GWI survey shows AI insights are wrong more often than marketers think

The gap between confidence and accuracy

Nearly 90% of senior marketing professionals use artificial intelligence every day, according to a GWI survey published this week by Marketing Dive. The problem is that 80% of those same professionals say they frequently encounter inaccurate or misleading AI insights.

The survey, titled "Artificial Intelligence: The AI Trust Gap," polled 1,883 senior insight, agency, and marketing professionals in the United Kingdom, the United States, and Singapore between July and September 2026. All of them work at companies with more than 500 employees or annual revenue above £20 million (about $26.5 million), in sectors such as media, agencies, gaming, and retail.

In short: marketing teams use AI to generate audience profiles, segmentation, and behavior analysis, trust the result, but rarely check whether that result is correct. Only 36% of respondents say they verify AI findings "most of the time." That's the core problem described in the survey, and also the most relevant point for whoever decides on martech tools and allocates media budget in Brazil: the shortcut AI promises only actually saves time when someone checks what it delivered.

Broadly, everyone is using AI, everyone feels very confident and good about it, but everyone also has very strong concerns and suspicions, and remarkably high numbers are seeing negative impact or less than good outputs. But we're carrying on using it regardless.

Jason Mander, chief insight officer at GWI

Where AI gets it wrong (and why nobody notices)

The report splits the error into two fronts that every growth marketer recognizes in practice: outdated data disguised as relevant data, and audience profiles built on stereotypes. On the first front, 72% of respondents agree that AI can make something seem more relevant than it actually is, and 63% admit they've used an insight without realizing at the time that it was later found to be outdated. More than half (68%) think their teammates make the same mistake without noticing.

On the second front, the problem is more structural: 76% of leaders believe the technology can reinforce outdated assumptions about who the audience is, and nearly 70% have already seen AI fail to accurately reflect their own company's actual audience. Even so, 81% say they feel confident using AI for audience profiling and segmentation, and 84% believe the technology can accurately explain human behavior. This mismatch between stated confidence and acknowledged error is what GWI calls the trust gap, and it's where the risk lies for anyone who uses that audience profile to decide on creative, channel, and budget.

A more specific finding explains why confidence survives the error: 71% of respondents say AI can tell you the "what," but not the "why" behind a certain audience behavior. The model delivers the statistical pattern, not the motivation behind it. Eighty percent agree that understanding motivation still requires real people at the table, and 75% think AI will never replace direct research with humans. Even so, more than 90% of organizations already build or use AI tools that depend on external data, which means dependence keeps growing despite the caveat.

The bill that comes due later

More than half of the senior decision-makers surveyed have already made a significant business decision based on an AI insight that later turned out to be incorrect or misleading. This isn't an abstract model error: it turns into rework, delays, and, in a good share of cases, money that doesn't come back. Eighty-two percent of marketing leaders felt at least one direct negative effect from this kind of error.

It's worth separating two things the table lumps together: operational effect (rework, delays) and business effect (financial loss, reputation, campaign performance). The first group is process cost, it hurts the team's hours budget. The second is the cost of a wrong decision that's already been published, and that's what actually worries whoever is responsible for media budget: a live campaign with segmentation based on an audience profile that the report itself shows has nearly a 70% chance of not reflecting the real audience.

What this changes for whoever decides on tools and budget

For growth and martech professionals, the GWI report isn't an argument against using AI: 90% of the market already uses it daily, and backing off isn't a realistic option. The point is different: treat AI insight as a hypothesis to validate, not a fact to execute on. In practice, that means:

• Cross-checking AI-generated audience profiles against your own first-party data (CRM, research, actual conversion behavior) before using it to define campaign segmentation
• Treating the "what" AI delivers as a starting point, and reserving budget for qualitative research (interviews, focus groups) to answer the "why" the model doesn't provide
• Defining, in writing, who audits AI output before it becomes a campaign brief, since today only 36% do this "most of the time"
• Measuring the cost of rework and decision delays as a KPI of AI adoption, not just the speed gain, because both sides of the ledger show up in the same report

Jason Mander, chief insight officer at GWI, sums up the diagnosis like this:

I think we've got a little bit lost in all of the benefits because AI…allows you to become so much more productive, save so much time, increase all of your efficiencies. You perhaps haven't scrutinized some of the downsides. But it also suggests that, generally speaking, companies haven't put the type of rigorous frameworks in place that you would expect with a new type of technology.

Jason Mander, chief insight officer at GWI

What the report doesn't resolve

GWI doesn't detail which validation framework works best, nor does it compare tools head-to-head: the study measures perception and reported experience from people who use AI day to day, it doesn't audit any specific model. There's also no breakdown for the Brazilian market: the sample covers only the United Kingdom, the United States, and Singapore, which means an equivalent local number doesn't exist yet.

That doesn't cancel out the warning: agency structure, deadline pressure, and dependence on third-party tools to generate audience profiles are realities that cross borders. The risk described in the report, treating as fact what is only an unverified statistical pattern, applies to any growth team working with generative AI today.