AI Referral Traffic Statistics 2026: How Much AI Assistants Actually Send

Two things are true about AI referral traffic at the same time, and most commentary only tells you one of them. The first is that AI assistants send a rounding error's worth of visitors to the average website. On the best available measurements in mid-2026, everything from ChatGPT to Claude to Perplexity combined accounts for roughly a third of one percent of tracked referrals. Google alone still sends around eighty times more. The second is that the same channel is growing faster than any other acquisition source has in a decade, that visitors arriving through it behave differently, and that the search traffic it is displacing is falling fast enough for publishers to notice. What follows is what the primary datasets say, where they disagree, and what to do with numbers this small and this volatile.

Key AI referral traffic statistics for 2026

• AI platforms accounted for 0.32% of all website traffic in 2026, up from 0.02% in 2024, roughly 1 visit in every 312, according to SE Ranking [1]
• AI assistants supplied 0.35% of tracked referrals in July 2026 against 30.04% for traditional search engines, per Ahrefs Web Analytics data across 107,300 sites [3]
• ChatGPT holds 74.78% of AI referral traffic, ahead of Gemini on 11.56% and Perplexity on 7.23%, on SE Ranking figures [1]
• ChatGPT's outbound referral traffic grew 206% year on year between January 2025 and January 2026, on Semrush clickstream data [4]
• AI-sourced traffic to US retail sites rose 393% year on year in March 2026, with travel up 233% and financial services up 158%, per Adobe Digital Insights [7]
• 68.01% of US Google searches ended without a click in January to April 2026, up from 60.45% in 2024, according to SparkToro [9]
• AI Overviews cut position-one click-through rate by 34.5% across 300,000 keywords studied by Ahrefs [10]
• Google search referrals to news and media sites fell 34% between December 2024 and December 2025, and 60% for the smallest publishers, on Chartbeat network data [13]
• Anthropic's crawler made close to 70,900 page requests for every referral it sent in June 2025, the widest crawl-to-refer gap Cloudflare measured [15]
• Visitors arriving from AI platforms spent 9 minutes 19 seconds on site versus 5 minutes 33 seconds for organic search visitors, per SE Ranking [1]
• 54% of UK adults now use AI tools such as ChatGPT, Copilot or Gemini, Ofcom found in a survey of 7,533 people [16]

How much traffic do AI assistants actually send to websites?

Somewhere between a quarter and a third of one percent of all web traffic, depending on whose panel you trust. That is small enough to sit below most analytics noise floors, and it has been small enough for two years running while growing rapidly.

SE Ranking analysed 101,574 websites across 250 countries over 16 months and put AI platforms at 0.32% of all website traffic in 2026, against 0.02% in 2024 [1]. That is a sixteenfold rise over two years, and still only one visit in every 312.

Ahrefs, using its own Web Analytics tool across 107,300 sites, landed in almost exactly the same place: 0.35% for AI assistants in July 2026, against 30.04% for search engines as a category and 27.00% for Google specifically [3]. Two independent panels converging on the same third-of-a-percent figure is about as good as measurement in this space gets.

The interesting part is the ratio, not the percentage. Search still sends roughly 86 times more traffic than AI assistants do, so anyone claiming AI has replaced search as a traffic source in 2026 is describing a future state, not a measured one.

What makes the number worth watching is the direction of the two lines. AI referrals are compounding from a tiny base; search referrals to content sites are shrinking from a huge one, a squeeze the wider AI search statistics lay out in full.

Which AI platforms send the most referral traffic?

ChatGPT dominates, but by less than it did. SE Ranking put ChatGPT at 74.78% of AI referral traffic in 2026, with Gemini second on 11.56% and Perplexity third on 7.23% [1].

The two panels rank the middle of the table differently, which is a useful reminder that these are estimates from different samples rather than a census. Ahrefs, for instance, sees Meta AI as the second largest AI referrer, which SE Ranking does not break out at all.

ChatGPT's share is also eroding at the edges. SE Ranking's May 2026 update shows its worldwide share of AI referrals falling from 79.74% to 75.96% year on year, and from 83.03% to 76.77% in the EU [2]. In the UK, though, it barely moved, going from 86.88% to 86.27%, meaning British sites are far more ChatGPT-dependent than the global average.

That UK concentration matters for planning. If almost seven in eight of your AI referrals come from one assistant, a single product change at OpenAI is your entire channel risk.

How fast is AI referral traffic growing year on year?

Fast, but unevenly, and the headline growth rates hide a lot of month-to-month chop. Semrush measured 206% year-on-year growth in ChatGPT's outbound referral traffic between January 2025 and January 2026, drawn from over a billion lines of US clickstream data [4].

Growth by platform looks very different once you scale for base size. SE Ranking recorded Claude up 320% and Gemini up 231% year on year, against ChatGPT on 27% and Copilot on 31%, whilst Perplexity slipped slightly [1].

Product changes move this channel more than user behaviour does. Similarweb tracked a 157.7% rise in total ChatGPT referrals in the fortnight after OpenAI's 7 May 2026 update made brand links more prominent, with homepage-specific referrals up 354.7% [6]. SE Ranking saw the same event in its own data as a jump from 0.23% to 0.32% of all website traffic in a single month, with the UK up 38.7% and the EU up 42.7% [2].

Be careful reading that as adoption. A referral count that triples because a model started rendering links differently tells you about interface design, not about demand, and the same lever can be pulled back.

Which sectors get the most AI referral traffic?

Retail and travel are growing fastest; tech and software have the highest absolute share. Adobe Digital Insights, which measures server-side across its Analytics customer base, reported these year-on-year changes for March 2026 [7].

Adobe notes that tech and software had the highest AI visit share of any sector, roughly twice media and entertainment and four times retail [7]. Growth rate and share are inversely related here, which is what you would expect: the categories with the most room left are growing quickest.

The seasonal peaks are wilder still. Over the 2025 holiday period Adobe recorded retail AI traffic up 693% and travel up 539% year on year, with November 2025 alone at 769% for retail [8]. Those are the numbers that end up in conference slides, and they are real, but they are comparisons against a near-zero base in late 2024.

News sits at the opposite end. Chartbeat found AI platforms accounted for less than 1% of total pageviews across its network, and that AI referrals to news sites produced the lowest engagement it measures, just 4.8 pageviews per article [13].

Volumes there are absolute, not relative, and they are small. Similarweb data reported by Digiday counted 35.9 million global AI referral visits across 1,000 news sites in June 2025, of which ChatGPT supplied 81.7% [18].

Is AI traffic better quality than organic search traffic?

On engagement metrics, consistently yes. Whether that survives contact with your own conversion data is another question.

SE Ranking found AI-referred visitors spent 9 minutes 19 seconds on site against 5 minutes 33 seconds for organic search visitors, about 68% longer [1]. Similarweb reported ChatGPT referrals to transactional sites converting at 7% versus 5% for Google, with 15-minute sessions against 8 minutes [17].

Adobe's conversion picture is the most interesting because it shows movement rather than a snapshot. Retail traffic from AI converted 128% worse than non-AI traffic in October 2024 and 42% better by March 2026, a full reversal in seventeen months [7]. Travel closed a similar gap from 86% behind to 14% behind.

There is a selection effect buried in all of this that nobody has cleanly removed. People who reach a site after a multi-turn conversation with an assistant have already done their comparison work, so of course they convert better than someone who clicked the fourth blue link. The AI channel is not creating that intent, it is arriving later in the journey and taking credit for it.

That distinction decides whether you treat AI referrals as a new acquisition channel or as a reattribution of demand you were already getting, and I dig into it in the AI search conversion rate statistics. On current evidence it is mostly the latter, with a thin and growing layer of the former.

What has AI referral traffic done to organic click-through rates?

It has not replaced what Google's own AI features took away. That is the central asymmetry of 2026.

SparkToro, using Similarweb clickstream data, found 68.01% of US Google searches ended without a click in January to April 2026, up from 60.45% in 2024 and 49% in 2019 [9]. Ahrefs measured a 34.5% reduction in position-one click-through rate on keywords carrying an AI Overview, comparing 150,000 AI Overview keywords against 150,000 matched informational keywords across March 2024 and March 2025 [10].

Seer Interactive found organic CTR on AI Overview queries falling from 1.41% to 0.64% year on year, whilst being cited inside the AI Overview lifted CTR from 0.74% to 1.02% [11]. In 2025, the Pew Research Center put the click rate at 8% of visits with an AI summary present against 15% without, from 68,879 searches by 900 US adults [12].

The publisher side quantifies the damage. Chartbeat recorded Google search referrals down 34% between December 2024 and December 2025, with large sites down 22%, medium sites down 47% and small sites down 60% [13]. Press Gazette reported the same Chartbeat dataset showing a 33% global decline in the year to November 2025 and 38% in the US [14].

Set those against AI referrals at under 1% of pageviews and the arithmetic is brutal. A channel worth a third of a percent cannot offset a third of your search traffic disappearing, and it will not for several more years of compounding.

How many times do AI crawlers hit a site for every visitor they send?

Thousands, in most cases. The crawl-to-refer ratio is the cleanest way to express the value exchange, and it is heavily lopsided.

In June 2025, Cloudflare measured Anthropic making close to 70,900 HTML page requests for every referral it sent, whilst Mistral sent ten times more referrals than it made crawl requests [15]. Cloudflare itself flags the caveat that matters most here: traffic referred by Claude's native app carries no Referer: header, which inflates the ratio by hiding referrals that did happen.

Crawler share keeps shifting. Similarweb noted ClaudeBot rising from 9% to 20% of global AI crawler traffic by May 2026, with GPTBot at around 10% [17].

The practical use of this ratio is capacity planning and robots.txt policy, not moral outrage. A crawler consuming meaningful origin bandwidth and returning nothing measurable is a cost line with no matching revenue line.

Why does GA4 undercount AI referral traffic?

Because a large share of AI-referred visits never carry a referrer at all, and land in Direct instead.

Native apps are the main culprit. Cloudflare documented this directly for Claude, whose app sends no Referer: header [15], and the same problem applies to any assistant used outside a browser.

There is also a definitional mess. Referrals from chatgpt.com and chat.openai.com arrive as separate hostnames, Copilot traffic can appear under Bing, and Gemini referrals sometimes carry Google hostnames that get bucketed as organic search.

Nobody has a clean server-side number for total AI referrals. The panel estimates from Similarweb, Ahrefs, Semrush and SE Ranking exist partly because the analytics stack cannot answer the question, so treat your own GA4 figure as a floor rather than a measurement.

How to read these numbers

Panel data and server-side data measure different things. Similarweb, Semrush and Datos infer behaviour from clickstream panels, which are large but skew towards desktop and towards certain browser extensions. Adobe and Chartbeat measure server-side across their own customer bases, which is accurate for those sites and not necessarily representative of the web.

Growth percentages off tiny bases mean little alone. A 693% rise in retail AI traffic is real and also compatible with the channel staying under 1% of visits, which it is.

Watch for the difference between traffic to AI platforms and traffic from them. Similarweb's figures showing ChatGPT falling from roughly 76% to 53% of worldwide generative AI web traffic describe who is using which assistant, not who is sending referrals [5]. The two get conflated constantly.

Survey figures on AI usage are self-reported and generous. Ofcom's 54% of UK adults using AI tools came from 7,533 respondents interviewed between 29 September and 28 November 2025 [16], and there is a wide gap between having used ChatGPT once and using it instead of Google, a gap I unpack in the AI search adoption statistics.

Finally, several of the most-quoted CTR studies compare 2024 with 2025. They are the best evidence available, but they predate a year of further interface change, and Google does not release the data that would settle it.

What this means going into 2027

Stop budgeting AI referrals as a traffic channel and start budgeting them as an influence channel. At 0.35% of referrals with 86x more coming from search, the click volume will not pay for a programme on its own. The case rests on conversion quality and on visibility inside answers people never click through from.

Expect the measurement gap to close before the traffic gap does. GA4's handling of AI sources is improving, and more assistants will pass clean referrers under regulatory and commercial pressure. Reported AI traffic will jump for reasons that have nothing to do with real growth, so baseline your numbers now.

Plan for single-platform concentration risk, especially in the UK. With ChatGPT at 86.27% of UK AI referrals [2], one product decision at OpenAI moves your entire channel, as May 2026 demonstrated in both directions.

Judge sector benchmarks against your own base, not the headline. Retail growing 393% and tech holding the largest share are different problems requiring different responses, and the sector you are in determines which one you have.

Treat crawl-to-refer as a cost metric. If AI crawlers are taking thousands of requests per referral and your origin costs are rising, that is a decision to make deliberately rather than by default.

If you want help working out whether your own AI referral numbers are a floor or a real reading, and where to invest to grow them, you can book a call with me and we will look at your analytics together.