The number of advertising creatives that use artificial intelligence in some part of the creation process has exploded in recent years as the technology matures. However, as AI becomes more commonplace, a discussion around the need for AI disclosures and labeling has intensified. As a result, many states have passed laws requiring disclosure when AI is used in certain ways. The European Union also requires the labeling of select AI-generated content under the AI act.

While some question if AI disclosure will impact ad performance, a recent study by MediaScience found that AI disclosure did not negatively impact ad performance across key metrics including brand recognition, consumer sentiment toward the ad and brand attitude. 

“What we discovered is consumers are accepting of AI if they know,” said Duane Varan, founder and CEO of MediaScience. “A lot of governments are now considering regulating and requiring AI labeling, but there's a question about how that labeling should be done.”

The “AI Labeling Impact Study” was conducted by MediaScience, an advertising research company, in conjunction with MediaPet, its AI video content platform, and the Ehrenberg-Bass Institute for Marketing Science at Adelaide University in Adelaide, Australia. 

Researchers evaluated four labeling methods, meant to mimic the frameworks under consideration by EU and U.S. legislators, across 900 U.S. participants. The four labeling methods were a text label in the first three seconds, a text label from seconds four through six, a text label for the entire duration of the ad and an icon displayed for the entire ad. There was also an unlabeled control group.

The AI labeling effect

When it came to knowing if the ad was AI generated, continuous text proved to be the most effective: Nearly half of respondents, 49%, were able to realize the ad was generated using AI. The control group scored the worst at 36%, and those who were shown an ad with an AI icon scored just two points higher at 38%. The group who received a text disclosure in the first three seconds had an AI awareness score of 46%, compared to 40% of those who received disclosure text later on in the ad. 

In terms of unaided brand recall, there was only a seven point difference across the five study groups, with certain AI disclosure methods actually assisting with recall. Both the control and the AI icon groups had an unaided brand recall rate of 54%, while the four through six seconds and continuous labeling groups each had a 61% recall rate. The rate for the group who saw a label in the first three seconds was 60%. 

When it came to brand recognition, there was a five point differential. The continuous labeling group had the lowest recognition rate of 87%, while the first three seconds group had the highest score at 92%. The control, four through six second and icon groups had recognition rates of 88%, 91% and 91% respectively. 

An AI icon elicited the worst brand attitude score, with just 44% of participants saying they had a positive attitude toward the brand. The control group and those who saw a label during the first three seconds had the highest brand attitude scores of 51%. There was a slight dip to 49% for the four through six second group and the continuous labeling group.

The icon group also liked the ad the least, with 56% of participants saying they enjoyed it compared to 63% of the control group. Those who saw a text disclosure liked the ad the most. Seventy percent of both groups who saw text for just a portion of the ad and 69% of the continuous text group liked it.

“Labeling at the end of the day really is actually a win-win proposition, so it's not a problem for the advertiser, provided the ad is good,” said Varan.

When to label

When and how to label an ad as AI generated is also a question the industry has to grapple with as more laws take effect. New York’s law, which took effect in early June, only requires disclosure in the case of “synthetic performers,” meaning AI that is meant to look like a real person.

The report found that consumers largely agree with this, with 60% of the study population indicating a disclosure should be used when AI is used to simulate humans. However, that falls 14 percentage points when it comes to animals, dropping to 46%. Less than half, 45%, of participants said disclosure is needed when AI is used for product placement or voices. When it came to coloring and lighting, just 21% of participants thought AI labeling was necessary. Other use cases measured include animation (41%), background (35%), script writing (33%), music (28%), text/subtitles (24%) and translation/dubbing (23%). 

A key takeaway for marketers is that consumer preference may not line-up with actual AI awareness. AI icons are the favored method of disclosure by 13 percentage points, but that method had the lowest awareness rate of the four tested at just 38%, compared to continuous disclosure, which came in at the top (49%). 

“The problem is that the AI icon doesn't actually increase awareness of the AI content, so for consumers that's a loss,” said Varan. “This is probably something that would be remedied [by] educating people about what the icon means. But to start, you really need those texts, and unfortunately, having it appear later in the body was not really a viable option or alternative.”