How to rank in Gemini

To rank in Gemini, a page has to be in Google's search index and allowed for Gemini's use. Google says "some Gemini responses are grounded on Search results" (opens in a new tab), and that a robots.txt token, Google-Extended, controls a page's use for grounding in Gemini Apps, which Google describes as "providing content from the Google Search index to the model at prompt time" (opens in a new tab). So let Googlebot crawl and index the page, do not disallow Google-Extended, and keep the facts in the page's text.

Google documents the index and the control. It does not document when the Gemini app searches or how it chooses which results to cite, so everything after crawler access comes from outside studies. This guide keeps the two apart. It covers the Gemini app; Google's AI answers inside Google Search are covered in how to rank in Google AI Overviews and AI Mode. It is one of a set on how AI assistants find and cite sources, and its sources were checked on September 27, 2026.

Scope and definitions

The Gemini app is Google's AI assistant. Google's privacy hub defines "Gemini Apps" as the web app at gemini.google.com and in browser sidebars, the Gemini mobile apps, and Gemini in Chrome, Messages, Maps, Android Auto, and other Google products (opens in a new tab). Google says more than 1 billion people use the Gemini app every month (opens in a new tab) (company figure, August 2026). Developers reach the same models through the Gemini API and Google Cloud, and several studies below used those routes, which can behave differently from the app.

Work to be cited in AI answers is called answer engine optimization (AEO) or generative engine optimization (GEO).

• Grounding is basing an answer on retrieved sources. For Gemini, Google defines it as "providing content from the Google Search index to the model at prompt time to improve factuality and relevancy" (opens in a new tab).
• The search index is the store of pages Google has crawled with Googlebot and processed.
• Google-Extended is a robots.txt token, not a separate crawler: "Crawling is done with existing Google user agent strings" (opens in a new tab), and the token is read as an instruction.
• Parametric knowledge is what a model learned in training. Google says search grounding lets Gemini "cite verifiable sources beyond its knowledge cutoff" (opens in a new tab), so a company too new for the training data can reach an answer only through search.

How Gemini finds and cites pages

What Google documents

Google's statements about the Gemini app and search are spread across several documents (official documentation):

When Gemini searches

Google does not document when the Gemini app searches. It does document the developer version. In the Gemini API, "The model analyzes the prompt and determines if a Google Search can improve the answer" (opens in a new tab), and "If needed, the model automatically generates one or multiple search queries and executes them" (opens in a new tab) (developer documentation, updated September 23, 2026). One question can therefore become several searches, and a page can be cited because it answers one of them.

In an academic audit, Gemini 2.5 Flash with Google Search, called through the API, searched on every simple factual question it was given and retrieved about 9 web pages per query on average across 4,706 queries (opens in a new tab) (preprint; 2025). The app may behave differently: the Gemini app's displayed sources and the API's shared 14.8% of domains for the same product questions (opens in a new tab) (preprint; September 2026).

What Google-Extended controls

Google-Extended has two uses: training future Gemini models, and grounding in Gemini Apps and in Google Cloud's grounding service. Google's cloud documentation states the second plainly: its grounding "does not use web pages for grounding that have disallowed Google-Extended" (opens in a new tab). By Google's account, blocking the token does not remove a page from Google Search.

Two consequences follow. First, one token covers both uses, so a site that disallows Google-Extended to stay out of training also leaves grounding in the Gemini app. Second, because the token has no user agent of its own, it is set in robots.txt alongside your other rules. Google's crawlers follow the most specific group that matches them and fall back to the * group when no group names them (opens in a new tab), so check both any Google-Extended group and your * group. The rules themselves are covered under robots.txt.

A study shows what blocking does in practice. In December 2025 tests of Gemini 2.5 Flash with search grounding, 21 popular publishers that Google Search and AI Overviews each retrieved for at least 20 queries were never cited by Gemini, and all of them block Google-Extended (opens in a new tab). Facebook, Instagram, TikTok, IMDb, Yelp, and Tripadvisor also received no Gemini citations, and they block it too (opens in a new tab) (peer-reviewed). The authors conclude that Google respects the choice. They also found that sites blocking the token were less likely to be retrieved by AI Overviews (opens in a new tab), which Google says the token does not affect; that finding is an association, not a tested cause.

Crawling, rendering, and fetches for users

Grounding draws on the Google Search index, which Googlebot builds. A page therefore has to be crawled and indexed like any page in Google Search; the steps are covered in how to rank in Google AI Overviews and AI Mode. In one practitioner test of five AI systems, only Gemini ran JavaScript when it fetched a live page (opens in a new tab) (one test page, October 2025). Server-rendered text avoids depending on that.

Google also runs fetchers that act for a user rather than for the index. Its documentation lists Google-Agent, "used by agents hosted on Google infrastructure to navigate the web and perform actions upon user request," and a Gemini Notebook fetcher for URLs that users add as sources; because users request these fetches, they "generally ignore robots.txt rules" (opens in a new tab). The Gemini privacy notice says some Gemini features are designed to help users complete tasks, including "interacting with websites" (opens in a new tab).

What Google does not document

• When the app searches. Only the API's rule is published, as described above.
• How it chooses and orders sources. For AI Overviews and AI Mode, Google says its generative AI features on Search are "rooted in our core Search ranking and quality systems" (opens in a new tab). The documents reviewed for this guide make no equivalent statement about the Gemini app.
• A report for site owners. Search Console's generative AI report counts impressions in AI Overviews and AI Mode (opens in a new tab), and its opt-out setting covers AI Overviews, AI Mode, and generative AI features in Google Discover (opens in a new tab). Neither lists the Gemini app.
• How it describes itself. Google warns that Gemini "can misrepresent how it works, including in response to prompts asking how it was trained or how it carries out various functions (like citing sources, or providing fresh information)" (opens in a new tab). Asking Gemini how it picks sources is not a reliable way to find out.

What research shows about Gemini's citations

Studies of Gemini differ in version, access route, and topic, so each finding below carries its setting:

• Often no links. 38% of Gemini's responses cited no website (opens in a new tab) (preprint; 55,936 queries; July–August 2025). In a study by 22 public broadcasters of the free apps, 42% of Gemini's answers gave no direct source, meaning no URL pointing to specific content (opens in a new tab) (news questions; late May to early June 2025).
• Named sources without links. In the same study, 54% of Gemini's answers made sourcing claims that were wrong or could not be checked, such as "according to" a named news organization while linking elsewhere or nowhere; no other assistant exceeded 4%. 72% of Gemini's answers had a significant sourcing issue (opens in a new tab). An assistant that names organizations without linking them can attribute claims to a company that did not make them.
• Broken links in early 2025. When asked to identify the source of news excerpts, more than half of Gemini's responses cited fabricated or broken URLs (opens in a new tab) (1,600 queries across eight tools; March 2025).
• More brand-owned sources. In API tests with ranking-style prompts, brand-owned sites made up 25.1% of the domains Gemini cited for well-known brands and 21.2% for niche brands, more than Claude or ChatGPT; for banking prompts, brand sources were often half or more of Gemini's citations (opens in a new tab) (preprint; Gemini 2.5 Flash with search grounding; mid-2025). A company's own pages have a better chance of being a Gemini source than a ChatGPT one.
• Beyond the most popular sites. When only the top-ranked source was counted, domains in the 1,000 most visited sites supplied it for 52.7% of queries in Google Search, 40.0% in AI Overviews, and 32.6% in Gemini, and Gemini cited the most distinct domains ranked outside the top 10,000 (opens in a new tab) (peer-reviewed; December 2025).
• A different set of sources from Google's other products. Of all the sources AI Overviews and Gemini returned for a query, about 11% were returned by both; for Gemini and Google Search the figure was 16%, and for Amazon shopping queries 8% (opens in a new tab). In the consumer apps, ChatGPT and Gemini shared only 5.4% of the domains they displayed for the same query (opens in a new tab) (preprint). An AI Overview citation does not imply a Gemini citation.
• Reviews, forums, and video. Gemini favored YouTube more than traditional Google Search did (opens in a new tab). In product questions, the domains the Gemini app displayed most often were pcmag.com (in 12.4% of answers with sources), reddit.com (8.0%), and youtube.com (6.9%) (opens in a new tab) (logged-out app; preprint).
• Younger pages. In a vendor study of about 17 million cited URLs, Gemini's citations averaged 1,118 days since publication and 831 days since their last update, against 1,416 and 1,047 days for Google's organic results (opens in a new tab) (July 2025).
• Answers change between runs. Two runs of the same query returned source lists with a Jaccard similarity of 0.46 in Gemini, against 0.66 in AI Overviews and 0.78 in Google Search (opens in a new tab), meaning that less than half of Gemini's combined sources appeared in both runs. In the logged-out app, repeated product questions shared 29.8% of their displayed domains (opens in a new tab). In German-language tests, Gemini's cited sources were the most stable of four engines within 24 hours (Jaccard 0.505), though its brand lists were not (0.409) (opens in a new tab) (preprint; first author affiliated with a tracking vendor).
• Few first-person picks. In product questions, Gemini stated a first-person preference in 7% of the answers that recommended products, against 79% for ChatGPT, and labeled a product the best in 43%, against 73% (opens in a new tab) (preprint).
• Well-known brands start ahead. In API tests of skincare recommendations with three models including Gemini 3 Flash, a real brand was chosen in every trial against nine fictional ones when specifications were equal, but a fictional brand won half the time with a 0.075-star rating edge, 1.6 times the reviews, or a 7.3% lower price (opens in a new tab) (preprint; laboratory).
• Manipulation. Across 13 model back ends, an attack that plants endorsements in web content succeeded 31.4% of the time against Gemini 3 Flash, the highest rate, and 0.0% against Claude Sonnet 4.6 (opens in a new tab) (preprint; we read the abstract only).

How much traffic Gemini sends

• Share of referrals. In a mostly B2B referral panel, Gemini's share of AI referrals rose from 2.4% in May to August 2025 to 10.6% in March to April 2026 (opens in a new tab) (vendor panel). In an earlier study of 3,000 sites, Gemini sent 17.6% of AI referral traffic (opens in a new tab) (vendor study; February 2025).
• Analytics. GA4's AI Assistant channel names Gemini (opens in a new tab), and Microsoft Clarity counts visits from Gemini's standalone site but not from AI features embedded in other products (opens in a new tab).
• Use. 24% of U.S. adults said they ever use Gemini (opens in a new tab) (survey, February 2026). Among U.S. B2B professionals who use AI at work, 62% use Gemini for work and 61% for product research (opens in a new tab) (vendor survey; 519 respondents; March–April 2026).

Steps

Steps 1 to 3 follow Google's documentation. Steps 4 to 10 are AEO HQ's recommendations based on the evidence above.

1. Get the page crawled and indexed in Google Search. Grounding draws on the Google Search index (opens in a new tab). Indexing steps are in the AI Overviews guide.
2. Do not disallow Google-Extended on pages you want Gemini to use. Read robots.txt and find the group that applies: a Google-Extended group if you have one, otherwise the * group (opens in a new tab).
3. Decide on training and grounding together. One token controls both (opens in a new tab). If you disallow it to keep content out of training, expect to lose Gemini grounding too.
4. Put the answer and the key facts in the page's text. State prices, scope, dates, and definitions in server-rendered HTML near the top, not only in scripts, images, or structured data.
5. Make your own site a good source about your company. Gemini cites brand-owned pages more often than ChatGPT or Claude do (opens in a new tab). Give your offer, pricing, and method pages the facts a buyer's question needs.
6. Answer the sub-questions. Because one prompt can become several searches (opens in a new tab), give each distinct buyer question a page or section that answers it, with no near-duplicate pages.
7. Earn coverage on independent sites and in video. Most of the domains Gemini cites are independent media, review, or comparison sites (opens in a new tab), and it favors YouTube more than Google Search does (opens in a new tab). See brand mentions and AI recommendations.
8. Keep pages current and dates true. Change a page's date only when its content changes.
9. Check what Gemini says about you, and whom it credits. Ask it about your company and your category, with and without signing in, and check both the facts and the attributions; Gemini has named sources it did not link (opens in a new tab).
10. Measure without a first-party report. Count Gemini visits in GA4 or Clarity, and run a fixed set of buyer questions several times over several weeks. Server logs cannot isolate Gemini grounding, because Google-Extended has no user agent of its own (opens in a new tab). Methods are covered in how to measure AI visibility and how to track AI referral traffic in GA4.

Other assistants use other indexes. Microsoft's assistant searches Bing; see how to rank in Microsoft Copilot.

What the evidence shows and does not show

Antipatterns

An antipattern is a practice that looks helpful but fails or backfires.

1. Blocking Google-Extended and expecting Gemini citations. The token governs grounding in Gemini Apps (opens in a new tab), and blocking publishers were never cited in one study (opens in a new tab). Instead, decide on the token knowing it covers both training and grounding.
2. Using the Search Console setting to manage Gemini. It covers AI Overviews, AI Mode, and Discover (opens in a new tab), not the Gemini app. Instead, use Google-Extended for Gemini.
3. Assuming an AI Overview citation means a Gemini citation. The two share about 11% of sources (opens in a new tab). Instead, test Gemini separately.
4. Asking Gemini how it chooses sources. Google says it can misrepresent how it works (opens in a new tab). Instead, rely on Google's documentation and on repeated tests.
5. Hidden instructions or planted endorsements. They succeeded against Gemini 3 Flash in a lab test (opens in a new tab), and Google's spam policies cover attempts to manipulate generative AI responses in Google Search (opens in a new tab), the index Gemini grounds on. Instead, state true facts plainly.
6. Facts only in scripts or structured data. Gemini ran JavaScript in one test (opens in a new tab), but other systems did not. Instead, put facts in server-rendered HTML.
7. Judging from one answer. Less than half of Gemini's sources repeated between two runs (opens in a new tab). Instead, repeat questions over several weeks.
8. Ignoring wrong attributions. Gemini has credited named organizations for claims their pages did not make (opens in a new tab). Instead, check what Gemini attributes to your company, and correct the public pages it could be drawing on.

Checklist

FAQ

Does Gemini use Google Search?

For some answers. Google says "some Gemini responses are grounded on Search results" (opens in a new tab), and describes grounding as providing content from the Google Search index to the model (opens in a new tab). Not every answer searches, and not every answer includes sources (opens in a new tab).

If I block Google-Extended, will my pages leave Google Search or AI Overviews?

Google says the token "does not impact a site's inclusion in Google Search nor is it used as a ranking signal in Google Search" (opens in a new tab). One study found that sites blocking it were less likely to be retrieved by AI Overviews (opens in a new tab), an association the study could not explain. The token does keep your content out of grounding in the Gemini app.

Is Gemini the same as AI Overviews and AI Mode?

No. They are separate products with separate controls: Search Console's generative AI setting covers AI Overviews, AI Mode, and Discover (opens in a new tab), while Google-Extended covers Gemini Apps (opens in a new tab). They also cite different pages: about 11% of sources in common (opens in a new tab). See how to rank in Google AI Overviews and AI Mode.

Can I see how often Gemini cites my site?

Not directly. Search Console's generative AI report covers AI Overviews and AI Mode (opens in a new tab), not the Gemini app. Visits from Gemini appear in GA4's AI Assistant channel (opens in a new tab). Citations have to be measured with repeated prompt tests.

Why does Gemini answer without links?

Google says not all responses include related links or sources (opens in a new tab). In one audit, 38% of Gemini's responses cited no website (opens in a new tab).

How long until Gemini cites a new page?

No study has measured it. The page first has to be crawled and indexed by Google, which the AI Overviews guide covers.

Change log

• September 28, 2026: First published.

Next steps

To check whether AI crawlers can reach your pages and how AI assistants answer your buyers' questions, AEO HQ's Instant AEO Audit ($499) is an automated audit of your site and up to three competitors. It includes a live crawl of your key pages and checks of AI crawler access, schema, and llms.txt.