---
title: "How to get cited by Gemini and AI Overviews"
description: "What Google-Extended controls, why AI Overviews are governed by different rules, what query fan-out is, and why Google says no special optimisation is needed."
url: https://aeoly.ai/blog/how-to-get-cited-by-gemini-and-ai-overviews
published: 2026-08-09
author: "Aeoly"
publisher: "Aeoly"
---
# How to get cited by Gemini and AI Overviews
Gemini and AI Overviews are different surfaces with different controls, and Google-Extended only governs one of them. Google also says neither needs special optimisation.
Google-Extended is the control people reach for when they want to influence Gemini, and it is the one most often described wrongly. Google's documentation states plainly that Google-Extended does not impact a site's inclusion in Google Search nor is it used as a ranking signal in Google Search.
## How to get cited by Gemini and AI Overviews
Get the page indexed and keep it eligible to be shown with a snippet, because that is the gate Google names for both surfaces. Leave Google-Extended allowed unless you want to opt out of Gemini training, since Google states it has no effect on Search. Then answer each question in its opening paragraph.
Two surfaces, and they are not governed by the same control, which is the single most common mistake made about this. AI Overviews are a feature of Google Search. The Gemini app and the Vertex grounding products are what Google-Extended governs. A rule aimed at the wrong one changes nothing about the surface you cared about.
The robots.txt side is one line, and the important thing about it is what the line does not do.
``` robots.txt
# Governs Gemini training and grounding only.
# Google states this has no effect on Search inclusion or ranking.
User-agent: Google-Extended
Allow: /
```
Which means the work that actually earns a Gemini citation is the work that earns a ranking, and the failure modes are the familiar ones. Before writing anything new, establish three numbers that most teams have never written down.
- How many pages you have published, as a number rather than an impression.
- How many of them Google has indexed. Search Console reports this, and the gap is usually wider than anyone expects.
- How many of the indexed ones answer a question in their opening paragraph. This is the one you have to check by reading.
The first two set a ceiling on everything else: a page no index holds cannot be quoted by a system that grounds in that index, however well written it is. The third is the only one that writing fixes.
It is also worth being precise about which surface you mean, because the advice differs and most posts blur all three together. AI Overviews are a feature of Google Search and follow Search's rules. The Gemini app and the Vertex grounding products are the surfaces this token governs. A control aimed at the wrong one does nothing at all.
## What does Google-Extended actually control?
Google-Extended is a robots.txt token that controls whether content Google has already crawled may be used for training Gemini models and for grounding. Google names three surfaces: Gemini Apps, the Vertex AI API for Gemini, and Grounding with Google Search on Vertex AI. It is not a crawler and it does not change what Googlebot fetches.
[Figure: A diagram showing Googlebot crawling a page once into Google's index. Two paths leave the index. The left path to Google Search ranking and AI Overviews has no gate. The right path to Gemini Apps and Vertex AI grounding passes through a gate labelled Google-Extended. Closing that gate leaves the left path untouched.]
_Google states that Google-Extended does not impact inclusion in Google Search and is not a ranking signal._
That last point catches people out repeatedly. Google-Extended does not crawl anything. Googlebot crawls your site once, and the token decides what that crawl may be used for afterwards.
[Figure: A comparison of what people assume and what is true. The assumption is that Google-Extended is a separate crawler that fetches pages and can be blocked to stop it. In fact Googlebot does all the crawling and Google-Extended is only a robots.txt token expressing what the already-crawled content may be used for afterwards.]
_This is why blocking it saves you no crawl budget and changes nothing about what Google fetches._
## Does blocking Google-Extended protect my rankings?
Some publishers add Google-Extended to robots.txt believing it is a general AI opt-out that also keeps them out of AI Overviews in Search. It is not. Google's documentation is explicit that the token has no effect on inclusion in Search and is not a ranking signal.
> You can block Google-Extended and still appear in an AI Overview. The two are governed by different things.
It takes a few seconds to confirm which of the two your own site allows. The AI crawler checker puts Googlebot and Google-Extended on separate rows, because they are separate decisions and only one of them can cost you an AI Overview.
In our survey, 10 of 48 readable robots.txt files disallowed Google-Extended, about 21 percent. Seven of the ten were news publishers. No documentation or reference site disallowed it, and exactly one SaaS marketing site did.
[Figure: A bar chart of which categories disallow Google-Extended. News accounts for seven of the ten. Documentation and reference sites disallow it in zero cases, SaaS in one, retail in one and community in one.]
_If you sell software, effectively none of your competitors are opting out of this. Aeoly's survey, September 2026._
Figures come from Aeoly's robots.txt survey of 52 well-known sites, run on 12 September 2026. Each site's robots.txt was read once with our own user agent and all seven crawler tokens were evaluated against that one file. 48 sites returned a readable robots.txt and the other 4 are excluded rather than counted as permissive. The survey reads robots.txt only and never crawls the sites themselves.
## What actually earns a Gemini citation?
Gemini's grounding leans on Google Search. Which means the work that earns a citation is, unglamorously, the work that earns a ranking: being indexed, being crawlable, answering the query, and being the page a system would reasonably quote.
That is good news for anyone who has been doing technical SEO properly. It also means the failure modes are the familiar ones. A page Google has not indexed cannot be quoted by a system that grounds in Google's index, and no amount of robots.txt tuning changes that.
[Figure: A funnel showing pages published, pages crawled, pages indexed, and pages eligible to be cited. Each stage is smaller than the last. A note explains that a system grounded in an index cannot quote a page that index does not hold, so the indexed count is the ceiling on everything else.]
_Google's URL Inspection reports this one URL at a time, which is why most teams have never established a site-wide figure._
It is worth being precise about which surface you mean, because the advice differs. AI Overviews are a feature of Google Search and follow Search's rules. The Gemini app and the Vertex grounding products are the surfaces Google-Extended governs. A single post that treats all three as one thing will hand you a control that does not apply to the surface you care about.
## What is AI Mode and how is it different?
AI Mode is a Search surface that answers a question by running many searches at once rather than one. Google calls the technique query fan-out and defines it as a set of concurrent, related queries generated by the model to request more information and fetch additional relevant search results. AI Overviews may use the same technique.
[Figure: A diagram showing one question in AI Mode expanding into several concurrent sub-queries run at the same time, each returning its own results, which are then combined into a single answer carrying links. Google defines query fan-out as a set of concurrent related queries generated by the model to fetch additional relevant search results.]
_Google's own definition. AI Mode breaks a question into subtopics and runs the searches simultaneously; AI Overviews may use the same technique._
The practical difference from a blue-link result is the number of doors into your site. One search matches one query against your pages. A fanned-out question matches several sub-questions at once, and Google says the models identify more supporting pages while the response is being generated, which allows a wider and more diverse set of links than a classic web search.
Which sounds like an invitation to write a page per sub-question. Google explicitly says not to, and this is the part worth reading twice.
## Should I write pages that target fan-out queries?
No. Google's guidance on optimising for generative AI features says creating separate content specifically to target fan-out queries violates its spam policies. The queries are generated by the model at answer time, they are not a keyword list, and a set of thin pages built to match them is the thing the policy exists to catch.
This is the sharpest example in the whole series of advice not transferring between assistants, so it is worth being exact about it.
[Figure: A comparison. Third-party research on ChatGPT finds cited pages have titles closer to the assistant's internal sub-questions, which reads as an argument for writing to them. Google states that creating separate content specifically to target fan-out queries violates its spam policies. The resolution offered is that answering a real sub-question inside a page is ordinary structure, while generating pages to match model-produced strings is not.]
_Both statements can hold. One is an observed correlation on a closed system measured by somebody else; the other is the platform owner stating a rule it enforces._
- Third-party research on ChatGPT finds that cited pages have titles closer to the assistant's internal sub-questions than to the original prompt, which reads as an argument for writing to those sub-questions.
- Google, writing about its own surfaces, calls building content specifically for fan-out queries a spam policy violation.
- Both can be true. One is an observed correlation on a closed system somebody else measured; the other is the platform owner stating a rule it enforces.
The way through is the distinction between writing a page that genuinely answers a question somebody has, and generating pages to match strings a model produced. A section covering a real sub-question your readers ask is ordinary good structure. Forty pages spun from a fan-out list is a doorway farm with a new name on it, and Google has said which of those it treats as spam.
> The queries are generated at answer time by a model. They are not a keyword list, and treating them as one is the mistake the policy names.
## Do I need to do anything special for AI Overviews and AI Mode?
Google says no. Its guidance states there are no additional requirements to appear in AI Overviews or AI Mode and no special optimisations necessary, because the generative features are rooted in its core Search ranking and quality systems. The same page says structured data is not required for generative AI search and there is no special schema.org markup to add.
That is a plain statement from the platform owner, and a large industry is currently selling the opposite. It is worth holding both facts in view: nobody has to take Google entirely at its word, and nobody selling you a generative-AI schema package has a better source than this one.
What the same guidance does ask for is unremarkable and checkable.
- The page must be indexed and eligible to appear in Google Search with a snippet. That is the one hard technical gate, and it is the one a stray nosnippet or noindex quietly closes.
- The page must meet the technical requirements for Search and follow the Search policies.
- The content should be helpful and written for people, which is the same sentence Google has been publishing for a decade.
So the honest instruction for Gemini surfaces is the boring one. Get indexed, stay snippet-eligible, answer the question in a paragraph that stands alone, and do not buy a separate discipline for it. Then read the generative AI performance report to see whether it worked.
## How do I see whether Gemini features are showing my pages?
Google now reports it. Search Console has a generative AI performance report covering impressions in AI features on Search, including AI Overviews and AI Mode, and a companion report for Discover. Google has described it as rolling out to a subset of sites, so check whether yours has it before assuming the data is missing.
That is a real measurement from the party that owns the surface, which puts it in a different class from asking Gemini questions by hand and writing down what you happened to see. Two things are worth knowing about how the numbers behave.
[Figure: An illustration of Search Console's generative AI performance report, listing pages against impressions in AI features on Search. Two notes record that pages in AI features are also counted in ordinary Web search traffic, and that an AI Overview occupies one position with every link inside it sharing that position.]
_Google has described the report as rolling out to a subset of sites, so its absence from your account is not evidence that your pages are absent from the feature._
- Pages appearing in AI features are also counted in ordinary Search traffic, in the Web search type. The generative AI report is a dedicated view rather than a separate universe, so do not add the two together.
- An AI Overview occupies a single position, and every link inside it is assigned that same position. So average position for a page cited in an Overview means something different from average position for a blue link, and comparing them directly will mislead you.
Read it the way you would read any impressions report: as a trend over weeks, filtered to the pages you were trying to get in. A single week tells you about that week.
## How should I structure a page for AI Overviews?
Answer the query in the opening paragraph under a heading that asks it, keep the page eligible for a snippet, and make sure the page is indexed. Google publishes guidance on AI features that does not amount to a separate optimisation discipline: the pages eligible for AI features are the pages eligible for Search.
Which is a duller answer than the field would like, and it is the one supported by what Google actually says. Some specifics do follow from it.
- Do not block snippets if you want to be quoted. A page carrying nosnippet, or a restrictive max-snippet, is telling Google not to show the passage that an AI feature would have shown.
- Structured data helps a machine parse the page and is not a shortcut past being useful. Add it properly where the type fits; expect nothing from it on its own.
- Alt text is the only description of an image a text system can read. On a page whose argument depends on a chart, an unlabelled chart is an argument the system cannot see.
- Keep one question per section. A heading covering three questions produces a passage that answers none of them extractably.
``` HTML head
```
That distinction catches people out constantly, which is why it is worth stating twice in one post. There are three separate controls here and they do three separate things: snippet directives shape what Search shows, Google-Extended shapes what Gemini may use, and noindex removes the page from the index entirely.
## How do I check if Google has indexed my pages?
Ask how many of your pages Google has actually indexed, as a fraction of how many exist. Most teams have never established that number, and it is the ceiling on everything else.
[Figure: Four Search Console verdicts and what each means. URL is on Google means it can be retrieved and cited. Crawled but currently not indexed means it cannot. Discovered but not currently indexed means it has not been fetched. Excluded by noindex tag means you asked for this outcome.]
_Only the first row is eligible. The other three are pages you believe are working._
Search Console will tell you page by page, for free, and it is the only source that speaks for Google itself. Aeoly runs the same check on a schedule through Google's URL Inspection API where Search Console is connected, so the number is a trend rather than something you remembered to look at once. Either way, look: indexing coverage is the cheapest diagnostic in this whole field.
- Confirm the page is indexed, using URL Inspection rather than assuming.
- Give every answer its own heading and a self-contained opening paragraph.
- Keep the fact and its source in the same sentence, so a grounded answer carries your attribution.
- Give images real alt text. It is the only description of a picture a text-based system can read, and it is free.
## How to get cited by Gemini and AI Overviews: the short version
Leave Google-Extended allowed unless you have a specific reason to opt out of model training, and understand that the choice is about training and grounding rather than about ranking. Then do the ordinary work: get indexed, answer the question in extractable paragraphs, and publish something worth quoting. Getting cited by Gemini is mostly the same job as ranking in Google, which is either reassuring or disappointing depending on how you have been spending your time.
The other assistants make you choose a crawler rather than a permission. ChatGPT and Perplexity both document theirs, and it is worth knowing which of your lines does what.
## More questions
### Does blocking Google-Extended remove me from AI Overviews?
No. Google-Extended governs the use of crawled content for training Gemini models and for grounding in Gemini Apps and Vertex AI. Google's documentation states it does not impact inclusion in Google Search and is not a ranking signal. AI Overviews are a Search feature and are not controlled by this token.
### Is Google-Extended a crawler?
No, and this is the most common misunderstanding about it. It is a robots.txt user-agent token used to express a permission. Googlebot does the crawling. Google-Extended decides what that crawled content may subsequently be used for, which is why blocking it saves you no crawl budget.
### Will blocking Google-Extended hurt my traffic?
Not through Search, according to Google's own documentation. What you lose is eligibility for the Gemini surfaces the token governs. Whether that costs you anything depends on how much of your audience asks Gemini questions your pages could answer, which is worth measuring rather than guessing.
### What is query fan-out?
Google defines it as a set of concurrent, related queries generated by the model to request more information and fetch additional relevant search results. AI Mode uses it to break a question into subtopics and run many searches at once, and AI Overviews may use it too. The queries are produced at answer time rather than drawn from a fixed list.
### Should I create pages targeting fan-out queries?
No. Google's guidance on optimising for generative AI features states that creating separate content specifically to target fan-out queries violates its spam policies. Covering a genuine sub-question inside a page your readers want is ordinary structure; generating pages to match model-produced strings is the thing the policy names.
### Do I need structured data for AI Overviews or AI Mode?
Google says no: structured data is not required for generative AI search and there is no special schema.org markup to add for it. Structured data remains useful for the search features it actually governs, so add it properly where a type genuinely fits your page, and do not expect it to buy generative visibility.
### Does Search Console report AI Overview impressions?
Yes. Google has introduced a generative AI performance report in Search Console covering impressions in AI features on Search, such as AI Overviews and AI Mode, plus a version for Discover. Google described it as rolling out to a subset of sites, so its absence from your account does not mean your pages are absent from the feature.
### Does nosnippet stop me appearing in AI Overviews?
It removes the text passage Google would otherwise show, which is the material an AI feature draws on. It is a Search-level directive rather than an AI-specific one, and it is a different control from Google-Extended, which governs Gemini training and grounding instead.
### Is AI Mode the same thing as an AI Overview?
They are different surfaces in Google Search, and Search Console reports on both under its generative AI features. The practical advice does not diverge much, because both draw on the same index and the same eligibility rules.
### How do I check whether Google has indexed a page?
Use URL Inspection in Google Search Console. It reports whether a URL is indexed and, when it is not, gives a reason such as crawled but not currently indexed or excluded by a noindex tag. It inspects one URL at a time, which is why most teams never establish a site-wide baseline.
## Sources
- Optimizing your website for generative AI features on Google Search (Google Search Central): https://developers.google.com/search/docs/fundamentals/ai-optimization-guide
- Expanding AI Overviews and introducing AI Mode (Google blog): https://blog.google/products-and-platforms/products/search/ai-mode-search/
- Spam policies for Google web search (Google Search Central): https://developers.google.com/search/docs/essentials/spam-policies
- Introducing generative AI performance reports in Search Console (Google Search Central blog): https://developers.google.com/search/blog/2026/06/gen-ai-performance-reports
- Generative AI performance report (Search) (Search Console help): https://support.google.com/webmasters/answer/16984139
- AI features and your website (Google Search Central): https://developers.google.com/search/docs/appearance/ai-features
- Google crawlers and user-triggered fetchers, including Google-Extended (Google Search Central): https://developers.google.com/search/docs/crawling-indexing/overview-google-crawlers
- URL Inspection tool (Google Search Console help): https://support.google.com/webmasters/answer/9012289
- AI crawler robots.txt survey, 52 sites (Aeoly. Method and site list in scripts/ai-crawler-survey.ts)