Meta's artificial intelligence (AI) ad-generation tool, which it has encouraged advertisers to use, has been found to repeatedly produce errors, such as arbitrarily changing product shapes and the people who appear.
U.S. business outlet Business Insider reported on the 13th (local time) that Meta's AI ad tool has repeatedly made text in ad images hard to read or generated products that look different from the real ones.
U.S. outdoor retailer "Ray" claimed that an ad placed through Meta's ad tool showed a baffling image of a bicycle with two handlebars attached to both the front and back.
There were also cases where an image of a one-piece pajama product was switched to a shirt-and-pants set without the advertiser's consent, or a woman appearing in the ad was changed to a man.
Advertisers say Meta is automatically applying the AI feature during ad production even though they did not indicate they would use it. A bookstore brand said that after it complained to Meta's customer center, the representative described it as a "one-time error" and submitted a request for a refund of the ad spend.
Advertisers say they have to check every time they run an ad to see whether the AI feature is enabled. Ad agencies and others say that because they run hundreds to thousands of ads at once, checking the settings creates a significant workload.
Robert Webster, chief executive officer (CEO) of ad firm TAU Marketing, said, "The default settings (of Meta's AI tool) are aggressive and it's easy to miss the toggle," adding, "In reality, most advertisers cannot withdraw their social media ad contracts, so the corporations (Meta) can reap the benefits (of using the AI tool) without any sanctions."
Meta said of AI ads, "AI can make mistakes, and it is the advertiser's responsibility to review AI-generated outcomes."