adtestbench Higgsfield
EN

AI news · GWI · AI insights

GWI finds 58% of marketing leaders acted on a wrong AI insight

A survey of senior marketers, researchers and agency staff in three countries finds 88% use AI daily, while 36% check its insights most of the time.

By Alexandre S. , 01:45 UTC

Most marketing teams now run audience research through AI, and most of them keep running into wrong answers from it. Of the 1,883 people GWI questioned for Artificial Confidence: The AI Trust Gap, 58% say they took a significant decision on an AI insight that later proved wrong. Daily AI use stands at 88%, and 80% often run into misleading findings. Respondents hold senior roles in marketing, research and agencies across the UK, the US and Singapore.

Retro-futurist illustration: a striped sunset over a grid horizon under a starry sky, with a software window standing on the horizon.
Drawn by adtestbench from “Artificial Confidence: The AI Trust Gap”,

Checking has not kept up. Only 36% verify AI-generated insights most of the time, GWI says, and 36% check one AI tool’s answer with a different AI tool. Out-of-date material is a second problem: 63% say they have used figures they only later learned were stale.

The cost shows up in the results. Of those surveyed, 27% report a financial loss and 21% weaker campaign performance. Marketing Dive, which covered the report on 13 November 2026, notes that respondents work at companies with more than 500 staff or about $26.5 million a year in turnover, while GWI’s own methodology note dates the fieldwork to the second quarter of 2026.

Ad research tools sell this kind of insight by plan. On our price sheet, Atria’s Core plan, $159 a month, follows 50 brands with no AI insights; Plus, $429 a month, follows 100 brands with them (plan prices read 23 October 2026). More tools of this kind are in our creative research hub.

Atria puts AI insights on its $429 Plus plan, and GWI’s figures leave the checking to the buyer: only 36% of the leaders surveyed check such insights most of the time.