Something Inc.Schedule a free consultation
GEO

AI Search Visibility: Why AI Mode and AI Overviews Disagree

Google's two AI answer surfaces reach the same conclusion 86% of the time — and cite completely different sources 86.3% of the time. If your AI search visibility program treats "Google's AI" as one target, you are tracking the wrong thing.

JBJosh BernsteinManaging Partner · AUG 10, 2026 · 10 MIN READ

Most teams building an AI search visibility program still talk about "showing up in Google's AI." That phrase is doing more damage than people realize. Google runs two separate AI answer surfaces — AI Overviews, bolted onto classic search results, and AI Mode, a standalone conversational surface that has already crossed 1 billion monthly active users this year. New data on how those two surfaces actually behave shows they are not the same target. They reach the same conclusion to a query 86% of the time, but they cite different sources 86.3% of the time. Same company, same underlying model family, wildly different bibliographies. If a single company's own product line can't agree with itself on who to cite, no brand should assume one optimization playbook covers both.

TL;DR · 60 SECONDSAhrefs analyzed 1 billion data points across 14 studies and found AI Mode and AI Overviews agree on the answer to a query 86% of the time but cite different sources 86.3% of the time — only 13.7% source overlap. Wikipedia (29.7%), brand homepages (23.8%), and app stores (6.6%) dominate the citation mix, and 67% of ChatGPT's top citations come from sources brands don't control at all. The takeaway: AI search visibility is not one metric against one engine. It is a per-surface citation problem, and Google's own product line proves it.
86%
of the time AI Mode and AI Overviews give the same answer
86.3%
of the time they cite different sources for that answer
13.7%
source overlap between the two surfaces

AI Mode vs AI Overviews: Same Answer, Different Sources

The data comes from Ahrefs' Tim Soulo, in a June 2026 analysis spanning 1 billion data points across 14 studies, first shared publicly in a LinkedIn post and later summarized by industry outlets. It builds on the kind of large-scale citation research an Ahrefs analysis has been running throughout 2026, and it's the largest dataset yet comparing the two Google surfaces head to head. The core finding is almost a paradox. Run the same query through AI Mode and AI Overviews and you'll get the same conclusion 86% of the time — Google's retrieval systems are clearly drawing from a shared model family and largely agree on what's true. But ask which pages backed up that conclusion, and the two surfaces point to almost entirely different source sets. Only 13.7% of citations overlap. Flip that number around: 86.3% of the time, if you're cited in AI Overviews, you are not cited in AI Mode for the exact same query, and vice versa.

Think about what has to be true for that gap to exist. Both surfaces sit on top of Google's index and a related family of models, yet they clearly run different retrieval steps before generating an answer. AI Overviews leans on signals closer to traditional ranking — the pages already earning visibility in classic organic results. AI Mode behaves more like a research agent, chaining queries and pulling from a broader, less rank-dependent pool. The output looks similar to a user. The mechanism behind it is not, and that mechanism is exactly what a GEO program has to reverse-engineer if it wants consistent citations rather than lucky ones.

WHY THIS MATTERS MORE THAN IT SOUNDSAgreement on the answer with disagreement on the source means these are two different retrieval pipelines wearing the same brand. AI Overviews is stitched into the classic search index and ranking signals. AI Mode runs a more conversational, multi-step retrieval process closer to a chat product. Optimizing your generative engine optimization program for "Google AI" as a monolith ignores that split entirely.

This isn't a new theme for anyone tracking AI citations closely — we've written before about how engines disagree on sources across ChatGPT, Perplexity, and Google's surfaces. What's new here is that the disagreement shows up inside a single company's product line. If Google can't keep its own two AI surfaces aligned on citations, there is no version of "optimize for Google AI" that works as a single strategy. There is optimizing for AI Overviews, and there is optimizing for AI Mode, and they require separately tracked, separately built citation footprints.

Inside the Citation Mix: Wikipedia, Homepages, and the Long Tail

The same dataset breaks down what actually gets cited across the sample, by source type. It's a useful reality check for anyone assuming their blog content is the primary lever. Wikipedia alone accounts for 29.7% of citations in the dataset — nearly a third of everything cited traces back to a single reference site nobody on a marketing team controls. Brand and company homepages take 23.8%, which is the one category most teams can actually influence directly. App stores account for 6.6%, relevant mostly to product-led and mobile-first brands.

Wikipedia29.7%
Brand / company homepages23.8%
App stores6.6%

Share of citations by source type (Ahrefs, 1B data point analysis)

Read those three numbers together and a pattern emerges: over half of the citation surface area in this sample sits outside any single brand's content marketing efforts entirely. Wikipedia notability, structured entity data, and homepage authority carry more combined weight than most teams expect from a channel plan built around blog posts. That doesn't mean content doesn't matter — it means the content that matters most is often the entity-defining page, not the tenth explainer article. It also reinforces a point we've made before about why mention rate is the wrong dashboard metric to obsess over in isolation: knowing you're mentioned tells you nothing about which of these source types earned the mention, or which engine surfaced it.

THE CONTROL PROBLEM67% of ChatGPT's top citations come from sources marketers don't directly control — third-party sites, not owned domains. Combine that with Google's Wikipedia and homepage skew and the pattern holds across engines: most of your AI search visibility is decided by pages you don't operate. The owned-content playbook still matters, but it's one lever among several, not the whole system.

What This Means for AI Search Visibility Programs

Most GEO reporting still rolls everything into a single "AI visibility" number — one mention rate, one citation count, one dashboard tile labeled "Google AI." The 86.3% source-divergence figure makes that reporting structure indefensible. If AI Mode and AI Overviews draw from almost entirely different source pools 86.3% of the time, a blended score hides exactly the information a marketing team needs: which surface is citing you, which one isn't, and why. A brand could be strongly cited in AI Overviews and functionally invisible in AI Mode and never see it in a combined metric, because the aggregate number would still look healthy.

DIMENSIONAI OVERVIEWSAI MODE
Integration pointEmbedded in classic search results pageStandalone conversational surface
Retrieval behaviorTied closer to organic ranking signalsMulti-step, chat-style retrieval
ScaleShown across a large share of eligible queriesCrossed 1 billion monthly active users in 2026
Source overlap with the other surface13.7% shared citations13.7% shared citations
Answer agreement with the other surface86% same conclusion86% same conclusion

This is where ai citation tracking has to change shape. Treating AI Overviews and AI Mode as one line item in a report is the reporting equivalent of merging Google and Bing organic rankings into a single number because they're both "search engines." No one would accept that for organic SEO. There's no reason to accept it for generative engine optimization either, especially once you have real data showing the two surfaces barely share a source list. Teams running visibility work across B2B SaaS buyer journeys, where a single AI-surfaced comparison can influence a six-figure deal, feel this gap most acutely — a strong AI Overviews citation on a category query means nothing if the buyer's actual research happens inside AI Mode.

It also changes how a CMO should read a flat or declining visibility number. A drop in blended citations could mean the brand lost ground everywhere, or it could mean AI Mode's query mix simply shifted toward a topic where the brand never had strong sources to begin with, while AI Overviews held steady. Those are two completely different problems with two completely different fixes — one is a content and authority gap, the other is a tracking blind spot. A single dashboard tile can't tell them apart. Separate tracking can.

Why AI Search Visibility Needs Per-Engine Tracking

The practical fix isn't complicated, but it requires giving up the one-dashboard-fits-all model. Ai search visibility has to be tracked as a set of engine-specific citation footprints, not a single composite score, and each footprint needs its own baseline, its own source mix, and its own action plan. We laid out the mechanics of doing this properly in our per-engine GEO framework, and the AI Mode/AI Overviews split is the clearest evidence yet for why that framework exists.

01Split the citation baselineLog AI Overviews citations and AI Mode citations as separate datasets, not a merged total. A 13.7% overlap means a shared number is mostly noise — you need to know which surface is actually citing you.
02Weight entity and reference signalsWith Wikipedia at 29.7% and homepages at 23.8% of citations, entity clarity — notability, structured data, an authoritative homepage — competes directly with blog content for citation share. Audit both.
03Track third-party dependencyWith 67% of top citations on other engines coming from sources you don't own, map who is citing you indirectly — review sites, comparison hubs, industry press — and treat that list as a channel, not an afterthought.

None of this replaces the fundamentals. A page still has to be crawlable, structured, and worth citing before any engine will surface it. But once the fundamentals are in place, the remaining work is engine-specific: understanding that a citation win in AI Overviews doesn't transfer to AI Mode, that a Wikipedia mention moves both, and that a strong homepage is now doing SEO and GEO work simultaneously. Programs that still report "Google AI visibility" as one number are, by this data, guessing at which half of the picture they're actually improving.

01This weekSplit your citation tracking by surface
THE MOVES
Pull your current AI visibility reporting and separate any blended "Google AI" metric into AI Overviews and AI Mode line items
Re-run your top 20 tracked queries in both surfaces and log which sources each one cites
Flag any query where you're cited in one surface and absent in the other — that gap is your starting priority list
DONE WHENA citation tracker with AI Overviews and AI Mode as separate columns, not one combined score.
02This monthAudit your entity and reference footprint
THE MOVES
Check whether your brand has an accurate, notability-supported Wikipedia presence or entity page
Review your homepage for the structured, citation-ready facts an AI surface would pull directly
Identify the third-party sites currently citing you indirectly and treat outreach to them as its own workstream
DONE WHENA short list of entity and third-party gaps ranked by which surface they'd move.
03This quarterBuild a per-engine GEO roadmap
THE MOVES
Set separate targets for AI Overviews citation rate and AI Mode citation rate rather than one combined goal
Assign content and digital PR work against the specific gaps found in the audit, not a generic 'get cited more' brief
Re-test quarterly, since a 13.7% overlap figure means the gap won't close itself as content improves broadly
DONE WHENA roadmap with distinct owners, targets, and content plans for each AI surface you're tracked in.

See where you are cited today

A free snapshot audit of your rankings and AI citations before we ever talk.

JB
Josh BernsteinMANAGING PARTNER, SOMETHING INC.

Josh leads work at the intersection of SEO and generative engines at Something Inc., helping B2B brands get ranked and cited across every major AI engine.

Free consultation

Let us be the last SEO agency you ever work with

A 30 minute call and a free audit of your SEO and GEO position. You keep the findings either way.