Google I/O 2026: AI Overviews Hit 2.5 Billion Users, AI Mode 1 Billion — What Changes for GEO

Updated September 25, 2026: After new 2026 research and Google product changes, we corrected the AI opt-out section (the Search Console control has been available to all websites worldwide since August 31, 2026), dated the 4.4× conversion figure to 2025, updated which engines to check monthly, and added Google's July 2026 guidance on chunking and schema, plus what 2026 research says about relevance and ranking. Why the GEO evidence changed.

Google I/O 2026 did not just ship product updates — it redrew the rules for how businesses get cited in search. Gemini 3.5 Flash is now the global default behind AI Overviews and AI Mode. The two surfaces have merged into a single AI search experience that has passed 1 billion monthly users. And according to BrightEdge data from February 2026, only about 17% of AI Overview citations come from pages in the organic top 10 — down from roughly 75% in early 2024.

That last number is the one that changes the strategy. Here is what happened and what to do about it.

What changed at Google I/O 2026

Google's annual developer conference in May 2026 announced several changes that directly affect how search surfaces content.

Gemini 3.5 Flash is the new default. The model powering AI Overviews and AI Mode was upgraded to Gemini 3.5 Flash, replacing Gemini 3 Flash, the prior default that had launched in March 2026. Gemini 3.5 Flash brings improved multi-step reasoning and faster response latency — practically, this means AI answers that are more elaborate, more multi-sourced, and generated faster.

AI Mode and AI Overviews are now one experience. Google had been running two overlapping AI answer products: AI Overviews (the box that appears above organic results for many queries) and AI Mode (a dedicated AI-search tab). At I/O 2026, Google announced these are merging into a unified AI search experience. For citation purposes, you are now optimizing for one surface, not two.

Scale: 1 billion monthly users on AI Mode, 2.5 billion on AI Overviews. Google confirmed AI Mode passed 1 billion monthly users at I/O 2026. AI Overviews reach roughly 2.5 billion monthly users globally. These are not experimental features — they are the primary search interface for a large share of queries. A business that is invisible to AI citation is increasingly invisible to a large portion of its potential customers.

The top-10 overlap collapse — why ranking #1 no longer guarantees citation

This is the most operationally significant data point of the last six months.

In early 2024, roughly 75% of AI Overview citations came from pages ranking in the top 10 for that query (Semrush, 2024). By October 2025, BrightEdge had tracked that overlap down to ~54%. By February 2026, the same research put it at approximately 17% — meaning more than 80% of AI citations now come from pages outside the top 10.

Other research estimates range from 17% to 38% depending on query type and methodology — treat the BrightEdge figure as the most conservative benchmark.

Why is this happening? AI engines run their own background searches on narrower, related questions (query fan-out) and choose sources from what those searches return. A 2026 review of 45 GEO studies (preprint) found topical relevance to be one of the most reliable levers, so a page that answers one specific question well can beat a page with more overall authority. Pages ranked #1 are almost always optimized for traditional ranking signals: backlinks, keyword coverage, internal linking depth. Pages ranked 25th are often niche guides written to answer one specific question completely — exactly what AI engines need.

A business that writes answer-first content — direct answer first, structured sections, FAQ at the bottom — can get cited at scale while a larger competitor with better overall authority stays invisible in AI answers. The leverage point has shifted from "who has the most links" to "who wrote the clearest answer to this specific question."

The new AI opt-out toggle — what it is and who it's for

On June 3, 2026, Google launched a Search Console toggle that allows publishers to exclude their content from AI Overviews and AI Mode without affecting their organic snippet placement. The first rollout was to a subset of UK publishers only, under an order from the UK Competition and Markets Authority (CMA), which has been reviewing Google's AI search practices under UK competition law. Google's help page does not mention the Gemini app.

What to read into this: It started as a regulatory concession in one jurisdiction, but Google has since made it a global setting: since August 31, 2026, every website can choose Include, Exclude or Inherit under "Search generative AI" in Search Console settings. Excluding also removes your site from the AI features in Discover and takes effect within 1-2 days; it does not stop AI training, which Google-Extended controls separately.

Should you opt out? For most small businesses, no. Being cited in AI answers is still a net positive — AI-referred visitors converted at roughly 4.4× the rate of organic search visitors (Semrush, 2025). The businesses that might legitimately consider opting out are publishers whose business model depends on driving large volumes of traffic from informational queries — news sites, recipe sites, large how-to publishers — where AI summaries may reduce click-through without compensating benefit. That describes almost no small local businesses or service providers.

Check the setting in Search Console and keep it on Include; if a property is set to Inherit, check the parent property. Page-level controls are due by March 2027, so follow the Google Search Central Blog for changes. Do not opt out speculatively.

What a small business should do now

The practical translation of all of the above comes down to one reframe: stop optimizing pages only to rank and start optimizing passages to be excerpted.

Most small business content was written to satisfy Google's traditional ranking signals — comprehensive coverage, internal links, keyword density in headings. That content may rank well but fails at the passage-level test: does any 200-400 word section of this page answer a single question completely enough for an AI to excerpt it?

One caveat from Google itself: its AI optimization guide (updated July 10, 2026) says there is no requirement to chunk content for its AI features and no special schema to add. And a 2026 review of 45 GEO studies (preprint, not yet peer-reviewed) found that topical relevance and a page's position in what the engine retrieves are the levers that reproduce most reliably, so being indexed and ranking for the questions your customers ask still comes first. Treat answer-first structure as a way to make pages clearer for readers and for other AI engines, not as a Google-specific switch.

The four moves that matter most right now:

1. Audit for passage readiness before building anything new. Pull your top 20 pages by impressions from Search Console. For each section, ask: if an AI model quoted only this paragraph, would it be a complete and useful answer? If not, that page goes on your editing list.

2. Restructure, do not rewrite. Most of the information your pages need is already there. The work is reorganizing: answer in the first sentence, supporting evidence next, specifics last. Add FAQ blocks. Convert headings to questions. This is editing, not content production.

3. Build the entity layer. AI engines cross-reference entities — your business name, your category, your author's credentials — before deciding whether to cite you. Keep your Google Business Profile complete (Google's AI optimization guide recommends it) and your name, category, and author details consistent everywhere. Schema markup still helps with clear entity facts and eligibility for Google's rich results, but Google says its AI features need no special schema, and an Ahrefs test (May 2026) found no significant citation lift from adding it.

4. Measure monthly with real queries. Share of Model Voice is the metric that matters. Run 10-15 realistic customer queries across Google AI Mode, ChatGPT Search, and Gemini once a month (Perplexity is optional: it had 0.9% of generative-AI web traffic in August 2026, per Similarweb). For Google, add Search Console's generative AI performance report (available to all sites since August 31, 2026): it shows your AI Overviews and AI Mode impressions by page, though not clicks or queries, and low-volume sites may not see it. Tracking these numbers over time is how you know whether any of the above changes are working.

The Google I/O 2026 announcements confirm a direction that has been visible in the data for over a year: AI search citation and traditional ranking are decoupling. The top-10 overlap collapse is not a bug being patched — it is the product working as designed. Businesses that adapt the structure of their content now are building an advantage that compounds. Those that wait for rankings to recover will find the gap harder to close.