ChatGPT referral traffic to the B2B sites Demandbase tracks rose 303% year over year. Demandbase puts it at 2.6 million visits a month in June 2026, up from roughly 645,000 a year earlier. The vendor published the benchmark on August 12, 2026; Demand Gen Report covered it. That number is real, and it is also the smallest part of the story the report tells.
One Product, Not a Category
Demandbase’s dataset covers June 1, 2025 through July 31, 2026. Demandbase counts more than 11 billion website visits, 75,645 campaigns and over $123 million in media spend. The 303% is a two-point comparison inside that window, June 2025 against June 2026.
Demand Gen Report’s writeup puts the other assistants in one line: “Referrals from Perplexity declined and Gemini and Claude remained flat.” ChatGPT accounted for “the overwhelming majority of AI-referred traffic,” per the same coverage. Demandbase publishes no baseline figures for the other three assistants. There is no way to size the decline, the plateau, or the distance between them.
What ChatGPT Referral Traffic Averages Per Account
Divide the numbers Demandbase published and a smaller picture appears. Spread 2.6 million monthly ChatGPT-referred visits across the 1,584 Demandbase instances behind the study and the result is roughly 1,640 visits per account a month, about 55 a day. This is our calculation, built from figures Demandbase published, not a number the company stated itself.
The average needs two caveats it cannot supply on its own. A Demandbase “instance” is a customer account, and one account can run several websites. That makes 1,640 a per-account average. At best it is an upper limit for any one site inside the account, not a fact about that site.
It is also only a mean. Demandbase publishes no distribution across its 1,584 tenants, no industry breakdown, and no sense of spread. A small number of large accounts could be carrying most of the curve.
A Step in May, Left Unexplained
The growth did not arrive as a straight line. Demand Gen Report describes a sharp inflection in May 2026, when monthly ChatGPT-referred volume “more than doubled versus prior months.” Trade coverage in PPC Land put it more bluntly: “The 303% increase in monthly ChatGPT referral visits was not smooth.”
Neither Demandbase nor the outlets covering it name a cause. PPC Land states the dataset’s limits directly: “The figures are not independently audited, the methodology is described rather than documented.” For a curve with an unexplained step in it, that gap matters more than the headline percentage.
Measure It in Your Own Analytics
None of this is a reason to dismiss the report. It is a reason to measure AI-assistant traffic in your own analytics. GA4 gives AI assistants their own channel row now, automatically, per Google’s release notes. The channel is not retroactive, so it will not rebuild the year Demandbase compares. From here on it beats taking a vendor’s word for the trend.
Undocumented methodology is not a new problem on this beat: we covered a case where two AI studies disagreed because their methods did, on a different question entirely. A growth curve without a published distribution is worth measuring independently. So is a step without a stated cause, before either gets repeated as fact.