Most “AI referral traffic in 2026” listicles open with a number, stack more numbers on top of it, and never say which question each number is actually answering. This one will. Every statistic below names its source, its method, and its measurement window — and the places where the trackers disagree are the point, not a footnote.

The three questions that get conflated

When someone asks “how much AI search traffic is there?”, they are usually mixing three different questions:

  1. Volume — how many total visits to websites arrive from AI surfaces?
  2. Share — of those visits, which engine sends them (ChatGPT vs Gemini vs Perplexity)?
  3. B2B relevance — how much of that volume reaches B2B SaaS sites, or moves their pipeline?

Each of these has a different data source, a different method, and a different reliability. A large share of the “conflict” in the public numbers is just three different questions being answered with one kind of chart. If you are reading any AI-traffic number — including anything we publish — first figure out which of the three it is answering.

Volume: what Similarweb reports

The most-cited volume figure of 2026 comes from Similarweb’s AI search blog: 770.7 million average monthly referral visits to websites from AI search surfaces, up 117.4% year over year across a June 2025 – May 2026 window (Similarweb, measurement window June 2025 – May 2026: https://aisearch.similarweb.com/blog/ai-referral-traffic-by-industry/).

Four caveats before you reuse that number:

  • It counts referral visits — sessions that started in an AI surface and landed on a website — not searches, not impressions, not citations. Those are much bigger numbers and much further from revenue.
  • It covers the whole web across Similarweb’s tracked industries, which are consumer-skewed.
  • The +117.4% year-over-year growth is measured against a much smaller base a year earlier. Doubling is impressive; it does not mean the traffic is now large in absolute terms for most sites.
  • The 12-month window hides any recent slowdown or acceleration. If the second half of the window ran slower than the first, the average will not show it.

The honest read: AI referral volume is real, it grew a lot over the past year, and it is still a small slice of total web traffic for most sites.

A practical rule for reusing any of these numbers: never quote a figure without its window and method beside it. The same tracker, run over a different twelve months, produces a different number — and the gap between those numbers is method, not mystery. “AI traffic grew 117%” is only as defensible as the measurement window you attach to it.

Share: where the trackers disagree

For the “which engine sends the most” question, StatCounter’s monthly market share tracker is the most-cited source. The latest release, September 2026, puts ChatGPT at 80.48%, Gemini at 10.91%, Perplexity at 5.85%, Copilot at 1.79%, and Claude at 0.95% of AI chatbot referral traffic (StatCounter, September 2026 monthly data — the page’s chart window runs September 2025 – September 2026: https://gs.statcounter.com/ai-chatbot-market-share). Compare that with the widely cited April 2026 press release — ChatGPT 76.85%, Gemini 9%, Perplexity 7.73%, Copilot 3.76%, Claude 2.66% (StatCounter, April 2026: https://gs.statcounter.com/press/chatgpt-falls-to-all-time-low-as-ai-chatbot-referral-market-continues-to-fragment). Same tracker, same method, five months apart: the leader’s share rose, the tail shrank. That movement is the window-dependence in action, not an error in either figure.

Other trackers report a different ChatGPT share in a different window. Trakkr, for example, has reported ChatGPT at 90.7% share with +66.2% growth over a 30-day window (Trakkr, 2026, 30-day window — we did not use a source link for this one, and that absence is itself a data-hygiene flag worth noting).

We are not going to pick a winner between 80.48% and 90.7%. Market share is window- and method-dependent:

  • Window: 30 days versus a quarter versus a year captures different product moments and different user habits.
  • Sample: a global panel of sites versus a single vertical versus one country’s browsing data produces different shares.
  • Detection method: matching referrer domains, parsing UTMs, and browser-level telemetry all find different sets of “AI visits.”

What is stable across all of this: ChatGPT sends the most AI referral traffic, Gemini is second, and the long tail (Perplexity, Copilot, Claude) is real but small in share. For planning purposes, treat share as directional for the next twelve months, not as a constant to build forecasts on.

One operational consequence follows directly: if your analytics setup matches only one engine’s referrer, you are undercounting the total by construction. Whatever the true split turns out to be, tracking a single engine’s referral rows will never show you the whole number, and a rise in the smaller engines will look like noise.

The B2B SaaS number does not exist (yet)

Here is the gap the listicles skip: there is no free, industry-specific number for AI referral traffic to B2B SaaS sites. Similarweb’s industry breakdowns do not include B2B SaaS as its own tracked industry, and the share data above is web-wide.

So when an article, a tool, or a vendor tells you “X% of B2B SaaS sessions now come from AI”, they are either inventing it or re-labeling a consumer-average with your logo on it. We will not fill that gap with a made-up number either: the B2B SaaS figure is not available, full stop.

What you can do is produce your own number. It takes three sources — GA4 referral rows, Search Console’s generative AI performance report, and Bing Webmaster Tools’ AI Performance report — and our measurement guide walks through all three. Separately from referral volume, there is the citation question: whether your brand appears in AI answers to your buyers’ money queries at all. That is a different measurement, and it matters even if AI referral volume stays small — we break it down in how to get cited in ChatGPT if you run a B2B SaaS company.

“A signal, not a flood”: what to do operationally

If AI referral traffic is a signal rather than a flood, three operational rules follow.

1. Track it monthly and track direction, not level. Your own monthly AI-referral sessions will be a small number. A small number that goes from 10 to 30 in a quarter is more informative to you than anyone’s 770.7 million. Set a baseline, record the month-over-month delta, and review direction quarterly. When the monthly count is small, resist the urge to react to a single month — at those levels, most of the movement is noise, and the only defensible statement is the trend across several months.

2. Do not let anyone promise it. No vendor — including us — should commit to “AI will bring you N extra sessions or M extra deals by a certain date.” The base is too small and too noisy to support that kind of promise. The defensible offer is measurement and citation presence: a number you can defend, and evidence of where you were and were not cited.

3. Weight citations over referrals for B2B. A buyer who asks “what’s the best for ” in ChatGPT is at the top of your funnel. Whether they click through to you in that session is downstream of whether you were in the answer at all. The click-through evidence behind that framing — AI Overviews cutting organic CTR, with cited brands holding a premium — lives in the studies comparison piece we linked to above, and the honest framing is: being uncited is an active loss, and being cited is a measured win.

The full measurement setup — harness, scoring rubric, month-over-month diffs — is documented in our methodology.

Try the free quick check

If you want the first pass without a meeting: we run your brand’s money queries across ChatGPT, Perplexity, Gemini and Claude, and send back a snapshot scorecard — cited or not, and what the answers actually said about you. Get the free quick check.