Two findings about AI search traffic circulated in the search trade press on September 28, 2026, and at first glance they seem to contradict each other. A court filing from news publishers suing OpenAI and Microsoft cites Microsoft data showing click-through rates between 51% and 94% lower in Bing Chat than in Bing web search, depending on the publisher. Meanwhile, a Profound study of more than two million AI conversations found that visits to a brand's website ran above their expected rate in the week after an AI assistant mentioned the brand.
Both findings can be accurate because they measure different moments in the buyer's path: the click on the answer itself and the subsequent visit. For marketing leaders, the practical risk is that a report based on referral traffic will capture the drop in clicks but miss the subsequent visits. In this piece, we look at what each source shows, where each one has limits, and how we measure AI search for our clients so that neither trend gets lost.
What the Publishers' Court Filing Shows About AI Search Clicks
The filing argues that AI answers reduce clicks to the sources they draw from and cites statements from OpenAI employees and Microsoft data as evidence. Publishers including The New York Times, the Daily News, and Ziff Davis submitted the combined summary judgment brief in federal court in New York on September 17, 2026.
According to Search Engine Journal's coverage of the filing, the brief quotes an OpenAI software engineer saying that "no matter how prominently we show the links, users won't click," a message The New York Times, one of the plaintiffs, reported was written in February 2023. The brief also states that Nick Turley, OpenAI's head of ChatGPT, said, in the context of ChatGPT's browsing function, that there was "no good reason to click" a link to the original source once ChatGPT gave an answer. The Microsoft data compares click-through rates in Bing Chat with those in Bing web search. The reported reductions vary by publisher: 87% to 93% for The Times' websites, 83% to 91% for the Daily News sites, and 51% to 94% for Ziff Davis sites.
This evidence deserves careful handling, because the brief represents one side's argument in a case the court has not decided. OpenAI and Microsoft argue that their use of news content is fair use. Search Engine Journal also notes that the brief leaves out much of what an analyst would want to know: it does not say when Microsoft collected its data, how many queries or impressions it covered, how the data was selected, or what the low and high ends of each range represent. Microsoft renamed Bing Chat to Copilot in November 2023, but Search Engine Journal notes that the product name does not date the data and that nothing in the passage ties the figures to the current Copilot. The filing therefore does not show how today's products perform. The direction of the finding is intuitive all the same, since an answer that fully resolves a question gives the reader less reason to leave it.
What Happens After an AI Assistant Mentions a Brand
Profound's research found that an AI mention is followed by more visits to the brand's website and that most first visits arrive more than a day later. In ChatGPT's case, more than 97% of those visits carry no UTM that ties them back to AI. Profound, which sells AI visibility tracking software, published the study, titled "The AI Mention Effect," on July 1, 2026, and Search Engine Land cited it in a September 28 article debunking common AI search myths.
The study analyzed more than two million US conversations across ChatGPT, Gemini, and Google AI Overviews from January through June 2026, using a double opt-in panel that recorded both AI interactions and later browsing. It counted a mention only when the brand appeared in the AI response and was not named in the user's prompt. It compared the rate of brand site visits in the seven days following a mention with a forecast based on those same users' visits to the same brands across three seven-day windows before the mention. The study reported three main findings:
- Visit lift: Mentions were followed by brand site visits at roughly 1.5 to 2.5 times the forecasted baseline rate over the following seven days. The lift was about 1.5 times for ChatGPT, with 6.39% of mentions followed by a visit versus a forecasted 4.33%; about 1.6 times for Google AI Overviews; and about 2.5 times for Gemini.
- Timing: About 20.5% of first visits to a mentioned brand occurred within an hour, and 42% within 24 hours, indicating that most first visits occurred after the first day.
- Tracking: For ChatGPT, more than 97% of brand site visits after a mention carried no UTM parameter, and the tagged share, which sat around 1% from January through April, rose to 2.47% in June 2026, a partial month that Profound treats as directional.
Profound is candid about the limits of its work, noting that the study is not a randomized experiment and therefore cannot fully rule out selection bias, and that it tracks site visits but does not follow them through to purchases. Because Profound sells tools built on the premise that AI mentions matter, its conclusions also warrant the same scrutiny we would give any vendor research, although forecasting each user's baseline from their own earlier behavior is a more careful method than a simple before-and-after count.
AI-influenced visit: A visit to a brand's website that follows exposure to the brand in an AI-generated answer may have been prompted by it and carries no AI campaign tag. When the person arrives later via search or a typed URL, the visit also has no AI referrer, so analytics tends to record it as direct traffic or organic search, often from branded queries.
Why Both AI Search Findings Can Be True at Once
The court filing measures clicks from an AI answer to its sources, while Profound measures visits to the mentioned brands in the following week, so a website can lose the first type of traffic and gain the second in the same month.
| Publishers' Court Filing (Microsoft Data) | Profound AI Mention Study | |
|---|---|---|
| What it measures | Clicks from an AI answer to its sources | Brand site visits within seven days of an AI mention |
| Time window | Immediate click from the answer (data collection dates not disclosed) | Seven days after the mention |
| Headline finding | Click-through 51% to 94% lower than web search, varying by publisher | Visits 1.5 to 2.5 times the forecasted baseline, varying by platform |
| How it shows up in analytics | Referral traffic that never arrives | Mostly untagged in ChatGPT's data, likely recorded as direct or organic, often on branded queries |
| Main limitation | One party's evidence, with undisclosed data dates and sample sizes | Not randomized, and tracks visits without purchases |
The two findings also concern different kinds of websites, which changes what each one means for a business. The filing is about publishers whose content an AI answer summarizes and who depend on the click for revenue, while the Profound study is about brands that an AI answer names and that depend on being chosen. The site an AI assistant cites as a source, and the brand it mentions in the answer, are not always the same, which is why we track citations and mentions separately for every organic client.
For a publisher, a lower click-through rate is a direct revenue problem. For a B2B company, a university, or a service business, what matters most is whether AI assistants name the brand when buyers ask for recommendations. Profound's data associates being named with more brand site visits, and, in ChatGPT's case, nearly all of those visits lack a UTM that ties them to the answer.
Why Referral Reports Undercount AI Search Traffic
Referral reports depend on each visit announcing its source through a referrer or a campaign tag, and many visits that follow an AI mention likely carry neither. Profound's data shows how little the tag side captures, since fewer than 3 in every 100 brand visits after a ChatGPT mention carried a UTM. Not every one of those visits came because of the mention, since Profound's forecasted baseline suggests many would have happened anyway, but untagged visits give a referral report no way to tell which ones did. A referral report shows the AI search traffic that identified itself, which, for tags at least, is a small fraction of all brand visits following a ChatGPT mention.
Our own client work shows a related gap, this time in how revenue gets credited once a visit is recorded. For a regional B2B distributor, we traced 198 orders totaling about $116,000 from May through August 2026 that included at least one visit from ChatGPT, and 49% of that revenue came from orders where ChatGPT was not the final touchpoint before purchase. A last-click report would have credited that 49%, about $56,000 across 121 orders, entirely to Google Ads, organic search, or direct visits. From April through August, between 72% and 84% of ChatGPT-assisted customers each month checked out as guests with no account and no lead record, so self-reported attribution tied to a lead form would not have been captured for them. We walk through that analysis in what last-click attribution misses about AI search.
Those were only the journeys we could see, since each one included a trackable ChatGPT visit. Profound's consumer panel suggests there are likely other buyers who read an AI answer and arrive later in a way no referral report will connect to AI, although its data cannot tell us how many there are for a B2B business.
How to Measure AI Search Traffic Beyond Referrals
Measure AI search across three layers that check one another: how often and how accurately AI assistants mention and cite the brand, what people do on the site after AI appears in their journey, and the revenue those journeys generate. We call this triangulation, and it helps because each layer has blind spots that the other two can partly offset.
- Exposure: We track mentions and citations separately across the assistants a client's buyers actually use, through the AEO Index and GEO Audits, and we add AI visibility checks every two weeks to inform reporting and content recommendations. This is the layer where we look first for signs of change.
- Behavior: We follow what visitors do across every trackable touch in their journey through Galileo, Arcalea's multi-touch attribution platform. We also monitor branded search and direct traffic trends as indirect signals, since untagged visits likely land there.
- Outcome: We connect those journeys to revenue, orders, and pipeline status in the client's CRM, so that visibility and behavior tie back to business goals.
AI referral traffic still belongs in the report, where it serves as a floor showing the minimum AI search traffic we can confirm. When exposure rises while referrals stay flat, the pattern Profound describes offers a possible explanation, and branded search and assisted revenue are the next places to check. When exposure remains low and branded search does not move, flat referrals are likely accurate, and the priority becomes earning more mentions and citations. Our guides to why AI visibility reports are often wrong and how to read an AEO analysis cover the traps in that first layer.
This approach also changes what a monthly report should lead with, because AI referral traffic may stay small or even shrink while mention share (the share of relevant AI answers that name the brand), branded search, and AI-assisted revenue grow.
The Bottom Line
The available evidence suggests AI answers drive fewer clicks than traditional search results, although the clearest data comes from one side of a lawsuit that the court has not yet decided. The brands mentioned in those answers are still drawing visitors, though many arrive a day or more later. In ChatGPT's case, almost all of those visits carry no tag identifying AI as the source. A measurement plan built only on referral traffic will capture the first trend while the second remains hidden, so the first step for most brands is to put instruments in place to see both before deciding on the strategy.
If you want to know how often AI assistants mention your brand and whether that visibility is already showing up in your pipeline, talk to our SEO, AEO, and GEO team about an AEO Index and a Galileo attribution assessment.