Google is the awkward engine in any GEO conversation, because the honest advice is the least exciting. There is no separate AI index to court, no special file to publish, no schema trick to deploy. AI Overviews, AI Mode, and Gemini all read from Google's ordinary web index, which means the work that earns a citation is the work that earns an organic ranking. Google said as much, in writing, in its May 2026 AI-optimization guide. This is the per-engine playbook for Google's AI surfaces, sitting under the broader pillar on generative engine optimization.
Key takeaways
- What Google rewards — The same things classic SEO rewards: organic ranking, genuine E-E-A-T, entity authority, and a direct on-page answer. Gemini rewrites from pages that already rank, so ranking is the prerequisite, not a bonus.
- What it explicitly ignores — llms.txt, content chunking, and AI-specific structured data. Google's May 2026 guide names all three as non-levers. Schema still earns rich results, but it is not an AI Overviews eligibility signal.
- The crawlers — Googlebot crawls for the index that AI Overviews draws from. Google-Extended is a separate, optional opt-out for Gemini model training — blocking it does not remove you from AI Overviews.
- How long it takes — As long as classic ranking takes, which is the honest and unwelcome answer. There is no AI-specific shortcut; the levers are the slow, durable ones.
How Google AI Overviews decides what to cite
The defining fact about Google's AI surfaces is that they are not bolted onto a new pipeline. They sit on top of the same index, the same crawler, and broadly the same ranking that decides organic results.
- Crawl and index the web with Googlebot, exactly as for organic search.
- Rank candidate pages for the query using Google's organic systems, including the E-E-A-T and entity signals.
- Trigger an AI Overview when the query is one Google judges suited to a synthesised answer.
- Select passages from the highly-ranked candidate pages.
- Synthesise the overview with Gemini and attach citation links to the sources used.
Two implications follow, and they are the whole game. First, ranking is upstream of everything: a page that is not in the candidate pool for a query is invisible to the AI Overview for that query, so classic SEO is not a parallel track, it is the track. Second, because Google has publicly disavowed the AI-specific levers, the optimisation effort that pays off is the same effort that has always paid off — which is good news for sites with durable authority and bad news for anyone hoping a formatting trick will leapfrog them.
The seven-step playbook
Tactics ordered by leverage, calibrated for Google. The contrast with the ChatGPT and Perplexity playbooks is stark: several moves that matter elsewhere are explicitly named as non-factors here, and the list is dominated by classic fundamentals.
- Win the organic ranking first. This is not a tactic so much as the precondition for all the others. AI Overviews draw from pages Google already ranks, so a page sitting on result-page three is not a candidate. If a query matters for AI visibility, it has to be a query you compete on organically, with the keyword research, internal linking, and content depth that implies.
- Build genuine E-E-A-T and entity authority. Google reports that the overwhelming majority of AI Overview citations come from sources it already considers authoritative. Named authors with real credentials and bio pages, an organisation that is a recognised entity in its field, citations from and to trusted sources — these compound slowly and are the single largest durable lever for Google's AI surfaces.
- Answer the query directly, near the top. Gemini extracts passages, and a concise, self-contained answer to the likely question is far easier to lift than the same point buried three scrolls down. This overlaps with the featured-snippet discipline Google rewarded long before AI Overviews existed; the muscle transfers directly.
- Refresh priority pages quarterly. Freshness is a real but moderate signal here. A meaningful share of cited content is recent, and pages left static for several quarters fade as fresher competitors appear. Put your top pages on a quarterly review — update statistics, add new developments, revise recommendations — without the every-fortnight intensity Perplexity demands.
- Keep your structured data, but for the right reason. Maintain Article, FAQPage, Organization, and Person markup. It earns traditional rich results and helps Google resolve your entities. Just do not treat it as an AI Overviews eligibility lever, because Google's guide says explicitly that there is no special schema to add for AI features. It is an entity and rich-result investment, not a GEO one.
- Allow Googlebot; decide on Google-Extended deliberately. Blocking Googlebot removes you from the index and therefore from AI Overviews. Google-Extended is a separate training opt-out for Gemini that does not affect AI Overviews eligibility, so the choice to block it is a content-licensing decision, not an SEO one. Confirm Googlebot is reaching your pages with a live fetch test, not just by reading robots.txt.
- Skip the GEO tricks Google has ruled out. Do not build an llms.txt for Google, do not artificially chunk multi-topic pages, do not invent AI-specific markup. Google's May 2026 guide names all three as things its systems do not use. They may help on other engines — llms.txt in particular is read by Claude and the agentic crawlers — but on Google they are effort with no return. Spend that time on steps one and two.
What's different from ChatGPT, Perplexity, and the agentic engines
Most generative engines reward a similar shape of page, but Google is the outlier that rewards the classic shape hardest and the AI-specific shape least. CTAIO Labs rewrote one article under three optimisation frameworks and measured the per-engine deltas, including Google, in the framework test.
- ChatGPT runs on a Bing-backed index and rewards brand authority plus structured prose, and it reads llms.txt. The brand-authority overlap with Google is real; the llms.txt divergence is not. The ChatGPT playbook is at how to rank in ChatGPT.
- Perplexity prizes freshness and quotability on its own index and rewards an aggressive refresh cadence. Google rewards freshness too, but moderately, and never at the expense of authority. The Perplexity playbook is at how to rank in Perplexity.
- Claude retrieves through Brave Search and does read llms.txt, so a file that does nothing for Google can still earn citations there. This is the cleanest example of why per-engine calibration beats a single universal checklist.
- The agentic crawlers (the ones that fetch llms.txt and AI-specific signals) are exactly the audience Google's guide tells you not to optimise for on Google. Build those signals for the engines that use them, not for Google.
Measurement
Google is the one AI surface where your existing analytics already carry most of the signal, which is a quiet advantage. Build the loop in three layers:
- Google Search Console. AI Overviews impressions and clicks are folded into Search performance. The tell-tale pattern is a query with rising impressions and flat or falling clicks — often a sign an AI Overview is answering on your behalf using your content. Track your priority queries for that divergence.
- Citation tracker. Profound, Peec AI, AthenaHQ, Otterly, or one of the others now report AI Overviews presence alongside ChatGPT, Perplexity, and Gemini. The Radar's scored shortlist is at 6 GEO Tools the Radar Actually Recommends; CTAIO Labs tested ten head-to-head in the visibility tools test.
- GA4 channel grouping. Add
gemini.google.comas a referral source and watch the trend. Volume is lower than ChatGPT's referral, but it confirms when Gemini's standalone surface, as opposed to AI Overviews in Search, is sending users.
Field evidence
Related reads
Frequently asked questions
How does Google AI Overviews decide what to cite?
AI Overviews, AI Mode, and Gemini-in-Search all draw from Google's existing web index rather than a separate AI corpus. For a given query, Google assembles a candidate set from already-ranked results, and Gemini synthesises an answer from passages in those pages, attaching links to the sources it draws from. The practical consequence is that organic ranking is the entry gate: a page that does not appear in the candidate pool cannot be cited, no matter how it is formatted.
Does llms.txt help with Google AI Overviews?
No. Google's May 2026 AI-optimization guide states plainly that its systems do not reference an llms.txt file and that publishers do not need one for AI features. This is the clearest divergence from ChatGPT and the agentic engines, several of which do read llms.txt. On Google, time spent on an llms.txt is time not spent on the levers that actually work.
Should I chunk my content or add AI-specific schema for Google?
Google's guide rules out both as eligibility levers. It says its systems understand multi-topic pages without artificial chunking, and that there is no special schema markup to add for AI features. Structured data still earns traditional rich results and helps Google understand entities, so keep your Article, FAQPage, and Organization markup — but do not expect it to function as an AI Overviews ranking signal, because Google says it does not.
What is the difference between blocking Googlebot and blocking Google-Extended?
Googlebot crawls the web for Google's search index, which is the index AI Overviews draws from. Blocking Googlebot removes you from organic results and therefore from AI Overviews. Google-Extended is a separate token that controls whether your content is used to improve Gemini models; blocking it is a training opt-out and does not remove you from AI Overviews or organic search. Most publishers allow Googlebot and make a deliberate choice on Google-Extended.
How is ranking in AI Overviews different from ranking in ChatGPT or Perplexity?
It is closer to classic SEO than either. ChatGPT rewards brand authority and structured prose on a Bing-backed stack; Perplexity prizes freshness and quotability on its own index. Google's AI surfaces run on Google's organic index, so the fundamentals that win organic ranking are the same ones that win AI Overviews citations, and the GEO-specific tactics that help elsewhere are explicitly down-weighted here. CTAIO Labs measured the per-engine gap directly in the framework test at /en/labs/agentic-search/framework-test/.
Does freshness matter for AI Overviews?
Yes, though less dramatically than for Perplexity. A meaningful share of AI Overview citations point to recently-updated content, and pages left untouched for several quarters tend to fade from citations as fresher competitors appear. A quarterly review of your priority pages — updating statistics, adding new developments, revising recommendations — keeps them in contention without the aggressive cadence Perplexity demands.
What is Google AI Mode and does it change the playbook?
AI Mode is Google's more conversational, multi-step search experience, powered by the same Gemini models and the same index as AI Overviews. It fans a query out into several sub-queries and synthesises across more sources, which rewards comprehensive topical coverage, but it does not introduce a separate optimisation surface. The playbook is the same: rank well, demonstrate E-E-A-T, answer directly, and stay fresh.
What tools measure my AI Overviews visibility?
Most LLM-visibility trackers — Profound, Peec AI, AthenaHQ, Otterly, and others — now report AI Overviews presence alongside ChatGPT and Perplexity. Google Search Console remains the foundation: AI Overviews impressions and clicks are folded into Search performance, and a query showing impressions without proportionate clicks is often one where an AI Overview is answering on your behalf. The Radar's scored shortlist of trackers is at /en/radar/geo-tools/.
How does Google AI Overviews decide what to cite?
AI Overviews, AI Mode, and Gemini-in-Search all draw from Google's existing web index rather than a separate AI corpus. For a given query, Google assembles a candidate set from already-ranked results, and Gemini synthesises an answer from passages in those pages, attaching links to the sources it draws from. The practical consequence is that organic ranking is the entry gate: a page that does not appear in the candidate pool cannot be cited, no matter how it is formatted.
Does llms.txt help with Google AI Overviews?
No. Google's May 2026 AI-optimization guide states plainly that its systems do not reference an llms.txt file and that publishers do not need one for AI features. This is the clearest divergence from ChatGPT and the agentic engines, several of which do read llms.txt. On Google, time spent on an llms.txt is time not spent on the levers that actually work.
Should I chunk my content or add AI-specific schema for Google?
Google's guide rules out both as eligibility levers. It says its systems understand multi-topic pages without artificial chunking, and that there is no special schema markup to add for AI features. Structured data still earns traditional rich results and helps Google understand entities, so keep your Article, FAQPage, and Organization markup — but do not expect it to function as an AI Overviews ranking signal, because Google says it does not.
What is the difference between blocking Googlebot and blocking Google-Extended?
Googlebot crawls the web for Google's search index, which is the index AI Overviews draws from. Blocking Googlebot removes you from organic results and therefore from AI Overviews. Google-Extended is a separate token that controls whether your content is used to improve Gemini models; blocking it is a training opt-out and does not remove you from AI Overviews or organic search. Most publishers allow Googlebot and make a deliberate choice on Google-Extended.
How is ranking in AI Overviews different from ranking in ChatGPT or Perplexity?
It is closer to classic SEO than either. ChatGPT rewards brand authority and structured prose on a Bing-backed stack; Perplexity prizes freshness and quotability on its own index. Google's AI surfaces run on Google's organic index, so the fundamentals that win organic ranking are the same ones that win AI Overviews citations, and the GEO-specific tactics that help elsewhere are explicitly down-weighted here. CTAIO Labs measured the per-engine gap directly in the framework test at /en/labs/agentic-search/framework-test/.
Does freshness matter for AI Overviews?
Yes, though less dramatically than for Perplexity. A meaningful share of AI Overview citations point to recently-updated content, and pages left untouched for several quarters tend to fade from citations as fresher competitors appear. A quarterly review of your priority pages — updating statistics, adding new developments, revising recommendations — keeps them in contention without the aggressive cadence Perplexity demands.
What is Google AI Mode and does it change the playbook?
AI Mode is Google's more conversational, multi-step search experience, powered by the same Gemini models and the same index as AI Overviews. It fans a query out into several sub-queries and synthesises across more sources, which rewards comprehensive topical coverage, but it does not introduce a separate optimisation surface. The playbook is the same: rank well, demonstrate E-E-A-T, answer directly, and stay fresh.
What tools measure my AI Overviews visibility?
Most LLM-visibility trackers — Profound, Peec AI, AthenaHQ, Otterly, and others — now report AI Overviews presence alongside ChatGPT and Perplexity. Google Search Console remains the foundation: AI Overviews impressions and clicks are folded into Search performance, and a query showing impressions without proportionate clicks is often one where an AI Overview is answering on your behalf. The Radar's scored shortlist of trackers is at /en/radar/geo-tools/.
Ready to Find the Right AI Tools?
Browse our data-driven rankings to find the best AI tools for your team.