AI Search Stats 2026: Market Share, Referral, and Citation Data

For three and a half years, “ChatGPT” was practically shorthand for generative AI. That’s no longer true. Similarweb data now shows ChatGPT’s share of worldwide generative AI web traffic sliding from roughly 76% a year ago to around 53% today, while Gemini has climbed past a quarter of all traffic and Claude has gone from a rounding error to the fastest-growing platform in the category.

For marketers, that single fact changes the job. Optimizing for one AI platform used to cover most of the addressable audience. It no longer does. This article walks through what Similarweb’s latest data, according to the 2026 Generative AI Landscape report, shows about platform market share, ChatGPT’s referral behavior since its homepage-link update, how often AI platforms cite sources, and a new wrinkle that didn’t exist a year ago: ads inside AI answers.

Generative AI usage is still accelerating, and how it’s growing has changed

The category itself is still expanding fast enough to make any single platform’s slice look small by comparison, and the shape of that growth is telling its own story.

• Average monthly web visits across generative AI platforms worldwide grew 70% year-over-year, reaching 9.5 billion between June 2025 and May 2026, while unique visitors grew more slowly, by 57%, to 655 million, according to Similarweb data.
• Visits are outpacing the audience that’s generating them, meaning existing users are opening these tools more often, not just that more people are showing up for the first time.
• App downloads, up 58% to 4.4 billion, track much closer to the unique-visitor number than the visit number, which suggests mobile growth right now looks more like new-user acquisition, while web growth increasingly looks like habit formation among people who already use these tools.

The AI market has fragmented into a three-way race

The clearest story in the data isn’t growth. It’s redistribution. According to Similarweb’s worldwide traffic panel, ChatGPT’s share of generative AI website visits fell from about 76% in June 2025 to roughly 53% by May 2026, a drop of more than 20 percentage points in twelve months. Gemini is the biggest beneficiary, rising from under 9% to around 27–28% of traffic over the same period. Claude moved from barely 2% to close to 9%, the largest proportional gain of any platform tracked.

Worth separating those two winners, because they’re winning different games. Claude’s growth is the largest number in this dataset, roughly quadrupling its share in a year, but it’s still starting from a small base, so that quadrupling accounts for only around 7 points of ChatGPT’s 23-point decline. Gemini, growing at a far less dramatic rate on paper, has actually absorbed most of the share ChatGPT lost.

That gap between “fastest-growing” and “biggest winner” matters for how marketers weigh these platforms: the one generating the most headlines about explosive growth isn’t the one you’d need to prioritize first if the goal is reaching the audience ChatGPT used to have to itself.

Mobile app performance data tells a genuinely different story than the web numbers do, not just a smaller version of the same one. On US iOS and Android, the platforms that barely register in the web share chart are the fastest movers: Meta AI and Claude posted the sharpest US app growth of any platform between June 2025 and May 2026, while Gemini, the platform actually eating ChatGPT’s web share, grew the slowest of the challengers on mobile. ChatGPT itself still grew solidly on its own app even as its web share slid, which suggests it’s defending the app it controls better than it’s defending the open web, where a user can switch platforms mid-search without installing anything.

Claude’s app story also looks different depending on where you’re standing. Its US growth is strong, but its worldwide app MAU growth is dramatically higher, suggesting Claude’s expansion is disproportionately international rather than US-led right now. Meta AI’s worldwide app growth runs nearly as high as Claude’s, and that’s arguably the more interesting number: Meta AI isn’t a destination people seek out the way they seek out Claude or ChatGPT, it’s distributed for free inside apps billions of people already had open.

An embedded assistant that matches a standalone challenger’s growth rate without asking anyone to download anything new is a preview of where much of future AI usage is likely to come from. Anthropic’s own numbers point the same direction: Claude’s unique visitors grew 180% in the second half of the year compared with the first half.

None of this means ChatGPT is shrinking. Its absolute visit count has stayed roughly flat, and it’s still the largest platform by a wide margin. What’s changed is the denominator: the category as a whole has grown fast enough that a flat leader looks like a declining one. And Gemini and Claude are winning share through different mechanisms that carry different ceilings. Gemini is riding distribution it already owns (Search, Android, Workspace), which scales as fast as Google chooses to push it. Claude is winning through deliberate adoption by developers and power users, a smaller pool that grows more slowly but tends to stick around longer once it commits.

The practical consequence for anyone tracking AI brand visibility is that single-platform coverage now misses a meaningful share of the AI-using audience, and which platform to watch depends on what you’re optimizing for. If reach is the goal, Gemini’s distribution advantage makes it the bigger near-term prize. If the goal is a technically sophisticated, high-intent audience, Claude’s growth pattern says otherwise. A strategy built around ChatGPT alone reaches roughly half the market either way. Similarweb’s AI Search Intelligence tracks brand presence across most of these platforms.

ChatGPT’s homepage-link update kept compounding

Back in May, ChatGPT changed how it surfaces brand links, replacing footnote-style citations with clickable brand names inside its answers. Similarweb data through late May 2026 shows the effect wasn’t a one-week spike. It kept building, and it never gave any of it back.

That last point is the one worth sitting with. A brand-new UI element usually produces a curiosity bump: people click the shiny new thing for a week or two, then behavior drifts back toward whatever it was before. That’s not what happened here. Homepage referrals had already been drifting slowly upward for two months before the update, so ChatGPT users clicking straight through to a brand’s front door wasn’t a new behavior the update invented. What the update did was take a pattern that was already a minority habit and make it the dominant one almost overnight, and it’s stayed the dominant one since. A plateau, not a decay curve, is the signal that this is a permanent shift in how people use ChatGPT’s answers rather than a novelty effect.

There’s a second tell in the data, and it’s easy to miss if you only look at the total traffic number. Total referrals and homepage-specific referrals didn’t grow at the same rate after the update, homepage referrals grew several times faster. If the update had simply pushed more of ChatGPT’s existing traffic pattern through the door, both numbers would have moved together. They didn’t, which means the update didn’t just add volume, it changed what a chunk of that traffic is. A meaningful slice of users are no longer clicking through to read more about a brand. They’re treating the brand name inside the answer as the endpoint of their research and the homepage as a formality on the way to typing the brand into their own memory later, or converting on the spot.

That’s a different job for a website to do than the one organic search traffic has trained marketers to expect. A user arriving from a traditional search result already has a specific question that page is supposed to answer. A user arriving at a homepage from inside a ChatGPT answer has often already made up their mind about the brand before they ever land, which makes this traffic behave less like a search click and more like the tail end of an ad impression. Homepage or direct-traffic spikes that used to get filed under “no clear source” increasingly deserve a second look before marketers write them off as unattributed.

Citations are still uncommon, but they’re growing fast

The rate at which ChatGPT cites sources in its answers is still low in absolute terms, but the trend line is steep. Similarweb data shows the presence of citations in US ChatGPT prompts rising from about 1.6% in June 2025 to roughly 6.8% by May 2026, more than quadrupling over 11 months. The climb wasn’t a straight line: citations jumped sharply starting in October 2025, then eased back for about four months before resuming their climb into spring 2026. That kind of pullback right after a sharp rise is worth remembering before treating any single month’s citation rate as the new normal. It’s as likely to be one data point in a noisy climb as it is a plateau.

Citation rates vary a lot by category. Travel & Hospitality prompts carry a citation about 23% of the time, the highest rate Similarweb tracked, followed by Automotive at roughly 20%. On the other end, Professional Services sits under 4%. Categories where users are comparing specific, checkable options (a flight, a car, a hotel) get cited noticeably more often than categories built on general advice.

The practical read hasn’t changed even as the numbers have moved: most brands still aren’t showing up in AI citations most of the time, which means the competitive field for AI citation analysis and structured, referenceable content remains wide open rather than saturated.

Ads have entered AI answers

This is the part of the 2026 picture that didn’t exist when the market share numbers looked the way they used to. According to Similarweb’s 2026 Generative AI Landscape report, roughly 26% of ChatGPT responses now contain an ad, a number with no chart in this piece to back it up on its own, but one worth sitting with before looking at where those ads actually land.

Of the ads that do show up, timing is everything. About a third of all ad impressions are served in the very first response in a session, when a user hasn’t yet revealed much about what they’re actually trying to do. That share drops quickly with each subsequent message: by the third response, it’s down to single digits, and it keeps thinning out from there. But it never goes to zero. A meaningful fifth of all ad impressions still show up ten or more messages into a conversation, meaning long, exploratory sessions, the kind where a user has been thinking out loud with the AI for a while, remain a live opportunity for ad placement well after the obvious early real estate is gone.

For brands, this adds a genuinely new competitive layer to AI visibility. It’s no longer only about whether you’re mentioned or cited. It’s increasingly about who’s buying placement around the answer you’re trying to be found in, and the data above suggests that placement isn’t a one-shot bet on the first message, it’s a competition that resets and re-weights with every turn of the conversation.

Search behavior is shifting alongside the platforms

Google’s AI rollout looks less like a single feature launch and more like two different bets running at once, and the 2026 Landscape report shows one of them clearly winning out. AI Overviews, the version of AI that shows up automatically inside a search a user already knows how to run, have kept climbing without asking anyone to change their habits. AI Mode, which asks users to deliberately switch into a more conversational interface, grew fast early on and has since leveled off and started to soften.

That gap is worth highlighting: people seem far more willing to accept AI inside the tools they already use than to adopt a new tool built around AI. Passive integration is beating deliberate reinvention, at least for now, which says something about how much friction even a genuinely better interface has to overcome.

That same preference for passivity shows up in the way people type. Search queries have gotten noticeably longer since AI-native search became normal, which only makes sense if users have started trusting the system to do more of the interpretation for them rather than doing the keyword engineering themselves. Clipped, keyword-style searches were a habit built around a keyword-matching engine. Longer, conversational queries are a habit built around a system users believe can actually understand context, and that shift quietly changes what “ranking for a keyword” is even supposed to mean.

One myth worth breaking is that AI chat is a young person’s habit. Overall usage has kept climbing even as Gen Z’s share has shrunk, which means growth is coming disproportionately from people who weren’t first in line to adopt these tools. AI chat is moving past being a novelty for early adopters and starting to look like infrastructure that everyone eventually uses, which is a very different targeting assumption than the one most brands built their AI-visibility strategy around a year ago.

What this data means for marketers in 2026

Coverage now means more than one platform

With ChatGPT below 55% of AI web traffic and Gemini and Claude both holding meaningful, growing shares, a GEO strategy built around a single assistant reaches a shrinking fraction of the AI-using audience. Similarweb’s AI visibility momentum tracking exists largely because this fragmentation makes month-to-month platform comparisons matter more than they used to.

Homepage and direct traffic deserve a second look

ChatGPT’s referral behavior has settled into a pattern where roughly six in ten referred visits land on a homepage rather than a specific page. Traffic that used to get filed under “direct, no clear source” may increasingly be AI-driven discovery that simply doesn’t carry a clean referral tag.

The citation door is still open, but it’s narrowing by category

A sub-7% overall citation rate means most content still isn’t making it into AI answers verbatim, but categories built on comparable, checkable facts (travel, automotive) are already seeing citation rates three to five times the average. Brands in those categories have less runway than brands in advice-heavy categories like professional services.

Ads are a new variable in AI visibility, not just search visibility

With roughly a quarter of ChatGPT answers now carrying an ad, and ad density concentrated in the first response of a session, appearing early and being cited cleanly both carry more weight than they did a year ago.

Search query behavior is quietly changing what “ranking” means

Longer queries and a 43%+ AI Overview appearance rate in the US mean that a growing share of search sessions are being interpreted and summarized before a user ever sees a traditional results page. Tracking brand mentions inside AI answers is becoming as relevant as tracking keyword rankings.

The AI platform market went from a single dominant player to a genuinely competitive three-way field in about a year, and the rules governing referral behavior, citations, and now advertising are still being written in real time. The brands paying attention to how these numbers move month to month, not just the annual snapshot, are the ones positioning themselves for whatever the field looks like in 2027.

FAQs

Is ChatGPT still the largest AI platform in 2026?

Yes, by a wide margin in absolute visits, but its share of the category has fallen sharply. Similarweb data shows ChatGPT’s worldwide web-traffic share dropping from roughly 76% in June 2025 to about 53% by May 2026, even as its total visit count has stayed roughly flat. The drop reflects the category growing faster than ChatGPT specifically, not ChatGPT shrinking.

Which AI platform is growing fastest in 2026?

Claude has posted the fastest proportional growth of any major platform. Its worldwide app MAU grew roughly 1,318% between June 2025 and May 2026, according to Similarweb data, and its worldwide web-traffic share rose from around 2% to close to 9% over the same period.

Did the ChatGPT homepage-link update from May 2026 have a lasting effect?

Yes. Homepage referrals from ChatGPT rose from roughly 26–29% of all referral traffic before the update to about 62–63% by late May 2026, and that level has held rather than faded, based on Similarweb’s weekly referral data.

How often does ChatGPT cite sources in its answers?

Citation presence in US ChatGPT prompts rose from about 1.6% in June 2025 to roughly 6.8% by May 2026, according to Similarweb data. Rates are highest in Travel & Hospitality (about 23%) and Automotive (about 20%), and lowest in Professional Services (under 4%).

Do AI chatbots show ads now?

Yes. Similarweb’s 2026 Generative AI Landscape report found that roughly 26% of ChatGPT responses now contain an ad, with ad density concentrated most heavily in the first response of a session.

How many people still use ChatGPT compared to alternatives?

ChatGPT remains the largest AI assistant by both web and app usage, but Gemini now holds roughly a quarter of worldwide AI web traffic and Claude is approaching one-tenth, meaning a strategy that only accounts for ChatGPT no longer reaches the majority of the AI-using audience on its own.

by Maayan Zohar Basteker

Senior SEO Specialist at Similarweb

Maayan is a senior SEO specialist with 7+ years of experience in SEO. She loves complex research projects, creating SEO strategies and performing technical audits.