81% Say Their AI Search Bet Is Paying Off. 59% Can't Read the Data.

Scrunch and Scribewise surveyed 602 US marketing and PR professionals between May 19 and June 2. Two findings from that survey sit about as far apart as two findings can.

81% said their organization can clearly connect its AI search work to business impact.

59% said they can’t confidently read the data they’re collecting and turn it into a next step.

Those are the same people. The survey, which carries a margin of error of about 4 points at 95% confidence, also found that 73% are already paying for tools to monitor how their brand shows up in AI answers. So the money is committed, the dashboards are running, and a clear majority of the people watching them say they can’t act on what they see. A minority of that same group is nonetheless reporting proven business impact to somebody.

John Barham, managing partner at the agency Roast, described the environment that produces a number like 81%.

“Every CMO we know has been under excruciating pressure over the last 18 months from their boards and C-suites or investors to crack this nut.”

The tools do not agree with each other

Part of this is not the marketers’ fault. The trade body’s 36-page framework for measuring AI visibility, out earlier this month, contains one admission that explains a great deal: more than 20 companies now sell these tools using different methods that produce different answers for the same brand.

We covered what that standard asks of buyers when it landed, and why the scorecard was already conflicted in July. The relevant part here is simpler: if two vendors measuring the same brand return different results, at most one of them is right, and no buyer currently has a way to tell which.

There’s a small piece of comedy inside the framework too. It cites 16% as the share of brands systematically tracking AI visibility today, as PPC Land noted, without naming who measured it. A document arguing for measurement discipline is carrying an unsourced number. Set that 16% next to Scrunch’s 73% and the two figures can only be reconciled by deciding what “systematically” means, which is exactly the problem the framework exists to solve.

What the traffic actually shows

There is a real number underneath all this, and it’s worth separating from the rest. Demandbase measured ChatGPT-referred visits to B2B brands rising from roughly 645,000 in June 2025 to 2.6 million in June 2026, a 303% increase over 12 months, according to Digiday’s reporting.

That’s a measured, dated, single-source figure with a stated window, and it’s growing fast. It’s also a count of visits, not a count of sales, and 2.6 million visits spread across the entire B2B universe is a smaller pot than the growth rate makes it sound.

The operators quoted in that piece are all solving the same problem by hand. Mulenga Agley, founder and CEO of Growthcurve, has given up waiting for a vendor to do it:

“The only real way to do it is to model it yourself by putting together your first-party data from pages, excluding what you know to be AI traffic.”

Alicia Yoon, founder and CEO of the skincare brand Peach & Lily, is blunter about the ceiling on any of this:

“So much [conversion data] is not directly captured,” she said. “There’s going to be a gap.”

Why the confidence is running ahead of the evidence

The mechanism is ordinary and it has nothing to do with AI. A CMO under 18 months of board pressure to have an answer on AI search will find one, and the honest version of that answer is uncomfortable to say in a quarterly review. “We bought a tool, we watch it, and we can’t yet connect it to revenue” is a true sentence that sounds like an admission of failure. “We can connect our AI search work to business impact” is a sentence that ends the conversation.

Barham’s other line names the gap between those two sentences:

“There is no one tool out there that can paint you a picture of the universe.”

Nothing here says the spend is wrong. Showing up in an AI answer is worth something, the referral traffic is real and climbing, and we’ve reported on how quickly those citations decay, which is its own argument for tracking them. The problem is the reporting layer. A category where the tools disagree, the benchmark is unsourced, and the practitioners are building their own models in spreadsheets is a category where the confident answer and the accurate answer have come apart.

Read the 81% as a pressure reading. When the same survey shows 59% can’t interpret their own dashboards, the useful move for any marketing leader is to work out which of those two numbers describes their own team, and to be the one who works it out rather than the board.

Related coverage: the IAB’s new AI visibility standard, AI citations have a half-life measured in weeks, and why AI visibility is not referral traffic.