From the sandwich-board ads outside neighborhood bodegas to campaigns by multimillion-dollar companies and midterm political races, AI-generated advertising is becoming part of our everyday scenery. But figuring out which images were photographed, which were prompted, and which fall somewhere in between is getting harder. In a September study of more than 500 consumers, only 17 percent correctly identified the AI-generated ad in a lineup of 10 — even though half initially felt confident they could spot it.
So what gives an AI ad away? Often, it’s something small that doesn’t hold up as well as the overall image. These six signs can offer clues.
1. The product looks plausible until its parts stop making sense
An ad can have a believable setting and a perfectly ordinary model, then fall apart around the product. For example, in June, an REI Instagram ad featured a bicycle with an extra set of handlebars growing out of its saddle.
Screenshot contains AI-generated image.
Credit: Screenshot: Reddit/r/REI
According to the retailer, Meta’s AI personalization tool altered a vendor-provided image, turning a real photograph into slop. REI said it took steps to unenroll from the tool.
“While a two-handled bike might be interesting, it is not something you will find in our assortment,” REI told Fast Company.
Other advertisers have encountered similar changes. In July reporting by Business Insider, advertising consultant Jessica Gleim said Meta had suggested an image that turned a client’s pajama dress into a shirt and pants. For another client, a women’s networking group in Montana, the tool proposed adding men. As Gleim told BI, “It’s not usable to help my clients grow their business,” regarding AI-generation.
2. The same character or object looks different between shots
Sometimes, each shot looks plausible on its own. Put them together, though, and the main character — or main vehicle — doesn’t quite stay the same.
Coca-Cola’s 2025 viral AI-generated “Holidays Are Coming” commercial drew criticism for its poor shot consistency, and its harshest critics called it “soulless, bland, and hard to even watch.” Viewers and experts documented changing vehicle design across the truck scenes, including mechanically implausible configurations. (Of course, that didn’t stop Coca-Cola from airing the commercial all winter long.)
The same sort of inconsistency appeared in Toys “R” Us’s 2024 film, made with OpenAI’s Sora (RIP). The film depicts the company’s founder Charles Lazarus as a boy dreaming about a fantastical toy store. But the child’s appearance varied throughout the video.
This Tweet is currently unavailable. It might be loading or has been removed.
Similar inconsistencies can happen within a shot. Last year, UC Berkeley digital forensics expert Hany Farid told Mashable to look for “temporal inconsistencies,” such as “the building added a story, or the car changed colors, things that are physically not possible.” These changes can occur in the background, away from the main subject, making them easier to miss.
3. Movement doesn’t match the object — or the speech
Back to Coca-Cola. In their 2024 AI holiday advertisement, the trucks were moving along roads without their wheels even rotating. Very sloppy.
This Tweet is currently unavailable. It might be loading or has been removed.
McDonald’s Netherlands encountered its own visual problems in an AI-generated Christmas commercial from last year, with distorted faces and bodies, and the company pulled the ad after backlash. The Sweetshop CEO Melanie Bridge allegedly defended the production in a now-deleted statement, saying, “AI didn’t make this film. We did.”
Mashable Trend Report
This Tweet is currently unavailable. It might be loading or has been removed.
As for synthetic people, look for a mismatch between the face and the voice. Australia’s national science agency identified poorly synchronized lip movements, flickering around the eyes, and expressions that don’t fit the emotional tone of speech as potential deepfake clues.
4. Lettering and details fall apart
Sometimes, the overall composition survives while the smaller elements look like nonsense. Product labels, objects on shelves, scenery, and body anatomy can look convincing at a glance but make no sense up close.
This Tweet is currently unavailable. It might be loading or has been removed.
The promotional images for Glasgow’s creepy Willy’s Chocolate Experience offered a memorable example in 2024. Amid the “nightmarish” scene, the AI-generated ads promoting the event initially promised “enchanting entertainment,” “catchy tunes,” and “exaggerated lollipops.“
Those errors can also creep into real photographs when AI tools alter them. Photographer and marketer Abigail Hogue told Business Insider that a Valentine’s Day campaign she photographed for Quite Literally Books began running with AI changes. Friends sent screenshots showing garbled product lettering and merchandise that resembled knockoffs. She disabled the AI creative enhancements and republished the campaign.
That unevenness makes lettering a useful clue, though an increasingly imperfect one. As the Global Investigative Journalism Network’s guide explains, newer generators are much better at producing readable text and anatomically correct hands.
5. The action plays like a fantasy blockbuster
AI-GENERATED IMAGE
Credit: Screenshot: NFL
Sometimes, AI-generated ads lean into the absurdity, turning a simple sales pitch into a Marvel-level spectacle. The drama keeps escalating, the scenes get stranger, and a soda commercial starts to feel like the trailer for an entire cinematic universe.
Fanta’s latest Halloween campaign takes that approach. The ad features a vampire driving a sports car, a werewolf performing a DJ set, and a cyborg Frankenstein — all to promote soda. The brand spent a year and a half developing the world with AI-powered entertainment studio GRAiL, describing the technology in its announcement.
This Tweet is currently unavailable. It might be loading or has been removed.
And Kalshi’s deliberately chaotic commercial, which appeared in the YouTube TV stream of Game 3 of the 2025 NBA Finals, jumps through surreal scenes with the energy of an internet fever dream. Creator PJ Accetturo told NPR that he made it in two days using AI tools. See also: The NFL’s larger-than-life “Ride the Float” commercial.
This Tweet is currently unavailable. It might be loading or has been removed.
6. The typography looks familiar
Sometimes, it’s the font itself that can be a dead giveaway. Unrelated ads often share the same chunky, rounded headings or vaguely handwritten type, creating the impression that they all came from the same AI generator. In one Reddit discussion, users noticed similar-looking lettering across generated images, even when the words themselves were nonsense.
One particular Brushstroke font is even being called “the new Comic Sans” by some graphic designers.
This Tweet is currently unavailable. It might be loading or has been removed.
This Tweet is currently unavailable. It might be loading or has been removed.
Reporting on AI-generated flyers in Houston, Chron’s Gwen Howerton described a recurring combination of bright lettering against dark backgrounds, boxed bullet points, and words circled or underlined for emphasis. Many also used type resembling marker scribbles. The repetition extended beyond Instagram to bar screens, banners, and neighborhood signs.
Meanwhile, AI companies have cultivated another recognizable style through their own branding. Claude and Perplexity have openly embraced serif fonts — letters with small strokes at their ends, familiar from printed books. Asked why, Monotype’s senior executive creative director Charles Nix told Creative Bloq: “In order to seem more human.” He also described the historical associations of serif typography as a way to convey authority and distinguish these brands from the sans serif lettering that has dominated technology branding.
Designer and type practitioner Keya Vadgama offered a similar explanation to WIRED: “Serifs have an origin in calligraphy. It connotes a very human, fluid way of making letterforms.” Reporters added that Runway and Manus had also adopted serif typography in their interfaces and branding.
Some ads are starting to disclose AI use
Lately, the clearest answer to “is that ad AI?” may come from the ad itself. In a June update, Meta said it was rolling out an “About this ad” section containing AI labels for ads created or significantly edited with its generative tools. It also said it would begin labeling ads when it detected industry-standard signals from third-party AI tools.
That push for transparency is reaching state law, too. Pennsylvania Gov. Josh Shapiro signed the Artificial Intelligence Deceptive Advertising Act on Oct. 22. Taking effect Dec. 21, the law requires clear disclosures when AI-generated or significantly AI-altered content creates a false impression about the nature, origin, or authenticity of advertised goods or services. It won’t put a label on every AI-assisted ad, though. The legislation exempts evidently fictional content, parody, satire, commentary, criticism, and text-only content.
As AI ads get better and better, the clues keep changing, which means that by the time we know what to look for, there may already be something new to catch.
