A Pew Research Center survey of 5,119 U.S. adults, fielded in February, found that about half of respondents now use AI chatbots, up from a third in 2024, while roughly a quarter use the tools daily. Pew itself notes the 2024 question was asked differently, so the jump should be read as directional rather than exact.

The adoption curve is steep enough to outpace most annual marketing calendars. ChatGPT alone moved from 18% of U.S. adults in 2023 to 44% in 2026, the same Pew survey reported. Against that background, Google added an "AI Assistant" channel to GA4 in May and Microsoft shipped an "AI Performance" report inside Bing Webmaster Tools in February, giving site owners their first officially supported look at how often their pages surface in Copilot and Bing's AI answers.

Both reports, however, measure the tail of a decision — a click that happened or a citation that was displayed — not the influence that shaped the buyer before the click occurred.

What does the buyer arrive carrying?

Buyers now arrive mid-decision, not at the start of one. Duane Forrester, who sells a data measurement platform, walked into a truck dealership last week after feeding a VIN into an AI assistant. The tool confirmed the trim, reconstructed the truck's listing history across "every site it had ever been listed on," and reported that the vehicle "has sat for 280 days and has already dropped $5,000."

"Nobody did this to dealers," Forrester wrote. "It is not a plot; it is a pace difference."

The information asymmetry between buyer and seller, the central mechanic of retail, has compressed from "a week of evenings" into the length of a coffee. Forrester used the generated brief to ask technical questions — "was that transmission replaced or just reprogrammed" — the salesperson could not field.

Where does the shortlist actually get written?

The shortlist used to live on a search results page. Forrester argued it now lives inside the assistant, where "three names and a paragraph" do the work once done by ten blue links and a long tail.

For established brands, the shift is uncomfortable. Fifteen years of customer history, a working search footprint, and industry recognition "do not count the way they used to" against the evidence a model uses to decide whether a name belongs in a category answer. The companies asking "why not us," Forrester wrote, usually cannot say what is being said about them in their place.

Is the catalog being read or visited?

The category-page playbook — product by use case, service by city, course by technology — was built for a system that listed pages. The assistant reads forum threads, bulletin boards, and spec sheets directly and returns two paragraphs.

"Being absent from what the models read is worse than being absorbed by it," Forrester wrote. The funding logic for those catalogs still reports in clicks, which is the reason it never gets questioned.

What does Google and Microsoft's new reporting actually measure?

Google's GA4 AI Assistant channel and Bing's AI Performance report both instrument the click side of AI search. Forrester spent years building tools in that category, so his critique is informed: the reports can show that a citation appeared, not that the citation shaped the reader.

"If you are waiting for assistant referrals to grow until they rival what Google sends you, stop waiting," he wrote. "That is not how these systems work."

The metric that would actually move organizations — the share of category questions answered by an assistant instead of a SERP — does not yet surface in click-based dashboards.

Why does the gap keep widening?

Two clocks are running at different rates. The buyer's clock ran for 18 months between Pew's surveys. The company's clock runs on 12-month planning cycles, with budget lines, owners, measurements, and board stories each taking a cycle to secure. The math, not the marketing team's effort, explains the gap.

Forrester's ordered prescription: know what is being said about you first, decide what belongs in the answer second, decide what to measure third. Tooling comes after the answer is defined, not before.

What to monitor next

Watch the share of category questions answered by an assistant instead of a result page, and the total volume of clicks leaving Google's results as a whole rather than from any one site. Both are reported in the source as moving away from a wait-and-see posture. The catalog's purpose, its consistency across sources, and whether the answer's description of a brand is accurate remain the questions most companies still cannot answer with their own numbers.