ChatGPT's AI Search Market Share: 92%, 62%, or 53%?

Ask three research teams what ChatGPT's AI search market share is right now, and you'll get three different answers: 92%, 62%, and 53%. All three were published within about six weeks of each other, in the middle of 2026, by credible, well-resourced sources with real methodology behind them. None of the three is wrong. They're answering three different questions, and almost nobody citing them says which one out loud.

Previsible says ChatGPT accounts for 92.4% of standalone LLM referral traffic. Similarweb says ChatGPT's share of worldwide generative-AI platform visits fell to 52.7%. A B2B-specific referral dataset puts ChatGPT at 62.6% of measurable B2B AI referral traffic. Pick the wrong one of these three to anchor a strategy decision on, and you'll either panic about a concentration risk that doesn't apply to you, or ignore a competitive shift that does.

AI Search Market Share Depends on What You're Measuring

The instinct, when three numbers this far apart show up in the same quarter, is to assume one report is sloppy. That's not what happened here. Previsible, Similarweb, and the team behind the B2B referral dataset all built defensible, distinct methodologies. The problem is that the industry has collapsed all three into a single phrase ("ChatGPT's market share") and then argues about whose number is right, when the real answer is that they're not measuring the same market. If your reporting and analytics stack is built to track a single blended AI-share number, you've already built the wrong dashboard.

Three Reports, Three Different Questions

Strip each report down to what it actually counted, not what headline it produced, and the disagreement stops looking like a contradiction. Previsible measured what happens on your website: a visitor arrives with a referrer header from chatgpt.com, claude.ai, or a comparable domain, and that session gets logged as LLM-driven traffic. It explicitly excludes Google AI Overviews, because AI Overviews traffic mostly doesn't carry a clean referrer that GA4 can attribute, a limitation covered in more depth in our AI referral traffic concentration analysis. Similarweb measured something structurally different: total traffic to the AI platforms themselves, as destinations, across every device, regardless of what those visitors did once they landed. A ChatGPT session that never leaves chatgpt.com still counts in Similarweb's number and never counts in Previsible's. And the B2B dataset narrowed Previsible's referral question to a single vertical, tracking only referral sessions landing on B2B properties.

Previsible: ChatGPT Owns 92% of Referral Traffic

Previsible's "2026 AI Traffic Report," published in July 2026, is the most directly useful of the three numbers if your question is narrow and practical: when an AI system sends a visitor to a site, which AI system is doing the sending? The study pulled GA4 data from 166 properties spanning 10 industries, covering November 2024 through May 2026, and isolated 6.77 million sessions with a referrer traceable to an LLM platform. Across that window, monthly ChatGPT-referred sessions in the sample grew from roughly 47,606 to 610,910, a 12.8x increase in a year and a half. Of every trackable LLM-referred session in the dataset, 92.4% arrived via ChatGPT.

That 92.4% figure gets quoted constantly as "ChatGPT's AI search market share" without the caveat that sits right underneath it: this is referral share only, and it explicitly excludes Google AI Overviews traffic, which doesn't pass a clean referrer to analytics tools. That's a meaningful exclusion (AI Overviews sits in front of an enormous share of Google's search volume), but it's also a defensible one, because you genuinely can't attribute that traffic in GA4 the way you can a chatgpt.com referral. If you're deciding where to spend GEO engineering effort on your own referral traffic, this is the number closest to your actual reality. The methodology and its blind spots are worth reading in full alongside our own work attributing pipeline to organic and AI sources, which runs into the same referrer-tracking ceiling.

Similarweb's Platform-Visit Share Slips to 53%

Similarweb's "AI Search Stats 2026," published July 29, 2026, asks a completely different question: not "who sends traffic to third-party sites," but "who's winning the AI-assistant category itself." It tracks all-device web traffic to the generative-AI platforms as destinations (visits to chatgpt.com, gemini.google.com, claude.ai, perplexity.ai, and peers), independent of whether any of those visits ever produced a referral to somewhere else. By that measure, ChatGPT's worldwide share of generative-AI platform traffic fell to 52.7% by May 2026, down from roughly 76% twelve months earlier. That's a 23-point drop in a single year, inside a category that itself grew fast: total average monthly visits across generative-AI platforms worldwide rose 70% year-over-year to 9.5 billion, for the twelve months ending May 2026.

Two things are true about that number at once. First, ChatGPT is still the single largest generative-AI platform in the world by a wide margin: 52.7% of a 9.5-billion-visit category is not a company losing the market. Second, a 23-point share decline in twelve months, in a category growing 70% year-over-year, means the absolute number of visits going to competitors is growing faster than ChatGPT's own. Read via Similarweb, that's the number that should inform a genuinely different decision than Previsible's: not "where do I optimize my referral funnel," but "do I need a Gemini-specific or Claude-specific GEO strategy because the competitive field at the platform level is shifting under me." A team that only ever looks at its own referral logs will never see this shift coming, because none of it shows up in GA4 until a competitor's growth eventually shows up as a dent in ChatGPT's own referral volume, months later.

The B2B Referral Number: 62.6%

The third data point narrows Previsible's referral-traffic question to a single vertical. A B2B-specific referral dataset covering March-April 2026 found ChatGPT's share of measurable B2B AI referral traffic at 62.6%, with Claude at 18.5%, Gemini at 10.6%, and Perplexity at 7.3%. That's a meaningfully more concentrated field than the broader-market Similarweb number, and a meaningfully less concentrated one than Previsible's 92.4% cross-industry referral figure, which makes sense, because B2B buying behavior and B2B AI tool adoption simply don't mirror the broader consumer-heavy sample Previsible pulled from.

Claude's 18.5% share here is the detail worth sitting with. It's roughly the same order of magnitude as Gemini and Perplexity combined in this sample, which tracks with what a lot of B2B SaaS teams already sense anecdotally: technical and enterprise buyers reach for Claude more often than the broader consumer population does. If your company sells into B2B SaaS specifically, this is the most applicable of the three numbers on the table. It's not as extreme as Previsible's cross-industry figure, and it's a referral number rather than a platform-visit number, so it maps directly onto the traffic your own analytics stack can actually see.

“Everyone is trying to measure AI visibility, but the thing they're measuring isn't stable.”

Kevin Indig made that point at Growth Memo in his "AI Halftime Report: H1 2026," published August 3, 2026, naming attribution as the single defining theme of the first half of the year for AI and growth measurement. His argument, paraphrased: it's not just that different reports use different methodologies. It's that the tools and definitions practitioners rely on keep shifting underneath them, so a number that looked solid in Q1 can mean something different by Q3 without anyone changing the underlying reality being described. That instability is exactly why a bare "ChatGPT has X% market share" headline, stripped of what it measures, travels so easily and misleads so often. Our own study of AI citation source concentration ran into the same issue from the citation side: concentration numbers move depending on what population of queries and sources you sample.

Which ChatGPT Market Share Number You Should Actually Use

None of this means the three reports should be averaged, and it doesn't mean you should pick whichever number makes the best slide. It means the right number depends entirely on the decision sitting in front of you. Quoting a bare AI search market share percentage, without saying whether it's referral share, platform-visit share, or a vertical-specific referral share, isn't just imprecise. It's close to meaningless, because the three numbers imply three different actions.

Most of the anxiety circulating right now ("are we too concentrated on ChatGPT," "are we ignoring Gemini," "is our AI traffic story getting worse") gets resolved the moment you match the right lens to the actual decision. A team asking whether to build Gemini-specific optimization should be looking at Similarweb's platform trend line, not Previsible's referral share. A team reporting AI-driven pipeline to a B2B SaaS board should reach for the B2B referral number, not a cross-industry blend. And a team deciding where GEO engineering hours go this quarter should use Previsible's referral figure, because that's the traffic actually landing on their own domain.

The averaging instinct is the one to resist hardest. Blending 92.4%, 62.6%, and 52.7% into a single "ChatGPT is roughly 70% of AI search" line produces a number that describes nothing real: not your referral traffic, not the platform category, not your vertical. It's an artifact of arithmetic, not a measurement of anything a buyer, competitor, or board member is actually asking about. The same discipline applies to picking whichever figure is most dramatic for a slide. A 92.4% headline gets more attention in a deck than a 52.7% one, which is exactly why it shows up in more decks, regardless of whether the decision at hand is about referral traffic or platform share.

None of the three reports will hold still, either. Previsible's next update will reflect whatever referral mix exists by the time it publishes; Similarweb's monthly trend line moves every time a new platform launches or a pricing change shifts usage; the B2B dataset covers two months and will need a longer window before anyone should treat it as a trend rather than a snapshot. Treat each as a periodic read on a moving target, tied to a named methodology and a named publication date, and re-check the number against the decision it's meant to inform before quoting it in a board deck or a client report.

See where you are cited today

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