What Brands Should Test Before Launching AI-Created Work

The study points to more than one outcome worth measuring

A lot of AI testing is going to focus on whether an execution performs. Our study suggests that performance is only part of the reaction consumers can have.

Comfort varies depending on the use. Thirty-six percent were comfortable with AI helping compare products, 32% with AI recommendations, 20% with personalized messages, 13% with AI-written brand content, and 10% with AI-generated influencers or spokespeople. That spread shows that the same audience can react very differently depending on what role AI is playing.

Purchase likelihood is another piece of the picture

When we asked about AI-created advertising or messages, 15% said it would make them more likely to buy, 44% said it would make no difference, and 41% said it would make them less likely.

Twenty-nine percent of consumers said brand AI use makes them question whether the information is accurate, while 27% said it makes marketing feel less human. Consumers also associated brand AI use with being lazy, less personal, less authentic, and less trustworthy.

Those findings suggest that an AI execution can affect perceptions of the work itself and the company behind it. A simple like/dislike question would miss a lot of that.

Audience response is uneven

Regular AI shoppers react much more positively to AI-created advertising than everyone else. Forty-two percent of regular AI shoppers said it makes them more likely to buy, compared with only 9% of everyone else.

That difference is one of the strongest reasons to look at AI-shopping behavior when evaluating an execution. The overall average can hide a lot of variation between people who already use AI regularly and people who do not.

What the data suggests testing

Taken together, the study points to a fairly practical set of measures for AI-created work: helpfulness, accuracy, trust, authenticity, brand fit, human connection, perceived effort, and purchase likelihood. It also makes sense to look at those results by AI-shopping behavior rather than only at the total sample.

Those measures come directly out of the patterns we saw in the study. They are not a universal checklist for every AI application, but they give researchers a stronger starting point than simply asking whether people liked the execution.

What stands out in the data

AI can create more than one kind of response at the same time. Consumers may see something as useful but still question its accuracy, feel less human connection, or react differently depending on how familiar they already are with AI.

That is the main reason this topic needs a broader testing lens. The interesting question is not only whether AI "works." It is what changes when AI becomes part of the experience.

Read the full report:

https://www.lab42.com/ai-shopping-report

Source: Lab42 AI Consumer Study, 2026. Q43-Q47, base all consumers n=500.