A TransUnion survey puts numbers on the gap between AI confidence and AI measurement

Marketers have never been more sure that AI is working for them, and have rarely been less able to show it. That is the finding underneath a TransUnion study published this week, and it bears directly on how search teams should be talking about visibility in AI answers.

The research, carried out for TransUnion by United Talent Agency's brand advisory arm, surveyed a hundred senior marketing and technology leaders at large American companies. TransUnion's own name for the result is a confidence-readiness paradox, and the numbers earn it. Eighty-nine per cent expect to put more money into AI-enabled marketing over the next one to two years, and 64% say they are confident of meeting their AI goals. Yet only 42% rate their people as highly ready, and 36% say the same of their data and processes. Fewer than half, at 48%, believe they can see enough of what platforms are doing with AI to optimise against it with any confidence.

Asked to rank those weaknesses, Matt Spiegel, who runs TruAudience growth strategy at TransUnion, put data ahead of the others without hesitation. Feed a system incomplete information, his argument goes, and no amount of capability elsewhere will make the output predictive. Skills and governance problems sit downstream in his reading: connect the customer records, the transaction history and the behavioural signals first, and the rest becomes tractable. Leave them disconnected and every other fix is harder than it needs to be.

What turns this from an operations story into a search story is what respondents blamed. Sixty-nine per cent pointed at walled gardens they cannot see into; 70% at blind spots between channels that stop them tracing AI's effect across a customer journey. Spiegel's account is that no single environment shows the whole path, because campaigns now run across connected TV, retail media, paid social, search and owned channels at once.

Pressed on whether visibility in AI answers is a version of that problem or a separate one, he treated it as the same thing: an extension of the walled garden, and another argument for measurement that does not originate with the platform being measured. People increasingly meet brands through a generated answer rather than a page of links, and nobody outside the system can see what shaped the recommendation.

That is worth sitting with, because every study currently attempting to map AI citations is trying to reconstruct the workings of something never designed to be inspected from outside. Here the same blindness is described by a hundred marketing leaders with no stake whatever in how the SEO industry settles its internal arguments. An unreadable retail media auction and an unreadable AI citation are one missing instrument pointed in two directions.

The practical consequence is about positioning. Treating AI search visibility as a discipline of its own, with its own tooling budget, puts it in competition with everything else on the list. Treating it as the search-shaped piece of a measurement failure the chief marketing officer is already raising in board meetings puts it alongside a problem that already has executive attention. The second framing gets funded faster, and it happens to be the more accurate one.

Three things follow from that. Check whether the customer data already inside the business is joined up enough to make an AI visibility tool worth buying — Spiegel's advice for a fresh AI budget was to spend first on data quality and identity resolution rather than on another application, and a citation tracker sitting on top of fragmented analytics will report activity and call it impact.

Build incrementality testing in now, while it is still a differentiator rather than an expectation. Spiegel notes that only a minority of marketers apply mix modelling or incrementality tests to AI at all, largely because time saved is easier to report than value created.

And measure lift rather than counts. A team that can tie traffic or conversions to specific AI citations will survive the scrutiny that a team counting citations will not.

Spiegel's closing observation is the sharpest thing in the study, and it is not in the press release. The costly mistake he sees is not a poor AI strategy but the question that precedes it — leaders asking what their AI strategy is, instead of asking which business problem they are trying to solve. The search version writes itself. Chasing citations without first deciding what those citations are supposed to produce leaves you with visibility and no way to argue that it mattered.