ChatGPT referral traffic to B2B websites grew 303 percent in a year and still accounts for roughly 0.33 percent of measured traffic. PPC Land reported on 16 August 2026 on a Demandbase Labs analysis covering 1,584 platform instances and more than 11 billion website visits between 1 June 2025 and 31 July 2026, which found monthly ChatGPT referral visits rising from about 645,000 in June 2025 to 2.6 million in June 2026.
Both halves of that sentence are the story. A near quadrupling is real growth. A share of measured traffic near a third of one percent is a rounding error in most B2B reporting. Anyone quoting one number without the other is selling something.
What Demandbase Labs actually measured
PPC Land described the Demandbase Labs dataset on 16 August 2026 as 1,584 distinct platform instances and over 11 billion website visits, collected between 1 June 2025 and 31 July 2026, with the headline comparison drawn between June 2025 and June 2026. Monthly ChatGPT referral visits went from roughly 645,000 to 2.6 million across that window, a gain of close to 2 million additional visits per month and a 303 percent year on year increase.
The unit here is a platform instance, meaning a customer of the vendor rather than a website or a company in the general sense. That matters for how the average is read. There is no industry breakdown of those tenants published, and no statement about whether the same tenants were present throughout all 14 months, so the June 2025 base and the June 2026 figure are not guaranteed to describe an identical population.
Why 303 percent and 0.33 percent are both true
The percentage growth and the share of traffic describe different things, and PPC Land published both on 16 August 2026. Growth is measured against a small base. Share is measured against the whole. Starting from 645,000 monthly visits inside a pool of billions, tripling produces a large percentage and a small slice.
PPC Land also noted the comparison against search: ChatGPT sends approximately 190 times less traffic than Google in this dataset. That ratio is the sentence to keep when a stakeholder asks whether the channel has arrived.
| Measure reported by PPC Land, 16 August 2026 | Figure |
|---|---|
| Monthly ChatGPT referral visits, June 2025 | About 645,000 |
| Monthly ChatGPT referral visits, June 2026 | 2.6 million |
| Year on year change | 303 percent |
| Share of measured traffic in the dataset | About 0.33 percent |
| Average extra ChatGPT visits per tenant per month | About 1,700 |
That last row is arithmetic, not a measurement. Divide 2.6 million by 1,584 tenants and you get roughly 1,700 visits each. PPC Land flags directly that no distribution data, no median and no dispersion figures were published, so the average is almost certainly hiding a concentrated set of winners. A handful of tenants with large audiences could account for most of the volume while the median tenant sees a fraction of 1,700.
The May 2026 inflection that nobody explained
PPC Land reported on 16 August 2026 that ChatGPT referral volume more than doubled against prior months in May 2026, described as the sharpest jump in the dataset, and that Demandbase provided no explanation for it. That gap is worth sitting with, because a single unexplained month sitting inside a 14 month trend line does a lot of work in a 303 percent figure.
A step change of that size in one month usually has a mechanical cause rather than a behavioural one. Product changes to how links are surfaced, changes to referrer headers, changes to how a measurement platform classifies a source, or a shift in how a large tenant tags traffic can all produce it. None of those were ruled in or out in what was published. Until someone explains May, the trend should be treated as a trend with an asterisk.
What happened to Perplexity, Gemini and Claude
According to PPC Land on 16 August 2026, the Demandbase release stated that Perplexity referrals declined while Gemini and Claude remained flat, but published no absolute volumes, percentages or shares for any of the three. That is a meaningful omission rather than a detail. Without a base, a decline could be a fall from a large number or noise around a tiny one, and flat could mean stable at scale or stable at nearly nothing.
The practical read is that this dataset supports statements about ChatGPT and supports very little about the rest. Anyone building a channel plan that ranks assistants against each other needs a source that published the underlying volumes.
The conversion quality evidence contradicts itself
PPC Land set out on 16 August 2026 that published evidence on whether AI referrals convert better than other channels points in opposite directions depending on who measured it. Invoca reported ChatGPT calls converting at 49 percent, described as the highest rate of any channel. Ahrefs reported AI visitors converting at 23 times the rate of conventional organic traffic. Microsoft Clarity reported AI referred conversions at three times traditional channels. An e commerce study found ChatGPT underperforming traditional channels outright.
Those results are not reconcilable as they stand, and the reason is almost certainly methodology rather than reality. Different definitions of a conversion, different attribution windows, different industries, and above all different treatment of visitors who arrive with high intent because they already decided before the click. A 23x figure derived from a self selected, late funnel visitor population is not comparable with a channel average.
The defensible position is that nobody has published a settled answer on AI referral quality, and a marketer quoting any single one of these multipliers in a board deck is choosing a number rather than reporting one.
Most AI influence never shows up as a referral
The attribution problem sits underneath all of the above, and PPC Land raised it on 16 August 2026 by citing Similarweb data showing only 8.8 percent of AI influenced visits arrive as direct referrals, with roughly 56 percent arriving through branded search instead. Those branded search visits land in analytics as organic search, indistinguishable from a user who found the brand any other way.
If that pattern holds, referral counts systematically undercount AI influence, and the 0.33 percent share is a floor rather than a total. It also means the honest version of AI referral attribution is a range with a wide error bar, not a line item. This is the same measurement gap that has always existed between what causes a purchase and what gets credited for it, and it has not become easier.
What the report did not say
PPC Land was explicit about the gaps in what Demandbase published. There is no industry breakdown of the 1,584 tenants. There is no statement on tenant consistency across the 14 month window. There are no absolute volumes for Perplexity, Gemini or Claude. There is no figure for AI referrals as a percentage of total visits across all assistants. There is no explanation for the May 2026 jump. There are no dispersion or median figures. The methodology was described rather than documented, and the underlying data is proprietary and has not been independently audited.
The report also said nothing about revenue, pipeline or deal size attributable to these visits, nothing about which countries the tenants operate in, and nothing about language. Any claim about what this means for a specific market is an extrapolation.
What to check in your own analytics this week
The useful response to a dataset like this is to measure your own version of it rather than adopt someone else's average.
- Pull the last 14 months of referral traffic from chatgpt.com and chat.openai.com in your analytics property, monthly, and look at whether your own curve has the May 2026 shape or a different one.
- Check what share of your total sessions that represents, then decide, in advance, what share would justify dedicated investment. Writing that threshold down before the number moves prevents a lot of arguing later.
- Look at branded search volume over the same window, since the branded search route is where most AI influence is believed to land, and a rise there with flat referrals is a signal your referral count is missing.
- Segment conversion rate for AI referred sessions separately, with enough volume to mean anything, rather than importing a multiplier from a vendor study.
- Confirm your measurement setup is not silently dropping or misclassifying these sources, which is a routine analytics configuration question rather than an AI question.
What this means for Thai marketers
The Demandbase analysis reported by PPC Land did not break its 1,584 tenants down by country, so nothing in it describes Thailand, and no share of measured traffic figure for the Thai market exists in this source. The reasoning that does transfer is about proportion and about patience.
For a Thai B2B business, a channel sitting near a third of one percent of traffic does not justify pulling budget out of anything that currently works. It does justify making the channel visible in reporting now, at low cost, so that a real trend can be recognised locally rather than imported from a US heavy dataset. Thai language behaviour is a genuine unknown here, and the branded search caveat probably bites harder in a market where a large share of discovery already runs through platforms that never pass a referrer at all. Teams doing structured generative engine optimisation work, or measuring their presence inside ChatGPT answers, will get more out of tracking their own baseline for two quarters than out of any single headline percentage.
FAQ
Is ChatGPT a meaningful traffic source for B2B yet?
Not on these numbers, at about 0.33 percent of measured traffic in the dataset. It is growing quickly from a small base, and PPC Land noted it sends roughly 190 times less traffic than Google in the same analysis, so it is worth tracking rather than reallocating budget toward.
Does this data cover Thailand?
The source did not state this. No country or regional breakdown of the 1,584 platform instances was published, and no language breakdown was given either, so the dataset cannot tell a Thai marketer anything specific about the Thai market.
Do AI referrals convert better than organic search?
There is no settled answer, and the published figures contradict each other. PPC Land listed results ranging from AI referrals underperforming traditional channels in an e commerce study to Ahrefs reporting 23 times the conversion rate of conventional organic traffic, with the difference driven by methodology rather than an agreed finding.
Why did ChatGPT referrals jump in May 2026?
Demandbase did not explain it, and PPC Land said so plainly. Volume more than doubled against prior months in the sharpest single move in the dataset, with no cause published, which is why the 303 percent annual figure should be read with that month in mind.
Do I need to change anything right now?
Nothing structural. The low cost step is to make the channel measurable in your own reporting and to watch branded search alongside it, because most AI influenced visits are believed to arrive as branded search rather than as a trackable referral.
If you want help separating the signal from the headline in your own analytics, the team at Relevant Audience can walk through your numbers with you. The full write up, including the caveats, is on PPC Land.
