50 informational queries with overview and citation status |
| Days 31-60 | Content upgrade | Top 10 pages rewritten answer-first with sub-question sections, original data or examples, named authors and updated dates |
| Days 61-90 | Trust and measurement | Entity clean-up and matching schema, new supporting pages for gaps, re-run the query log and prompt panel, compare with baseline |
Should you opt out?
You can limit use of your content with snippet controls, but a nosnippet or low max-snippet setting also removes your text from ordinary search snippets, which usually costs more than it saves. Most businesses are better off being cited accurately. If you are concerned about training, note that blocking Google-Extended is a separate decision that does not remove you from Search. See the complete GEO guide for the wider crawler policy.
This guide is one slice of generative engine optimization. The same fixes (crawlability, answer-first structure, original value and clear entity signals) also help in ChatGPT, Gemini, Perplexity and Claude. Reckona AI runs this as part of its GEO practice.
Frequently asked questions
Do I need special markup to appear in Google AI Overviews?
No. Google says there are no additional requirements. A page needs to be indexed and eligible to appear in Google Search with a snippet. Structured data is not required, but if you use it, it must match the visible content.
How does Google choose which sources to cite in an AI Overview?
Google has not published selection weights. Its documentation describes query fan-out, where the system runs several related searches across subtopics and combines the results. In practice, pages that already rank well, answer a sub-question directly and show clear expertise are the most likely candidates, but nothing guarantees a citation.
Can I block AI Overviews from using my content?
You can limit snippets with controls such as nosnippet, max-snippet and data-nosnippet, but they also affect ordinary search snippets. The Google-Extended token controls training and grounding of Gemini and does not affect Search or AI Overviews.
How do I track whether my pages appear in AI Overviews?
Combine Search Console data for your informational queries, a manual log that records whether an overview appears and whether you are cited, and a prompt panel across other AI assistants. Results vary by query, location and time, so track patterns over weeks.
Is optimising for AI Overviews different from SEO?
It builds on SEO. Indexing, helpful content and page experience remain the foundation. The extra emphasis is on answering a question and its related sub-questions directly, adding original value, making the entity clear and measuring appearances query by query.
Want to know if you are being cited?
Our free audit baselines your search health and checks how AI assistants describe your brand against your top competitor.
Get my free GEO audit →