Monthly referral visits from OpenAI's assistant to the business-to-business web properties measured by Demandbase climbed from roughly 645,000 in June 2025 to 2.6 million in June 2026, the account-based marketing vendor said on August 12, 2026. Rival assistants did not follow: referrals from Perplexity fell over the same period, while Gemini and Claude stayed flat.

The figures come from Labs by Demandbase, the research arm of the San Francisco company, and were distributed through a wire release timed for 9:00 a.m. Eastern on Wednesday, August 12. According to Demandbase, the 303% increase in monthly ChatGPT referral visits was not smooth. Volume rose through the year and then jumped sharply in May 2026, a month in which, the company said, referral traffic more than doubled against prior months.

That inflection is the most specific thing in the release, and also the least explained. Demandbase does not attribute it to a product change at OpenAI, a shift in its own customer base, or a measurement adjustment on its side.

A fourfold rise measured on one vendor's estate

The headline number describes a single dataset. Demandbase says the analysis draws on more than 11 billion website visits recorded across 1,584 distinct platform instances, which the company calls tenants, between June 1, 2025 and July 31, 2026. The year-over-year comparison sits inside that window rather than spanning it, using June 2025 as the base month and June 2026 as the endpoint. The fourteen-month collection period and the twelve-month comparison are not the same thing, and the release does not reconcile them.

Scale matters here for interpretation. Eleven billion visits across roughly fourteen months averages out to something in the order of 786 million visits a month. Against that denominator, 2.6 million monthly ChatGPT referrals works out to roughly a third of one percent of measured traffic. The arithmetic is approximate, since the release does not publish a monthly total visit figure alongside the referral figure, and it is not a number Demandbase itself states. But the order of magnitude is consistent with what other measurement firms have found. Research published in February 2026 concluded that ChatGPT sends approximately 190 times less traffic to websites than Google despite handling around 12% of Google's query volume.

Percentage growth on a small base produces large-looking numbers. A move from 645,000 to 2.6 million is a real gain of nearly two million monthly visits, distributed across 1,584 customer instances. Spread evenly, that is fewer than 1,700 additional ChatGPT-referred visits per tenant per month, though referral traffic is almost never distributed evenly and the release gives no dispersion data, no median, and no indication of how concentrated the gain is among the largest properties.

Where the growth is not happening

The platform breakdown is the more consequential finding, and Demandbase supplies it without figures. According to the company, ChatGPT accounts for the overwhelming majority of AI-referred traffic to the properties it measures, while Perplexity referrals declined and both Gemini and Claude remained flat. No absolute volumes, percentages, or shares are attached to any of those three platforms.

That asymmetry runs against the direction of travel in general web traffic to the assistants themselves. Similarweb data published in June 2026 put ChatGPT at 52.7% of worldwide generative AI website traffic, down from 76.4% twelve months earlier, with Claude tripling its share and Perplexity drifting from 1.8% to 1.3% over the same stretch. A month later, further Similarweb analysis showed ChatGPT's share of standalone chatbot web visits falling to just above 50% by May 2026, with Gemini, Claude, Perplexity and DeepSeek absorbing the difference.

Two measurements can both hold. Visits to an assistant's own website and outbound referrals from that assistant to third-party sites are different quantities, governed by different product decisions. An assistant that answers in full and links rarely will register high usage and low referral output. The gap between the two figures is precisely the territory that AI visibility vendors have grown up to occupy, and it is territory where measurement remains contested: an IAB framework published in early August 2026 found that only 16% of brands track AI visibility at all, while more than twenty vendors sell tools that disagree with one another about the same brand.

The May 2026 question

Demandbase places its sharpest jump in May 2026. Similarweb's 2026 landscape research separately documented a shift toward homepage referrals in ChatGPT output after May 7, 2026. Whether those two observations describe the same underlying change is not established by either dataset, and neither company links them. What can be said is that referral behaviour from a single dominant assistant has changed abruptly more than once in two years. In August 2025, Profound research found ChatGPT referral traffic dropping 52% from late July as OpenAI reweighted citations toward Reddit and Wikipedia over branded content.

A channel that can halve or double within weeks on the basis of undisclosed ranking changes behaves less like a distribution channel and more like a dependency.

The advertising half of the release

Alongside the referral data, Demandbase published spending figures drawn from its own advertising platform. The company says its Labs advertising analysis covers more than $123 million in media expenditures across 75,645 campaigns, an average of a little over sixteen hundred dollars per campaign.

Between June 2025 and June 2026, connected television spending on the platform grew 112%, while display advertising grew 55%. CTV share of overall advertising investment moved from 1.53% to 2.08%.

Those three numbers constrain each other in a useful way. If connected television spending more than doubled while its share of the pool rose only from 1.53% to 2.08%, a relative increase of about 36%, then the pool itself must have expanded by roughly 56% over the year. Display's 55% growth accounts for very nearly all of that. In other words, connected television is growing faster in percentage terms, but display is still the channel doing the arithmetic work. At 2.08%, CTV remains under one fiftieth of measured advertising investment on the Demandbase platform. The remaining 97.92% sits elsewhere.

That proportion is far below the broader market. Industry projections have placed connected television at 43% of United States television advertising budgets in 2026, and the IAB forecast published in January 2026 put CTV growth at 13.8% for the year against 9.5% for United States advertising overall. B2B is a late arrival to the format. MNTN's July 2025 partnership with ZoomInfo was framed at the time as opening streaming inventory to business advertisers, and Microsoft Monetize spent much of 2026 attaching LinkedIn professional data to CTV inventory across three demand-side platforms. Demandbase's own figures suggest the category is starting from a very low base indeed.

Display's 55% growth is the more surprising line. AdRoll reported in March 2026 that retargeting CPMs had risen 18% while display prospecting prices fell 11%, a split suggesting demand migrating out of upper-funnel web display rather than into it. Demandbase's data points the other way, at least within account-based programmes. The release does not break the 55% into prospecting and retargeting components, so the two findings cannot be squared directly.

The blind spot argument

Demandbase frames the referral growth as evidence of a measurement problem rather than a distribution opportunity.

"AI assistants are becoming an increasingly important way B2B buyers discover and evaluate companies, but much of that activity happens before a buyer ever reaches a brand's website," said Rachel Truair, Chief Marketing Officer at Demandbase. "That makes it more important than ever for marketers to reinforce their brand across the channels they control - from advertising and content to sales engagement - to build trust and deliver a consistent buyer experience."

The claim about pre-website activity has independent support. LinkedIn research published in June 2026 found that 94% of B2B buying groups use large language models before contacting a sales team, and argued that visibility inside AI retrieval systems has become a buying-stage variable rather than a post-purchase one. Dreamdata benchmarks published in March 2026 placed the average B2B buying journey at 272 days across ten stakeholders and 88 touchpoints, with the large majority of that activity occurring before a pipeline record exists. Brainlabs analysis covered in June 2026 described the same effect from the content side: buying groups arriving at vendor conversations having already run the comparison themselves.

A vendor selling advertising and account intelligence has an obvious interest in the conclusion that owned and paid channels deserve more investment when discovery moves off-site. That does not make the underlying observation wrong. It does mean the inference from the data to the recommendation is the company's, not the data's.

Conflicting evidence on what AI referrals are worth

Volume growth says nothing about value, and the published evidence on the quality of AI-referred traffic remains unresolved.

Invoca's benchmarks report, published July 14, 2026, found ChatGPT-referred phone calls converting to sales leads at 49%, the highest rate of any channel it measured and roughly ten percentage points above the cross-channel average. Ahrefs research found AI search visitors converting at 23 times the rate of conventional organic traffic. Microsoft Clarity analysis across more than 1,200 publisher and news sites put AI-referred conversion at three times traditional channels. Set against those figures, a study of 973 e-commerce sites with $20 billion in combined revenue found ChatGPT referrals underperforming traditional channels on both conversion rate and revenue per session. PPC Land's July 2026 review of that literature noted the spread runs from 3x to 23x depending on methodology and site category.

There is also a measurement question underneath the measurement question. Similarweb research published in June 2026 found that AI-recommended brands were 2.5 times more likely to receive a site visit within seven days, but that only 8.8% of AI-influenced visits arrived as a direct referral. Roughly 56% arrived through branded search instead. Referral counts of the kind Demandbase published capture the first category and miss the second entirely.

The plumbing has been catching up. Google Analytics added a dedicated AI Assistant channel to its Default Channel Group in May 2026, assigning an ai-assistant value to the medium dimension automatically when a recognised assistant referrer is detected. Before that, tracking required manually configured custom channel groups, and a meaningful share of AI-sourced visits landed in analytics platforms as direct traffic.

What the release does not say

Several gaps are worth recording. Demandbase does not disclose the absolute size of its non-ChatGPT AI referral volumes, so the words declined and flat carry no magnitude. It does not publish AI referrals as a share of total measured visits. It does not state whether the 1,584 tenants in the dataset were constant across the fourteen-month window, which matters when the headline metric is a raw volume comparison rather than a per-tenant one. It does not disclose the industry mix of those tenants, and it does not separate the advertising figures by customer segment.

The company also does not say what caused the May 2026 jump. In a channel where a single ranking change wiped out half of ChatGPT's outbound referrals in July 2025, that is not a trivial omission.

Why this matters for marketing

For B2B marketers, the practical significance of the Demandbase numbers is less the 303% than the composition behind it. Growth concentrated in one assistant means exposure to one company's product decisions. Independent research surveying 45 generative engine optimisation studies concluded in July 2026 that body-only page rewrites can lower a page's presence in AI retrieval by 16%, and that commercial promises in the category have outrun the evidence. A separate Fractl survey of 343 United States marketing decision-makers found 81% rejecting the GEO label as a distinct discipline.

For media buyers, the CTV and display figures describe a portfolio still overwhelmingly weighted toward display, with connected television growing quickly from a base too small to change the shape of a budget. The 0.55 percentage point move in CTV share over twelve months is the honest version of the more than doubled headline.

For publishers and vendors selling into this market, the Demandbase release joins a growing body of first-party platform data being published as marketing research. The figures are not independently audited, the methodology is described rather than documented, and the dataset is proprietary. Those are the standard conditions of vendor benchmark publishing, and they set the limits within which the figures carry weight.

Timeline

Summary

Who: Demandbase, the San Francisco account-based go-to-market platform, published the data through its Labs research arm. Rachel Truair, Chief Marketing Officer, provided the accompanying commentary. The findings concern OpenAI's ChatGPT, Perplexity, Google's Gemini and Anthropic's Claude, and affect B2B marketers, media buyers and demand generation teams.

What: Monthly ChatGPT referral visits to the B2B web properties Demandbase measures rose from roughly 645,000 in June 2025 to 2.6 million in June 2026, a 303% increase, with a sharp inflection in May 2026. Perplexity referrals declined while Gemini and Claude stayed flat, though no figures were given for those three. Advertising data from the same platform showed connected television spending up 112% and display up 55% year over year, with CTV share of investment rising from 1.53% to 2.08%.

When: The release was issued on Wednesday, August 12, 2026, at 9:00 a.m. Eastern. The underlying data covers June 1, 2025 to July 31, 2026, with the headline comparison running from June 2025 to June 2026.

Where: The company is headquartered in San Francisco. The dataset spans 1,584 Demandbase platform instances and more than 11 billion website visits, with no geographic breakdown published.

Why: Demandbase argues that AI-driven discovery is moving evaluation activity off brand websites, creating a gap in the measurable buyer journey, and that marketers are responding by investing in channels they control. The referral figures are consistent with other measurements showing AI assistant traffic growing quickly while remaining a small fraction of total volume, and with research showing that a majority of AI-influenced visits never register as AI referrals at all.