News
Most senior marketers have acted on a wrong AI insight, and only a third check what AI tells them
GWI's AI trust gap survey: 58% of senior marketers made a big decision on an AI insight that proved wrong, and only 36% routinely verify AI output.
Fifty-eight percent of senior marketing, insights and agency professionals say they made a significant decision on an AI insight that later proved wrong, according to GWI's AI trust gap survey, released November 10, 2026. Only 36% say they verify AI output most of the time, a gap that matters to any team building briefs and budgets from AI summaries.
Key points
- Audience research company GWI polled 1,883 senior people in marketing, insights and agencies across the UK, the US and Singapore in July 2026, all at companies with 500+ employees or £20m+ in annual turnover.
- 88% of them use AI every day, and 80% say they often see AI findings that are inaccurate, misleading or untrue.
- 82% report at least one cost of bad AI output: most often rework (38%), delayed decisions (36%) and customer complaints (34%), and 27% report financial loss.
- Only 36% check AI-generated insights most of the time (43% in the US, 29% in the UK) and 22% check now and then; a further 36% check one AI tool's answer with another.
- GWI sells survey data collected from real people, and its report concludes that AI insight needs exactly that kind of human research behind it.
Who did GWI survey for the AI trust gap report?
Marketers, not business leaders in general. The release headline speaks of "business leaders", but its methodology note says the 1,883 respondents were senior marketing, insights and agency professionals: 850 in the US, 665 in the UK and 368 in Singapore, surveyed in July 2026 at companies with 500+ employees or £20m+ in turnover. So the findings describe marketing teams like the ones CMO Times readers run. The release does not give a margin of error, the question wording or how respondents were recruited.
How often do marketers check what AI tells them?
About a third do so routinely. GWI reports three figures on how respondents check:
| How respondents check AI-generated insights | Share |
|---|---|
| Verify them most of the time | 36% (43% in the US, 29% in the UK) |
| Check them only occasionally | 22% |
| Cross-check with a different AI tool | 36%, in addition to the groups above |
GWI likens the last group to one machine marking another's homework. Stale data is the other weak point: 63% say they have acted on out-of-date data or insights without knowing it, and 72% agree that AI tools can make dated information seem more relevant than it is.
AI is only as reliable as the data behind it, and most leadership teams already know that. … What they're missing is the discipline to verify real human truths, because AI sounds just as certain when it's wrong as when it's right.
— Tom Smith, Founder and CEO, GWI
Why does AI sound certain when it is wrong?
Because most models still guess rather than admit what they do not know, according to independent testing. The AA-Omniscience benchmark from Artificial Analysis scores models on 6,000 factual questions across 42 economically relevant topics, penalizing wrong answers and scoring a declined question above a wrong one; in its November 2025 paper, only three of the models tested scored above zero, so every other model gave more wrong answers than right ones. GWI's report cites the same benchmark to argue that AI tools are built to produce a plausible answer, not a verified one.
What does the AI trust gap cost a marketing team?
Mostly time, sometimes money. 82% reported at least one negative consequence of inaccurate AI output, led by everyday friction:
| Consequence reported | Share of respondents |
|---|---|
| Internal rework | 38% |
| Delayed decisions | 36% |
| Customer complaints | 34% |
| Financial loss | 27% |
| Reputational damage | 26% |
| Weaker campaign performance | 21% |
GWI argues that because these costs get absorbed into daily work rather than traced to a flawed insight, teams count AI's speed and miss the price of its errors. 85% of respondents rate the risk of deciding on inaccurate AI insight as significant.
What should CMOs do about the AI trust gap?
Our analysis: the survey points at process more than tools. If only about a third verify AI output most of the time, and as many check one model with another, the cheapest fix is a named step before an AI-derived finding reaches a brief, a forecast or a budget: someone confirms the date of the data behind it and traces it to a source. That echoes the case for keeping human judgment in charge of AI-assisted marketing, and follows a recent study of B2B marketing organizations that found AI use widespread but rarely coordinated. Read the findings with their source in mind: these are self-reported answers, gathered by a company whose business is selling the human research it recommends. The full Artificial Confidence report also covers AI stereotyping of audiences and who in an organization answers for a decision that goes wrong.
Frequently asked questions
What is the AI trust gap?
It is GWI's name for the distance between how much senior marketers trust AI for insight and how often that insight holds up. In its survey, 88% use AI daily while 58% have made a significant decision on an AI insight that later proved wrong.
How many marketers verify AI-generated insights?
36% of the senior marketing, insights and agency professionals GWI surveyed say they verify AI-generated insights most of the time. Another 22% check only occasionally, and a further 36% cross-check with a different AI tool.
What does acting on bad AI insight cost?
Respondents most often reported rework (38%), delayed decisions (36%) and customer complaints (34%). 27% reported financial loss and 26% reputational damage.
Who was surveyed for GWI's Artificial Confidence report?
1,883 senior people working in marketing, insights or agencies across the UK, US and Singapore, all at companies with 500+ employees or £20m+ in turnover, surveyed in July 2026.
Sources
- GWI's release on its AI trust gap research: press release from GWI via GlobeNewswire, 10 November 2026.
- Artificial Confidence: The AI Trust Gap: the full report, published by GWI.
- AA-Omniscience: Evaluating Cross-Domain Knowledge Reliability in Large Language Models: the benchmark paper by Artificial Analysis, on arXiv, November 2025.
- GWI: the company's website.