How to Optimize for Gemini Search: What I Have Found After 6 Months of Testing

We will admit it. When Google first bolted Gemini onto Search, we filed it under "panicked answer to ChatGPT" and got on with our day. Another AI product destined for the Google graveyard, next to Google+ and Stadia.

That was a daft call. Gemini is now the engine behind AI Overviews and AI Mode, which means it sits in front of more eyeballs than almost any product in tech. Google told investors back in July 2025 that AI Overviews had crossed 2 billion monthly users and AI Mode had passed 100 million monthly active users in the US and India alone. You do not optimise for a footnote at that scale. You optimise for it properly.

So this is our practical guide to optimising for Gemini search, written from the actual work, not from a press release. We have been tracking Gemini and AI Mode citations across our client accounts for months, and the short version is this: it is SEO, but with the difficulty turned up. Let us show you what that means.

What is Gemini search, exactly?

Gemini is the AI model. You meet it in three main places: the standalone Gemini app, AI Overviews (the summary box at the top of a normal Google search), and AI Mode (the full conversational search experience inside Google).

The thing to grasp is that Gemini is not a separate index sitting in a cupboard somewhere. It is plugged straight into Google's web index and ranking systems. When AI Mode answers you, it is pulling from the same crawl, the same quality signals, and broadly the same understanding of authority that powers blue links.

That single fact changes everything about how you approach it. If you already do good search engine optimisation, you are most of the way there. If you do not, no amount of "AI optimisation" trickery will save you.

How is Gemini different from ChatGPT and Perplexity?

People lump all the AI engines together, and they really should not. The plumbing underneath each one is different, and the plumbing decides who gets cited.

ChatGPT leans on its training data plus a search partner for fresh, web-connected answers. Getting picked up there is part "are you in the index it queries" and part "are you authoritative enough to have soaked into the model".

Perplexity behaves like a search engine wearing an AI hat. It crawls broadly, loves recent content, and will happily cite a page that is days old if it answers the question well.

Gemini uses Google's index. So Google's ranking signals, the ones you have been chasing for years, directly shape what Gemini surfaces. If Google rates your page, Gemini is far more likely to lean on it. If Google has buried you on page four, you are effectively invisible to Gemini too.

Our honest take: this is good news if you take SEO seriously, and terrible news if you have been propping up thin pages with backlinks. We dig into the wider picture in our guide to getting your brand into AI answers, but for Gemini specifically, the Google connection is the whole game.

The myth that needs killing first

There is a small industry built on telling you that AI search needs its own special toolkit. Magic markup. An llms.txt file. Chopping your content into tiny robot-friendly nuggets. Rewriting everything in some "AI-readable" voice.

In May 2026 Google put out a new resource on optimising for generative AI in Search, and it is refreshingly blunt about all of that. Their official line is that, from Google's perspective, "optimizing for generative AI search is optimizing for the search experience, and thus still SEO".

The same guide debunks the usual GEO snake oil directly. In Google's words, "You don't need to create new machine readable files, AI text files, markup, or Markdown to appear in generative AI search," and there is "no requirement to break your content into tiny pieces for AI to better understand it." Search Engine Journal summed it up neatly when they reported that Google's new guide calls AEO and GEO "still SEO".

We are not saying chunking and structure are pointless. Clean structure helps everyone. We are saying do not buy the idea that there is a secret Gemini lever that bypasses being genuinely good. There is not.

What we actually see Gemini cite

Theory is cheap. Here is what we keep finding when we track which client pages earn a place in AI Mode and AI Overviews, and which get politely ignored.

Pages that already rank well, but the overlap is shrinking

For a long time, ranking in the top 10 was almost a guarantee of getting cited. Not anymore. Ahrefs found that only about 37.9% of AI Overview citations now come from pages in the top 10 organic results, down from roughly 76% in July 2025, across an analysis of 863,000 keyword SERPs and 4 million AI Overview URLs.

Why the drop? Google's own documentation explains that AI features use a "query fan-out" technique, issuing multiple related searches across subtopics rather than just grabbing the pages that rank for your exact phrase. So a page sitting at position 14 for the head term can still get cited because it nails a sub-question the fan-out asked.

The lesson we have taken from this: ranking in the top 10 still helps enormously, but covering the full cluster of related questions around a topic is what really earns citations now. Thin pages that win one keyword are losing ground.

Content with specific data and a real point of view

Gemini loves a number it can attribute. "Studies show" gets nothing. "37.9% of AI Overview citations come from top 10 pages, per Ahrefs" gets cited because it is a clean, sourced, quotable fact.

It is the same with opinion and first-hand experience. Google's guidance is explicit: "Don't just recycle what others on the internet have already said, or could easily be produced by a generative AI model." If your page reads like an AI wrote it from the other AI summaries, Gemini will skip you and cite whoever had something original to say.

Google's own properties, a lot

Worth knowing before you get too excited. Google self-cites heavily. An SE Ranking study covered by Search Engine Land found Google.com made up 17.42% of all AI Mode citations, tripling from 5.7% in June 2025, with Google-controlled properties together accounting for roughly a fifth of all sources.

You cannot out-Google Google. What you can do is win the slices it does not own, and there are plenty of them.

What Gemini quietly refuses to cite

The flip side is just as useful. After months of watching, these are the pages that consistently get left out in the cold.

• Thin content held up by links alone. Pages that rank on authority but say very little get skipped. Gemini judges the substance, not just the backlink profile.
• Content locked behind heavy JavaScript or paywalls. If the main text only appears after rendering or a login, it often does not make it in. Google needs to read it cleanly.
• Stale pages. A solid 2024 article that nobody has touched since loses to a fresher take on the same question, even when the old one ranks higher organically.
• Paraphrased copies. If your page is a reworded version of the page already ranking, Gemini cites the original. We have watched this happen to clients who "studied the competition" a little too closely.

Our practical checklist for getting cited by Gemini

This is the routine we run for any page we want Gemini to favour. It is not glamorous. It works.

1. Earn the organic ranking first. Gemini reads Google's index, so traditional technical SEO, solid on-page optimisation, and genuine link building remain the foundation. There is no shortcut around being a page Google trusts.
2. Cover the whole question cluster. Because of query fan-out, you want one page (or a tight set of linked pages) that answers the main query and the obvious follow-ups. Read the top results, list every sub-question they answer, then answer all of them plus one they missed.
3. Lead with the answer. Put a direct, two-sentence answer right under each heading, then expand. Buried answers do not get extracted.
4. Add something only you can. A statistic, a tested process, a contrarian opinion, a real client outcome. Google's helpful content guidance asks whether your page provides "original information, reporting, research, or analysis". If the honest answer is no, that is your homework.
5. Keep it crawlable. Make sure your main content is in the HTML, not hidden behind a script or a login, and that your robots.txt is not blocking the pages you care about. We still find this on audits more often than we would like.
6. Refresh on a schedule. Revisit your priority pages at least quarterly. Update the data, prune the dead bits, add what has changed. A visible "last updated" date helps.
7. Build your entity, not just your pages. Gemini understands brands through Google's Knowledge Graph and entity signals. Keep your name, details and positioning consistent everywhere, earn mentions on credible sites, and sort your Wikidata and Wikipedia presence if it applies to you.

How do you even measure Gemini visibility?

Here is the awkward bit. There is no Gemini Search Console. Google does not hand you a tidy dashboard of where you appear in AI Mode, so tracking it takes a bit of graft and a bit of cunning.

This is the rough setup we use across accounts:

• Manual prompt tracking. We run a fixed list of target queries through Gemini and AI Mode on a schedule and log whether the client is cited, and against whom.
• GA4 referral checks. Watch for sessions from gemini.google.com and Google AI surfaces in your referral reports. The volume is modest, but the trend tells a story.
• AI citation monitoring tools. Several platforms now track AI Overview and AI Mode citations at scale. Pick one, but always sanity-check it by hand.
• Brand mention alerts. Mentions across the web feed Google's understanding of your entity, so we keep alerts running and chase the gaps.

None of it is as clean as the organic data you are used to. That is just where the discipline is right now. We expect proper reporting eventually, but we are not holding our breath.

Where this fits in the bigger AI search picture

Gemini is one engine in a crowded room. The smart move is not to obsess over it in isolation but to build content that earns trust across all of them, which is exactly what our AI search visibility work is built around. The tactics overlap heavily with getting cited in ChatGPT and AI Overviews, and with the broader shift we cover in our piece on why so many brands get GEO wrong in 2026.

The common thread is depth, originality and authority. Engines may weight those differently, but none of them reward thin, recycled content, and Gemini is arguably the strictest of the lot because it borrows Google's quality bar.

Our honest bottom line

After all the tracking, the takeaway is almost annoyingly simple. Optimising for Gemini search is not a new discipline. It is SEO with a higher quality threshold and a fan-out twist.

Everything Google has preached for a decade, helpful content, real authority, clean structure, freshness, original analysis, is exactly what Gemini rewards. The difference is that Gemini is far less forgiving of mediocrity. You cannot coast on domain authority. The page itself has to be better than the alternatives, and now it has to answer the questions around the question too.

For anyone doing the work properly, that is a gift. The bar went up, but the playbook stayed the same. If you would rather not run the experiments yourself, that is literally our job, so tell us what you are trying to rank for and we will show you where Gemini is currently looking instead.