How to Optimize for Google AI Overviews and Gemini: Being Indexed Is Not the Same as Being Selected

The whole game changed in one sentence: Google already has your content

Every other AI platform has to discover you. ChatGPT crawls a slice of the web through Bing and OpenAI's own crawlers. Perplexity runs live retrieval against its own index. Claude leans on a curated set. They each decide, query by query, whether your domain even enters the candidate pool.

Google does not have that problem with you. If you have any organic presence, you are already in the index that feeds AI Overviews. AIOs are not powered by a separate training corpus. They are grounded in Google's live search index through a custom Gemini model, which is precisely why they hallucinate less than standalone chatbots. The retrieval layer is the same Search infrastructure that has ranked you for years.

That sounds like an advantage. It is actually the hardest version of the problem. When the entire web is already a candidate, the differentiator is not discovery. It is selection. Gemini 3 became the default AI Overviews model on January 27, 2026, and each AIO now pulls 32% more sources per response than the prior model, per SE Ranking's analysis. More sources per answer does not mean more brands win. It means the bar for which specific pages get pulled into a synthesized answer moved, and it moved underneath everyone at once.

AIOs are grounded in Google's live search index. Discovery is solved. The contest is selection, and selection runs on different signals than ranking.

This is the core split that governs everything below. On Google, visibility is cheap and influence is scarce. Being seen is the baseline. Being selected, summarized, and cited is the asset. Most brands optimize for the first and wonder why the second never arrives.

How Google's AI surfaces actually choose what to cite

AIOs run a query fan-out. The model takes one query, expands it into a cluster of related sub-queries, retrieves documents across each, and assembles a consolidated answer that cross-verifies across multiple domains. A featured snippet extracts from one page. An AIO synthesizes across several. That difference is the whole reason single-keyword optimization no longer works on this surface.

The candidate pool starts from Google's core ranking systems: RankBrain, BERT, PageRank, the Helpful Content and Reviews systems, passage ranking. Organic rank is the entry ticket. But rank alone does not get you cited, and the gap between the two has widened sharply.

• 97% of AI Overviews cite at least one source from the top 20 organic results for that query (seoClarity, 432,000 keywords). Ranking is still the most reliable on-ramp.
• And yet only 38% of cited pages also appear in the top 10 organic results, down from 76% in the prior study (Ahrefs). The rest split almost evenly: 31.2% from positions 11 to 100, and 31.0% from beyond position 100.
• Position 1 pages appear in AIOs more than 50% of the time (seoClarity), but a page in position 6 can earn more AIO citations than the page in position 1. Citation position inside the AIO, not organic rank, is now the visibility metric that matters.

Read those three numbers together. Ranking improves your odds. It does not secure your selection. Nearly a third of everything AIOs cite ranks nowhere near the top of the page it appears on. Google is deliberately reaching past rank to assemble a diverse, cross-verified answer.

After Gemini 3, the churn intensified. Gemini 3 replaced roughly 42% of previously cited domains in AI Overviews (SE Ranking), and 88% of AIOs now cite three or more sources while only 1% cite a single source (SE Ranking). The single-source AIO is effectively dead. If your strategy was to win one snippet-style answer box, that strategy expired on January 27.

The selection signals that survive the churn cluster into five buckets:

1. E-E-A-T and existing trust. Experience, expertise, authoritativeness, trust. LLMs read author bios for real credentials, not third-party domain scores. There is no "domain authority" input.
2. Semantic relevance, not keyword matching. Gemini can cite a page that contains none of the query's keywords if the page is semantically on-point and from a reliable domain.
3. Concise, structured formatting. Direct answers near the top, headings, lists, tables, FAQ blocks, clean semantic HTML.
4. Freshness. AIOs favor recently updated, reliable information. Stale pages get replaced, which is exactly what the Gemini 3 domain churn demonstrated.
5. Brand citations across the web, including unlinked mentions in authoritative, niche-relevant content. Quality of the context beats quantity of the links.

The signal almost nobody budgets for: YouTube

Here is where Google diverges from every other AI platform, and where most GEO programs leave the most value on the table. Google's AI surfaces lean disproportionately on Google's own ecosystem, and the single biggest beneficiary is YouTube.

• YouTube's coverage across tracked AIO keywords runs nearly 3x any other non-brand domain (BrightEdge), and YouTube citations inside AIOs are up roughly 25% since early 2024.
• YouTube enjoys a 200-fold advantage over any competing video source in AI citations broadly (TechEdge AI / 5W Citation Source Index 2026). There is no second place in video.
• In January 2026, YouTube officially overtook Reddit as the most-cited social platform across AI-generated answers (PikaSEO, GEORaiser). On Google specifically, YouTube and Reddit run near parity, while Perplexity still cites Reddit 6.1x more than YouTube.
• Google AI Overviews drives 36.6% of all YouTube citations observed across platforms, making it one of the two dominant engines pulling video into answers (OtterlyAI, 100M+ citations).

The execution detail that separates winners from spectators: format and timestamps. 94% of YouTube AI citations go to long-form video, not Shorts (OtterlyAI). And when a YouTube citation includes a specific timestamp, 73% of the time it surfaces in Google AI Overviews (OtterlyAI). Google can reach into a clearly chaptered, timestamped video and lift the exact 40-second segment that answers a sub-query in its fan-out.

On Google, a chaptered, timestamped, long-form YouTube video is not a marketing asset. It is a citation surface. Most brands fund text and ignore the one format Google cites most.

If your GEO budget has zero allocation to YouTube, you have a structural blind spot on the platform with 2.5 billion AI users. This is the clearest "being seen vs. being selected" gap in the entire study. A brand with strong organic rankings and no video can be perfectly visible to Google and still never get selected into the part of the answer Google trusts video to fill.

AI Overviews vs. AI Mode vs. the Gemini app: three surfaces, three rule sets

These are not interchangeable. They reach similar conclusions and cite almost entirely different sources, which means winning one does not win the others.

AI Overviews are the summaries injected above organic results. They are the default, mass-reach surface: 2.5 billion monthly users as of Google I/O, May 2026. They cite 7.7 domains per query on average, run a ~11% no-citation rate (answers with no external sources), and reference about 1.3 entities per response (Ahrefs, 730K responses).

AI Mode is the conversational, follow-up surface inside Search. It cites 9.2 domains per query, produces responses roughly 4x longer, references 3.3 entities, and has only a 3% no-citation rate (Ahrefs). For a brand that qualifies, AI Mode is the more reliable citation surface because it almost always cites something, and it leans harder on encyclopedic sources: Wikipedia appears in 28.9% of AI Mode responses vs. 18.1% of AIOs (Ahrefs).

The critical finding for anyone allocating effort: AI Mode and AI Overviews reach 86% semantic similarity but share only 13.7% of their citations, and 77% of unique cited domains appear on just one surface, not both (Ahrefs). Appearing in an AI Overview does not put you in AI Mode. They are two separate selection contests that happen to agree on the answer.

The standalone Gemini app is a third channel entirely. It is a conversational assistant where buyers research vendors and make decisions inside the chat, grounded through Grounding with Google Search and now Google Maps. Gemini 3's grounding metadata even added media_id for visual citations and page_numbers to pinpoint where information was found. The Gemini app behaves more like a true chatbot than a SERP feature, which means it pulls more brand-level recommendations than the deliberately commercial-shy AIO.

One more thing changed the map. On May 19, 2026, Liz Reid announced at Google I/O that AI Overviews and AI Mode are merging into "one seamless AI Search experience." The two surfaces will converge. The two citation pools, today only 13.7% overlapping, are the integration risk and the opportunity. Brands that qualify for both today are positioned for whatever the merged surface inherits.

Why Google barely mentions brands, and what that means for commercial intent

If you sell something, internalize this before you build a Google GEO plan: AIOs are an educational layer, not a commercial one, by design.

BrightEdge's ecommerce data is blunt. Google AI Overviews include a brand in only about 6.2% of ecommerce responses, averaging 0.29 brands per answer. Compare that to ChatGPT at 99.3% of responses and 5.84 brands, Perplexity at 85.7%, and Google's own AI Mode at 81.7%. On the brand-mention axis, AIOs are the minimalist of the entire field. Google routes commercial intent to organic results and ads, and reserves the AI summary for explanation.

This is a strategy choice, not a bug. 62% of ecommerce AIO citations come from sources that rank nowhere in the top 100 organic results (BrightEdge). Google is pulling explanatory, comparison, and how-it-works content into the answer, then handing the actual transaction to the links and ads below. Google Ads now appear on roughly 25% of AIO SERPs, up from under 1% in March 2025.

The contrast with ChatGPT is the whole point. ChatGPT and Google disagree on brand recommendations 62% of the time (BrightEdge), and Google AI Overviews are 44% more likely to criticize a brand than ChatGPT (BrightEdge). Google is cautious, hedged, and educational. ChatGPT is confident and recommendation-first.

Do not optimize Google AIOs to recommend your product. Optimize them to explain the category and cite you as the authority who explained it. The recommendation happens elsewhere on Google.

What does not work on Google, then, is the playbook that works on ChatGPT: product-led, "best X for Y" listicle bait engineered to win a recommendation. AIOs will mostly refuse to recommend at all. What works is owning the educational layer. Be the page Google cites when it explains how the category works, then let your organic listing, your ads, and the standalone Gemini app capture the intent the AIO deliberately won't.

What this does to clicks: the zero-click squeeze is real, but it is not uniform

The headline number is brutal and well-sourced. The most rigorous controlled study, Pew Research Center's analysis of 68,000 queries, found users are 47% less likely to click a traditional result when an AI Overview appears, and only 8% of visits that include an AIO end in any click. Other large datasets agree on direction: Seer Interactive measured a 61% CTR drop on AIO queries across 2.43 billion impressions, and Ahrefs estimated a 58% lower average CTR for the top-ranking page when an AIO is present.

That is the cost of zero-click. But three findings keep this from being a simple death spiral, and they are where the strategy lives:

• Citation is now a paid-tier signal. Brands cited inside an AI Overview earn 35% more organic clicks and 91% more paid clicks than non-cited brands on the same SERP (BrightEdge). The click pie shrank, but being cited buys you a far larger slice of what remains.
• There are early signs of a floor and a rebound. Organic CTR on AIO queries fell to roughly 1.3% in December 2025 and recovered to 2.4% by February 2026 (BrightEdge). Google is actively tuning, and AIOs are now abridged by default with a "Show More" expansion that pushes organic results back into view.
• The traffic that survives converts. Practitioner data consistently shows AIO click-throughs arrive with higher intent and longer time on site. Fewer clicks, higher quality, which is exactly the conversion economics that defines the AI-search era.

The strategic read: stop measuring success in raw organic clicks on AIO queries. That number is going to keep falling and there is no playbook that reverses it. Measure citation share. Being cited is the new ranking. On Google more than anywhere, the mention is the signal, and the click is a downstream bonus.

The execution playbook

Eight moves, ordered by leverage on Google's specific selection mechanics.

1. Earn the top 20, then stop expecting rank to do the rest. 97% of AIOs cite from the top 20 organic results, so ranking is the entry ticket. But with only 38% of citations coming from the top 10, treat strong rankings as necessary, not sufficient. Do not over-invest in moving position 6 to position 3 when the citation contest is decided by other signals.

1. Build topical clusters, not keyword pages. AIOs fan a query out across sub-questions. A seed page interlinked with a cluster of supporting pages gives Gemini complete coverage to assemble its answer from. Comprehensive beats narrow.

1. Fund YouTube as a citation surface. Publish long-form (not Shorts), with clean chapters and timestamps. Timestamped video surfaces in AIOs 73% of the time. This is the single highest-leverage move unique to Google and the one most programs skip.

1. Front-load direct answers. Put a concise, declarative answer near the top of every page, then expand. Use headings, lists, tables, and FAQ blocks. Gemini lifts the clean summary, not the buried one.

1. Own the educational layer, not the recommendation. With brands in only 6.2% of ecommerce AIOs, write to be cited as the authority that explains the category. Route commercial intent to organic, ads, and the standalone Gemini app, which mention brands far more.

1. Make E-E-A-T machine-readable. Real author bios with verifiable credentials. Match your structured data to visible content. JSON-LD for author, FAQPage, article, review, and sameAs. No special schema is required, but well-implemented schema raises the probability of correct interpretation; poorly implemented schema demonstrably lowers AI visibility.

1. Keep content fresh on a schedule. Gemini 3 replaced 42% of previously cited domains. Freshness is not cosmetic. It is how you survive model updates that re-pull the citation pool.

1. Qualify for AI Mode separately, and watch the merge. With only 13.7% citation overlap, appearing in AIOs does not earn AI Mode. AI Mode rewards longer, more entity-rich, encyclopedic content and almost always cites something (3% no-citation rate). With the two surfaces merging into one experience announced at I/O 2026, qualifying for both now is forward protection.

One thing to skip: do not block Google-Extended thinking it removes you from AI Overviews. Google-Extended only governs AI training use; blocking it has zero effect on AIO inclusion or Search rankings. And there is currently no way to opt out of AIOs while staying in regular Search, a gap now under UK CMA scrutiny. You are in this surface whether you optimize for it or not. The only choice is whether you get selected.

Every statistic in this piece follows Visigent's research standards: named sources, dated figures, corrections on request to help@visigent.com.