What the AI-Made Ads That Worked Had in Common

AI-made ads are no longer festival curiosities. One aired during the 2025 NBA Finals, several ran during the 2026 Super Bowl, and a few were torn apart within hours of release. After the Coca-Cola and McDonald’s holiday spots, a brand weighing its own AI ad has a larger sample to learn from.

Six cases from 2024 to 2026, and the audience research around them, point to one pattern. The ads that worked chose generated video because it served the idea and kept it away from what it renders badly. The ones that failed leaned on the novelty, asked the model to carry a human face through an emotional story, or sat on a brand whose value is visible craft.

An absurd brief, made for the medium

In June 2025, the prediction market Kalshi aired a 30-second AI-generated spot during Game 3 of the NBA Finals. NPR reported that P.J. Accetturo, an advertising veteran, made it in two days, with AI prompting costs under $2,000; his own fee was not disclosed. The ad strings together absurd scenes, from a bride fleeing police on a golf cart to a farmer in a pool full of eggs, each character calling a Finals pick.

The brief came first. Accetturo said Kalshi hired him to make “the most unhinged NBA Finals commercial possible.” That tone suits generated video. In a deliberately absurd world, an odd gesture or a too-glossy face reads as part of the joke, so the flaws a model still produces stop costing credibility. The labour was in the selection: 300 to 400 generations to get 15 usable clips. The ad passed 3 million views on Kalshi’s X account within a week, and the analyst Debra Aho Williamson told NPR that the AI angle itself was drawing attention.

That attention came from being early. A brand that copies the format a year later inherits the look of an AI ad without the surprise.

Generated where the model is strong, filmed where it is weak

Microsoft took a quieter route with a one-minute ad for its Surface Pro and Surface Laptop business computers. On the Microsoft Design blog, the team explained that generated shots went where cuts were quick or motion was limited, and that top-down framing helped keep the model from inventing errors. “Any intricate movement such as closeups of hands typing on keyboards had to be shot live,” the post says. The team reported thousands of prompts and estimated it saved about 90% of the usual time and cost, a figure from the company itself.

PetaPixel reported that the ad had been online for about three months, with roughly 50,000 YouTube views, before Microsoft revealed its use of generative AI, and that viewers had not noticed. The method sits in the shot list. Someone decided, shot by shot, what a model could render convincingly, and the edit kept generated shots short enough that drift within a clip had no time to show. Hands, the product in use and anything a viewer would study closely stayed on camera.

When the story rests on a face

Toys“R”Us premiered a brand film made almost entirely with generated video at Cannes Lions in June 2024. Produced by Toys“R”Us Studios and the agency Native Foreign, it tells the origin story of founder Charles Lazarus, with Geoffrey the giraffe appearing in a dream. Decrypt reported that the team narrowed hundreds of iterative shots to a couple dozen and added corrective visual effects, and WHP Global says the film carries an original score by the musician Aaron Marsh.

The reaction from the creative community was harsh; one author quoted by Decrypt called it “an abomination.” An AI video maker pointed to trouble with hands and with keeping a character’s appearance consistent from scene to scene. A sentimental origin story asks viewers to bond with one face, and separately generated shots are where faces drift. A human score could not fix that.

Dunkin’ ran into a version of the same wall at the 2026 Super Bowl. Its ’90s sitcom spot de-aged a celebrity cast, and Adweek critic Audrey Kemp, who attributed the effect to AI, found the result “more uncanny than comforting.” Meltwater’s analysis of social conversation on February 8 and 9 found Dunkin’ took 37% of the discussion about AI slop in the game’s ads, with 42% negative sentiment. Viewers know those famous faces, which makes any synthetic smoothing easier to spot.

Disclosure does not rescue a poor fit

Valentino labelled its late-2025 video for the DeVain handbag as AI-generated and described it as a digital creative project made with digital artists. Glossy reported hundreds of critical comments calling it “cheap” and “lazy.” The labelling was honest. The trouble was the match between a luxury house and a medium many viewers associate with low effort.

Research backs that reading. In three experiments published in the Journal of Advertising Research in 2025, R. N. To and colleagues found that revealing AI use in luxury ads made consumers judge the brand less favourably, because the ads seemed to require less effort and felt less authentic. The penalty shrank significantly when the generated visuals were highly creative, beyond what traditional methods could easily produce.

Svedka made the AI the point of its Super Bowl spot. TechCrunch reported that the 30-second ad, built with the AI studio behind Coca-Cola’s holiday ads, was primarily AI-generated, that rebuilding the brand’s Fembot character and teaching the model her expressions and movement took about four months, and that people wrote the storyline. Meltwater found Svedka drew 63% of the engagement around AI-related commercials, with largely neutral sentiment. Openness bought attention, without much warmth.

What audience research measures

NielsenIQ tested AI-generated ads of varying quality on more than 2,000 people, measuring brain activity in about 150 of them, and published the results in December 2024. Viewers rated the AI ads less engaging than conventional ones, and the ads produced weaker memory activation regardless of how good they looked. Low-quality visuals forced viewers to work harder to process the ad, which pulled attention from the message.

DoubleVerify’s July 2026 study of 22,000 consumers in 22 markets adds the quality split. Around the Super Bowl, Meltwater found that about 6% of ad mentions concerned AI, and close to half of those were negative, mostly calling AI ads uninspired or low quality.

Quality decides the verdict

Share of 22,000 consumers in 22 markets, July 2026

Consumer views of AI ads, DoubleVerify, July 2026

Statement	Share of consumers
A low-quality or uncanny AI ad would hurt their opinion of a brand	42%
View polished, professional AI ads positively	40%
Cannot consistently identify AI-generated content	56%

Read together, most viewers cannot reliably tell when an ad is generated, yet a visibly poor one costs the brand. The risk sits in the weak shots, which is why the shot plan matters more than the share of AI in the cut.

Before commissioning an AI ad

• Ask why this idea needs generated video. If the only answers are cost and novelty, expect viewers to sense it.
• Check the fit with your brand. A heritage or luxury brand pays more for perceived low effort; an irreverent brand can turn a model’s oddities into jokes.
• Ask for a shot plan that marks each shot as generated, filmed or composited, with a reason. Hands, your product in use and any recurring face need a clear answer.
• Budget the edit and the sound as you would on a filmed spot, and keep generated shots as short as the story allows.
• Decide disclosure before production, following platform rules and local law, and test the cut with real viewers before launch, checking what they remember about the brand.