Three facts about AI traffic are true at the same time, and most arguments about it come from holding only one. AI referrals are small: roughly 1% of website visits, according to Conductor's analysis of 3.3 billion sessions across 13,770 enterprise domains. They convert unusually well: Adobe Digital Insights found visitors arriving from AI assistants converted 42% better than non-AI traffic in March 2026, a reversal from converting 38% worse a year earlier. And they undercount AI's real influence badly: Similarweb's downstream study found people who received a brand recommendation from ChatGPT were 2.5 times more likely to visit that brand's site in the following week, mostly through branded search that your analytics credits to something else.
Here is the 2026 data on how much traffic AI actually sends, where it comes from, where it lands, and why the referral number is the wrong thing to manage.
How much traffic does AI actually send?
Across the large studies, AI referrals sit between 0.1% and 2.8% of total traffic depending on industry and methodology, with Conductor's 1% the most representative figure. That is not a rounding error, but it is not a channel that replaces search either, and anyone selling it as one is ahead of the data.
Who sends it is more interesting than how much. Previsible's analysis of 6.77 million AI-referred sessions across 166 sites from late 2024 through May 2026 found ChatGPT responsible for 92.4% of trackable referrals from standalone assistants. But the mix underneath is moving fast: Claude overtook Perplexity in March 2026 after a 64-fold increase in referral sessions, while Perplexity's referral sessions fell 61% from their March 2025 peak. A B2B panel run by Goodie tells the same story from the buyer side: ChatGPT's share of measurable B2B AI referrals fell from 89% in mid 2025 to about 63% by spring 2026, with Claude at 18.5%, Gemini at 10.6%, and Perplexity at 7.3%.
And the largest AI-influenced traffic source is not a chatbot at all. Google's own AI Overviews and AI Mode produce more AI-influenced traffic than ChatGPT, Claude, Gemini, Perplexity, and Copilot combined, which is one reason Google's new Generative AI report in Search Console matters even though it shows impressions only. For the wider platform picture, our AI market share reference tracks usage across the engines.
Why AI visitors convert
The Adobe finding is the one that changes budgets. AI-referred retail visitors in March 2026 converted 42% better than non-AI visitors, spent longer on site, viewed more pages, and generated more revenue per visit. Twelve months earlier the same channel converted 38% worse. Something changed, and it was mostly the visitors.
The mechanism is simple once you see it. A person arriving from a search engine is usually still researching. A person arriving from an AI assistant has already asked the assistant to compare the options, read a synthesized answer, and chosen to visit one brand out of it. The assistant did the middle of the funnel. What lands on your site is closer to a decision than a query, which is why per-visit value runs high even when volume runs low.
Where AI visitors land, and why it matters
This is the part most sites get wrong. On May 7, 2026, ChatGPT began surfacing prominent, clickable brand links inside its answers instead of burying sources in footnotes. Similarweb's weekly referral data shows the share of ChatGPT referrals landing on homepages jumped from roughly 26 to 29% before the change to about 62 to 63% by late May, and it has held there rather than faded.
At the same time, 65% of the URLs ChatGPT cites sit two or three folders deep in a site's structure, and Previsible found 28.8% of ChatGPT referrals land on internal search results pages. Put the three findings together: AI reads your deep, specific pages to build the answer, then sends the convinced visitor to your front door or your search box. If your homepage still opens with a slogan and a hero image rather than the answer to "is this the right option for me," you are re-pitching a visitor who arrived pre-sold. And if your internal search is an afterthought, nearly a third of your AI visitors are meeting it first.
The measurement blind spot
Now the number that should reframe the whole conversation. Similarweb found users who got a brand recommendation from ChatGPT were 2.5 times more likely to visit the brand's website within seven days. Most of those visits arrive as branded search or direct traffic, not as a click on a ChatGPT link. Analytics logs a branded search visit; the recommendation that caused it gets zero credit.
That is why the 1% figure understates AI's influence and why managing to the referral number leads teams to the wrong conclusion. The referral shows who sent a click. It does not show who named you, described you, or handed the recommendation to a competitor, and those are the events driving the branded visits three days later. It also explains why tracking one engine measures a shrinking slice of a growing category: Similarweb's data has ChatGPT's share of generative AI web traffic sliding from about 76% in mid 2025 to about 53% by May 2026, with Gemini past a quarter and Claude up from around 2% to close to 9%.
Diploria is built for the layer the referral cannot see. It polls the questions buyers ask across 11 AI surfaces, records whether your brand was named and how, and reports share of voice against competitors, so the recommendation is measured at the source. The LLM Traffic view connects to Google Analytics 4 to show the referral side alongside it, engine by engine, so the two numbers can be read together instead of one being mistaken for the whole.
What to do this quarter
Segment AI referrals in GA4 by engine and watch the trend rather than the total; the total will stay small while the mix shifts under it. Make your homepage answer-ready for a visitor who has already compared you with two competitors, and test your internal search with the specific queries an AI-referred visitor would type. Track branded search volume next to AI mention rates, because the branded lift is where AI's influence actually shows up in your numbers. And measure the answers themselves, per engine, since the engines cite very different sources, as our most-cited sources breakdown shows, and reward very different presences. The free AI visibility check shows where you stand today in under a minute.