How Gemini chooses what to cite: an SEO guide to Google's assistant

01What Gemini is, from an SEO point of view

Gemini is Google's standalone assistant and the model behind AI Overviews and AI Mode. The app and the search features share a model but not a citation pool: Omnia found Gemini and AI Overviews share only 38.5% of their top cited domains. Treat Gemini as its own surface.

Google has folded Gemini into Search, Android, Chrome and Workspace, which is why it matters even if your customers never open the Gemini app. When they ask a question in any of those places, the same retrieval habits apply: a live Google search, a short list of sources, and a confident answer.

02Which sites Gemini cites most

Community and reference sites dominate. In Ahrefs' September 2026 tracking of US queries, Reddit captured 28.5% of mention share among Gemini's top 50 cited domains, YouTube 14.6% and Wikipedia 8.8%. Retailers such as Walmart and expert publishers fill out the list.

The practical reading: Gemini leans on places where many people describe a brand, not on the brand's own site. Omnia's May–June 2026 citation database put 90.1% of Gemini citations on third-party editorial and independent content, with owned domains at 6.7%.

03Gemini cites less than AI Mode, so each slot matters more

Profound's tracking found Gemini pulls about 6.6 sources into an average answer and names around 4.4 brands, against 15.2 sources for AI Mode. Omnia measured 4.28 cited domains per answer. Fewer slots means a brand either makes the short list or is absent.

04How ranking in Google relates to Gemini

Ranking helps but does not guarantee anything. Ahrefs' analysis found about 38% of citations on Google's AI surfaces come from pages in Google's top 10. Query fan-out means Gemini runs many sub-searches, so niche pages that rank for specific questions can be cited even when your main page is not.

This is the same two-path model Ahrefs describes for all AI search: be present in the live retrieval (SEO), and be mentioned widely enough to be part of what the model already knows. For Gemini specifically, the second path runs through Reddit threads, YouTube videos and reference pages far more than through your homepage.

05What to do for Gemini

Five actions, in order of impact. Most overlap with good SEO; the Gemini-specific parts are crawler access and the third-party focus.

1. 1Allow Google-Extended. It is the robots.txt token Google documents for Gemini. Blocking it removes you from Gemini grounding without affecting Search rankings.
2. 2Be present where Gemini looks. Relevant Reddit threads answered properly, YouTube videos with clear transcripts, and the reference or review sites your category already cites.
3. 3Rank for the specific questions. Fan-out rewards pages that answer narrow sub-questions. Comparison, pricing and "near me" pages are the usual winners.
4. 4State facts plainly on your own site. With only a few slots, Gemini needs to repeat you accurately. Prices, locations, named people and dates in short, extractable blocks.
5. 5Keep it fresh. Ahrefs found AI-cited content is about 25% fresher than typical search results. Real updates, not new dates.

06Gemini and E-E-A-T: why entity clarity matters

Gemini is built on the same Knowledge Graph that powers Google Search, so brands that are clearly defined as entities (who you are, where you are, what you do, who vouches for you) are easier for it to retrieve and describe. Ambiguity costs you a slot.

In practice that means consistent naming across your site, Google Business Profile, LinkedIn, directories and the publications that mention you. If one source says you are an "SEO agency" and another says "digital consultant", the model has to guess. Pick one description and repeat it everywhere, including in your official FAQ and your About page.

It also means supporting the claims you want repeated. Gemini cites few sources, so it leans on ones it can verify. A page that states a price range and a service area, with a dated update and a named person behind it, is a stronger grounding candidate than a slogan.

07A worked example: a Melbourne services brand

Ask Gemini "best SEO expert in Melbourne" and you will typically see three to five brands drawn from a local publication, a review platform and a Reddit thread, not from the brands' own sites. If you are missing, the fix is on those pages, not yours.

The sequence I use: run the prompt list in Gemini monthly, note every cited domain, mark the ones that name competitors but not you, then earn a place on them with something useful to the author (data, an expert comment, a case study). In parallel, make sure the facts those pages would repeat are stated clearly on your own site so the model describes you correctly once it does include you.

Expect movement to be gradual. Gemini draws on the live search and on consensus across the web, and consensus takes months to build. A quarter is a fair window to judge whether the plan is working.

08How to measure Gemini visibility

Gemini passes referrer data inconsistently, so use three signals: a Gemini filter in your GA4 AI channel, a fixed monthly prompt list run in the Gemini app, and the "How did you hear about us?" question on your forms.

Goodie's B2B GA4 panel put Gemini at 10.6% of measurable AI referrals in March–April 2026, third behind ChatGPT and Claude, so expect modest but real traffic. Record mentioned, cited and accurate for each prompt, compare quarters rather than months, and watch which third-party sites are doing the citing; that list is your outreach plan.

09Gemini in the monthly report

Two lines: Gemini referral sessions and conversions from GA4 (expect undercounting), and the fixed prompt list run in the Gemini app with mentioned, cited and accurate recorded per prompt. Add the list of citing domains each month; the ones that name competitors but not you are next month's outreach.

If you also track AI Overviews and AI Mode, keep them as separate rows. The three Google surfaces share a model and little else, and a brand can be strong in one and absent from another. The fix is different for each, so the report should show which one moved.

10Common questions about Gemini and SEO

Does blocking Google-Extended hurt rankings? No; it only affects Gemini grounding and training. Does ranking first guarantee a Gemini citation? No; about 38% of citations on Google's AI surfaces come from the top 10, and fan-out means many sub-questions decide the answer. Is schema required? Helpful for Search features, but Ahrefs found no proven effect on AI citations.

One more that comes up: "Should we optimise for Gemini or for AI Overviews?" Both, but separately. They share a model and only about 38.5% of their top cited domains, so the prompt list, the cited-domain list and the outreach targets need to be tracked for each. The content habits (answer first, specific facts, freshness) are the same; the places you need to be mentioned are not.

11The short version

Gemini is grounded in Google but picky. Rank for the specific questions, be mentioned on the sites it trusts, keep your facts clear and fresh, and let Google-Extended in.