Google AI Overviews vs. AI Mode vs. Gemini: The Differences That Actually Matter for Your Visibility

I’m so confused by how Google’s different AI tools and modes work, I’m sure you are too. Here’s what my research found.

TLDR

These are my current notes on the three Google AI products people keep confusing, as of May 2026, right after Google I/O made the picture both clearer and more complicated.

• AI Overviews = the AI summary above your blue links. 2.5 billion monthly users. The one that’s gutting CTR for informational queries.
• AI Mode = the dedicated conversational search tab. 1 billion monthly users in 12 months. Built on Gemini 3.5 Flash. Queries are 3x longer than classic search.
• Gemini = the standalone AI app at gemini.google.com. 900 million users. Not really a search product. The “brain” that powers the other two.
• They share infrastructure but they’re three different optimization problems.
• May 19, 2026: Google announced AI Overviews and AI Mode are being merged into “one seamless AI Search experience.” The boundary blurs over the next few quarters. The optimization work doesn’t.
• If you’re an SEO or a business owner, AI Overviews is the surface that’s costing you traffic today. AI Mode is the surface that’s going to define visibility tomorrow. Gemini is interesting but mostly orthogonal to your search strategy.
• Being cited inside an AI Overview earns 35% more organic clicks and 91% more paid clicks than not being cited. That’s the entire ballgame.

The longform version is below.

The headline shift: three products, three problems

I’ve watched SEO threads on LinkedIn all year where people say “AI Overviews and AI Mode and Gemini” in the same breath like they’re the same thing. They’re not. And conflating them is the fastest way to misallocate budget.

Three different surfaces. Three different user behaviors. Three different ways of earning visibility. They share an underlying model called Gemini, but the way they’re stitched into Google Search is completely different. That’s what matters for anyone trying to get found.

Here’s the cheat sheet I keep open while doing audits for clients. For the full strategic frame on optimizing across these surfaces, we walk through it in our AEO best practices guide.

Quick comparison

That’s the orientation. Now the details that change how you operate.

Google AI Overviews: the surface that’s eating your clicks

AI Overviews are the AI-generated answer boxes that sit on top of the regular Google search results page. They launched broadly in the US in May 2024 and now reach more than 200 countries and 40+ languages.

Sundar Pichai said on the May 2026 I/O stage that AI Overviews crossed 2.5 billion monthly active users. That’s not a typo. To put it in context, ChatGPT, the product everyone breathlessly talks about, reportedly hit 900 million weekly users in February 2026. AI Overviews are an order of magnitude bigger.

The way they work under the hood: Google uses a technique called query fan-out, where it breaks your single search into multiple sub-queries, retrieves and ranks sources for each, then synthesizes the results into a short summary with citation chips. You get the answer. Sometimes you click. Often you don’t.

The CTR data, in one place

Every serious study on this has reached the same conclusion. The magnitude varies. The direction is unanimous.

• Seer Interactive, September 2025 update: Across 3,119 informational queries, 25.1M organic impressions, and 1.1M paid impressions, organic CTR dropped 61% (1.76% to 0.61%) when AI Overviews appeared. Paid CTR dropped 68% (19.7% to 6.34%).
• Ahrefs, December 2025: Analysis of 300,000 keywords showed a 58% reduction in CTR for the top-ranking result on AIO-containing queries. The earlier April 2025 study showed a 34.5% drop. It’s getting worse.
• Arc Intermedia case study: Of 1,000 commercial search terms analyzed, 86.8% triggered an AI Overview.
• SISTRIX, March 2026: Position one organic CTR on AIO queries fell from 27% to as low as 11%.

The number that matters most is the one from the same Seer study: brands cited in AI Overviews earn 35% more organic clicks and 91% more paid clicks compared to brands not cited on the same queries. Citation has become the new ranking.

We dug into this directly in our own AEO page analysis, where we looked at 5,000 URLs in the fintech space to figure out which pages get cited by ChatGPT and why. The short version: fundamental SEO signals (Ahrefs Domain Rating, keyword coverage, backlink profile) are still the strongest predictors of AI citation. That overlaps a lot with how AI Overviews behaves on Google’s side.

I’ve watched this play out on client sites. moveBuddha had a 643% lift in referring domains over our retainer, and we earn AI Overview citations on a meaningful slice of high-intent moving queries because the content is structured for extraction, not just for blue-link ranking. Upwind got cited consistently in cybersecurity definition queries before being acquired, which is part of why we built a productized cybersecurity marketing and content offer around that pattern. Fingercheck shows up on payroll software comparisons that send 13K monthly visits and represent $29K in monthly traffic value, per Ahrefs.

The CTR math sounds catastrophic until you realize the upside of being cited. Which is why the strategy isn’t to fight AIO. It’s to earn the citation.

Where AI Overviews are weak

Static. No follow-up. Designed for high-confidence informational queries where a clear consensus exists across multiple authoritative sources. Triggers more on “how to” and “best of” queries, less on transactional, less on highly local.

If you’ve been chasing position-one CTR for “how to clean suede” or “best budget laptops 2026,” AIO has already absorbed most of that demand.

Google AI Mode: where the next search behavior gets shaped

AI Mode is a separate tab inside Google Search, accessed via the “AI Mode” toggle near the search bar. Tap it and the experience changes. The blue links disappear. You get a full synthesized answer with inline citations and the ability to ask follow-up questions that maintain context across the session.

It’s Google’s bet on what search becomes when the constraint of “10 blue links” stops mattering.

Launched as a Search Labs experiment in March 2025. Opened to all US users without a waitlist in May 2025. Expanded internationally in October 2025 to 40+ countries and 35+ languages. As of May 2026 I/O, AI Mode crossed 1 billion monthly users, with queries doubling every quarter since launch. Average AI Mode queries run three times as long as conventional searches.

The model under the hood was upgraded to Gemini 3.5 Flash globally on May 19, 2026. Liz Reid, VP of Search at Google, described the upgrade as delivering “an even more powerful AI search experience” and said AI Mode is “the next step in our journey to bring together the best of a search engine with the best of AI.”

The query fan-out mechanic, in plain English

Here’s what AI Mode is doing every time someone runs a query.

1. User submits one query.
2. Google decomposes it into up to 16 parallel sub-queries.
3. Each sub-query retrieves and ranks its own set of sources.
4. Gemini 3.5 Flash synthesizes the results into a single answer.
5. Citations are surfaced inline, often as hover previews.
6. The user can ask a follow-up; the system retains the context.

What this means for SEO is significant. Ranking #1 for the original query is no longer enough. Your content might rank for the top-level query and still get skipped if a page that ranks #6 happens to have a stronger passage answering one of the sub-queries.

This is the data point that put it in stark relief: top-10 rankers accounted for 76% of AI Overview citations in mid-2025. By early 2026, that share had dropped to roughly 38%. Ranking position alone no longer determines citation.

Passages now matter more than pages. Entity-rich atomic answers, the kind that read like cleanly extracted definitions or comparison rows, win citations that the surrounding 2,000-word article doesn’t. The flip side: Digital PR work that earns coverage in authoritative outlets becomes structurally more valuable, not less, because those external mentions are what teach AI models who you are. We covered this intersection at length in Does Digital PR Matter in an AEO World?

The 93% zero-click reality

Seer Interactive’s behavioral data on AI Mode shows a roughly 93% zero-click rate inside the experience. Users get the answer, scan the citation chips, and bounce.

That sounds dire. The counter-argument: if your brand is cited consistently in AI Mode for queries your buyers run, the citation itself drives branded search lift later. People remember the names they keep seeing. The visibility compounds even when the click doesn’t happen.

That’s the operating model we use on cybersecurity client work. Get cited consistently for the category-defining queries. The branded search downstream is where the conversion actually happens.

Gemini: the chatbot, not the search engine

This is the one that confuses people the most because Google reused the name. Gemini is two things at once:

1. The family of large language models that powers everything we’ve been talking about.
2. The standalone consumer AI assistant that lives at gemini.google.com and in the Gemini app.

When most people say “Gemini,” they mean the second one. The chatbot interface that competes with ChatGPT and Claude.

The Gemini app hit 750 million monthly active users in Q4 2025, then 900 million by Google I/O on May 19, 2026. Sundar Pichai called Gemini 3 “the fastest adoption of any model in our history” on the Q4 earnings call in February. The model now processes more than 10 billion tokens per minute through direct API usage by Google customers.

What Gemini is built for

Gemini is the workspace. You go there for the long-running tasks, not the quick lookups.

• Writing essays, articles, drafts, marketing copy
• Coding: debugging, refactoring, generating new components
• Analyzing uploaded files (PDFs, spreadsheets, images, video)
• Long-context reasoning across documents thanks to the 1M-token context window
• Building Gems (custom personas)
• Persistent multi-day conversations

Gemini will search the web when it needs to. It just doesn’t default to it the way AI Mode does. The product is built around the model’s reasoning, not around the freshness of web results.

Why Gemini matters less for SEO directly

People sometimes ask me whether they should be “optimizing for Gemini.” The short answer is no, not directly. Gemini doesn’t return links to your site in any volume comparable to Search. It’s a chatbot, not a search engine.

But there’s an indirect angle. Gemini’s training data and retrieval behavior influence how the model represents your brand when asked. If your site is well-structured, your entity signals are clean, and your authority shows up across credible third-party sources, you’re more likely to be described accurately when a user types “what’s the best agency for digital PR” into the Gemini app.

That’s brand training data work, not citation work. Different optimization. Same underlying principle, and it’s where a well-run digital PR program earns compounding returns most marketers still miss.

The May 2026 I/O merge

On May 19, 2026, Liz Reid stepped on stage at Google I/O and called the redesigned search box “the biggest upgrade to our iconic search box since its debut over 25 years ago.” Then she announced that AI Overviews and AI Mode are merging into what Google is calling “one seamless AI Search experience.”

This matters. The boundary between “the summary on the SERP” and “the conversational tab” is being dissolved. The new intelligent search box accepts text, images, PDFs, videos, and even Chrome tabs as inputs. It expands to handle long conversational queries. Follow-up questions maintain context across what used to be two separate surfaces.

Liz Reid was careful to say “this new search box does not mean that you’ll only get AI responses.” The classic results page isn’t going away yet. But the operative word is “yet.” The trajectory is clear.

Three other things from the same announcement that you should be tracking:

1. Information Agents. Persistent background AI that runs queries on your behalf, “operating in the background 24/7,” monitoring web content and pinging you when conditions are met. Rolling out US-first this summer. This is search becoming agentic.
2. Generative UI. Google generates custom layouts for your specific question on the fly. Static result pages give way to dynamic interfaces.
3. Gemini 3.5 Flash as the default for AI Mode globally. Faster, cheaper, better at sustained agentic workflows. Available now in 200 countries and 98 languages.

The implication for marketers: the three-surface model I described above is going to collapse into one experience that picks up where the last conversation left off and runs background research while you sleep. The optimization work is still the same. The interface is what changes.

What this means for your SEO program

Here’s the operating model I’m running at Green Flag Digital, refined across cybersecurity, fintech, B2B SaaS, and consumer brand engagements through Q1 and Q2 2026.

1. Stop measuring only rankings. Start measuring citations.

The Ahrefs and Seer data make it impossible to ignore. Ranking #3 on a query that triggers AIO is materially less valuable than ranking #7 and being the cited source. Track citation share on your top 50 priority queries weekly. Tools like Semrush AI Visibility, Ahrefs Brand Radar, BrightEdge, Evertune, and Profound now expose this layer. Use any of them. Just start tracking.

2. Restructure content for passage extraction, not page ranking.

Query fan-out doesn’t pull pages. It pulls passages. A 40 to 60 word passage that directly answers a specific sub-question is more citation-likely than a 3,000-word article without clear answer boundaries.

What works in practice:

• Lead with a direct answer paragraph (under 60 words) that defines the term or answers the question
• Use definition boxes and clear sub-headings that mirror likely fan-out sub-queries
• Add comparison tables for “X vs Y” intents
• Embed FAQ blocks that match conversational follow-up patterns
• Cite original statistics, named experts, and recent studies. AI models lean on freshness signals.

3. Strengthen entity signals.

AI models cite entities they recognize. That means consistent naming across your site, social profiles, and structured data; named author bylines with clear bios; Organization and Person schema; and presence in third-party authoritative sources (industry publications, podcast appearances, conference talks). The plumbing side of that, the crawlability and indexation work nobody talks about, is what we tackle in our technical SEO consulting engagements. If AI crawlers can’t fetch your pages, none of the content work matters.

This is where our expert-led content bridge model comes in. The executive does the original thinking on podcasts, keynotes, and video. We amplify and structure it across the surfaces AI systems actually crawl. The exec leads. The distribution machine does the rest.

4. Build for the conversational follow-up.

People in AI Mode aren’t searching the way they search in classic Google. They’re starting conversations. Your content needs to anticipate the second question, not just answer the first. Cluster topics around full intent journeys, not isolated keyword targets.

5. Adopt a multi-surface evaluation framework.

This is what we call DeserveToRank at GFD. It’s a binary yes/no evaluation across the scope ladder (page+query, page, site, topic) crossed against the Google surface matrix (Search, Ads, News, AI Overviews, AI Mode). For every priority query, you ask the same question across each surface: do you deserve to rank here? Where the answer is no, you find the specific blocker and fix it.

The framework forces you to stop optimizing for the surface that’s already losing share and start optimizing for the surfaces where attention is migrating.

What the experts are saying

The SEO community is not in consensus on what to do, which is itself a useful signal. We’re early enough that no single framework wins.

Lily Ray, VP of SEO Strategy and Research at Amsive, has been one of the loudest voices arguing that AI search visibility is mostly E-E-A-T done well. The same authority signals that mattered in classic SEO still matter, just measured differently. Her recent framing on this:

“Large language models have to use search engines, that’s going to be true forever. They’re not going to be able to replace Google’s index anytime soon, or Bing’s index, so we have to be visible in search and do a good job in search as part of our AEO or GEO strategy. That’s going to be true forever.”— Lily Ray

Aleyda Solis, founder of Orainti, has emphasized crawlability as the prerequisite no one talks about. AI crawlers behave differently from Googlebot. They hit rate limits, JavaScript-heavy rendering blocks, and Cloudflare rules that screen out Perplexity and ChatGPT’s bots by default. Before optimizing a sentence for citation, the bots have to be able to fetch the page.

Mike King treats GEO as an engineering discipline. Schema, retrieval, embedding architecture. Rand Fishkin, who has been calling the death of click-optimized SEO for years, thinks the whole framing is already obsolete and that the real lever is brand presence in places people actually spend time.

Pick a side or hold all four. They’re all partially right. The point is that any agency selling you a complete framework right now is selling certainty that doesn’t exist. For ongoing analysis on how the AIO citation game is evolving in practice, Search Engine Journal’s coverage is the steadiest signal in the space.

Bottom line

AI Overviews, AI Mode, and Gemini are three different products solving three different jobs. The distinction matters because the optimization work for each is different. AI Overviews is where you fight for citation on the queries you’re already trying to rank for. AI Mode is where you build for passage extraction and conversational follow-up. Gemini is mostly indirect. Make sure your brand shows up correctly when the model is asked about you.

The May 2026 I/O merge is going to blur the lines over the next few quarters. The work doesn’t change. The interface does.

The companies that win the next five years of search aren’t the ones with the most pages or the most links. They’re the ones whose content gets pulled into the synthesized answer when it matters.

If you want to know how your site is performing across all three surfaces right now, that’s what we do at Green Flag Digital. Our AI, AEO, and GEO strategy services map your citation share across AI Overviews, AI Mode, ChatGPT, Perplexity, and Gemini for the queries your buyers actually run, then build the content and entity infrastructure to fix the gaps. Reach out and we’ll run the diagnostic.

Joe
Green Flag Digital