The headline going around is that ChatGPT has fallen below half the market. It has. The part that gets dropped is that it did so without losing a meaningful number of visits, and that distinction decides whether you should be reallocating effort or simply adding to it.
Similarweb's worldwide traffic panel tracked generative AI web traffic from June 2025 through May 2026 and published the results in late July. ChatGPT went from about 76% of that traffic to roughly 53%. Gemini climbed from under 9% to around 27–28%. Claude moved from barely 2% to close to 9%, the largest proportional gain of anything tracked.
Worth being precise about what is being measured: this is share of visits to generative AI websites, not share of AI-assisted queries overall. Assistant usage embedded inside other products — Copilot inside Office, Gemini inside Search results, model calls inside third-party apps — does not show up in a web traffic panel at all. The direction of the trend is well supported. The absolute levels are a floor, not a full census.
What the AI search market share numbers actually show
| PLATFORM | JUN 2025 SHARE | MAY 2026 SHARE | CHANGE |
|---|---|---|---|
| ChatGPT | ~76% | ~53% | −23 points |
| Gemini | <9% | ~27–28% | +19 points |
| Claude | ~2% | ~9% | +7 points (≈4.5x) |
Read those as points and the story is Gemini. Read them as multiples and the story is Claude. Both readings are correct and they answer different questions: Gemini is where the volume moved, Claude is where the trajectory is steepest. A headline that picks one and calls it the winner is picking a denominator, not reporting a finding.
Gemini took the points, Claude took the growth rate
Gemini's climb is a distribution story rather than a preference story. It sits inside Search, Android, Workspace and Chrome, which means a large share of its growth came from surfaces users were already standing on. That is an enormously durable advantage and it is nearly impossible for a standalone product to out-market.
Claude going from 2% to 9% is a different kind of signal. There is no default placement doing that work, so the growth reflects people choosing it, largely for long-document work, coding and analysis. For anyone doing generative engine optimization, that matters more than the raw share: a smaller audience that arrived deliberately tends to be further down the funnel than one that arrived by default.
Share of generative AI web traffic, May 2026 (Similarweb worldwide panel)
Why a flat leader still loses share
This is the arithmetic that makes the headline misleading. If a market grows 70% and one participant's volume does not move, its share falls sharply even though nothing bad happened to it. ChatGPT did not shrink. Everything around it got bigger.
Run the numbers and it is unremarkable. Take a category at 100 units where one player holds 76. Grow the category to 170 while that player stays at 76, and its share is now roughly 45%. The 94 units of new demand went somewhere else entirely. That is close to what the panel captured, and it describes a market adding users faster than any single product can absorb them rather than a product losing its audience.
Distinguishing the two matters because they call for opposite responses. A shrinking platform is a reason to divest. A flat platform inside a growing category is a reason to hold your position and fund the new surfaces separately. Teams that read share loss as decline tend to cut the thing that is still producing, then wonder why total citations fell in a quarter when every engine-level number said otherwise.
“Share is a ratio. Traffic is a count. Strategy built on the ratio when you needed the count is how teams talk themselves out of a channel that is still working.”
The practical error this produces is a reallocation decision. A team sees ChatGPT down 23 points, concludes it is in decline, and moves effort to Gemini. But the effort that was earning ChatGPT citations is still earning the same absolute volume it always was. What has changed is that there are now two more engines worth earning citations in — which is an addition problem, not a substitution problem. Tracking that properly is exactly the discipline behind AI visibility tracking through market share swings.
One more wrinkle sits underneath the referral numbers. Similarweb noted that ChatGPT's May 2026 interface change, which surfaces homepage links more prominently, pushed homepage referrals from roughly 26–29% of its referral traffic to 62–63%, and that the shift held rather than decaying. If your ChatGPT referrals suddenly skewed toward your homepage this spring, that was a UI change on their end, not a content problem on yours.
The engines are not interchangeable
Treating these three as one channel assumes they behave alike. They do not, and the differences are large enough to change which pages earn you citations on each.
The practical consequence is that a single content investment does not pay out evenly. A well-structured comparison page may earn citations across all three; a dense technical reference may only ever surface in Claude; a locally-inflected query may only surface in Gemini because of its Search inheritance. None of that is visible while the reporting collapses everything into one AI row, and it is the reason AI engines disagree on sources more often than most teams expect.
What this changes about AI search market share strategy
Three things follow from a fragmenting market where the leader is not actually shrinking.
That third point is the one that changes budgets. Global share tells you where the population is; it tells you nothing about where your buyers are. We have seen B2B accounts where a single-digit engine drives a disproportionate share of qualified pipeline, which only becomes visible once the reporting stops averaging. The same effect turns up in how AI search market share moved through 2026 — the aggregate curve and the account-level curve rarely match.
Where to put the next quarter of effort
Split your AI visibility reporting by engine if you have not already, and hold the split for a full quarter before drawing conclusions — these shares moved 20 points in under a year and a one-month read is noise. Keep doing whatever earned you ChatGPT citations, because that volume has not gone anywhere. Then add Gemini and Claude as separate tracked surfaces rather than as a rounding error inside an AI bucket.
The underlying work does not fork much. Extractable structure, demonstrated authority and clean machine access earn citations across all three engines; what differs is which queries each one surfaces you for and how often it links out at all. Measure per engine, build once, and let the reporting tell you where the buyers actually are. If the reporting cannot answer that today, that gap is the first thing to fix — it is the same one we work through in reporting and analytics engagements.
See where you are cited today
A free snapshot audit of your rankings and AI citations before we ever talk.
Tyler 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.