A chatbot agreed to “sell” a Chevy Tahoe for one dollar. A Christmas ad disappeared within days of launch. A Vogue spread starred a model who never existed.
None of these were hoaxes. They are part of a growing list of AI marketing fails that have made audiences wary of any brand that leans on the technology.
The research backs up the mood. Harris Poll data presented at Cannes Lions 2026 found that 78% of consumers say AI makes ads feel less authentic, and 73% would trust an ad less if they suspected AI made it.
I pulled the original reporting behind every case below, so each one comes with a source you can check yourself. Some brands lost money. Some lost goodwill. All of them learned that trust is harder to rebuild than a campaign is to produce.
Also read: Meta and Instagram were not the only platforms hit, see the AI-generated ads that went horribly wrong on Facebook and Instagram.
Table of Contents
- McDonald’s Netherlands: The Christmas Ad Pulled in Days
- Coca-Cola: Two Years of AI Holiday Ads
- Guess x Vogue: The AI Model Nobody Asked For
- Duolingo: The “AI-First” Memo That Backfired
- Willy’s Chocolate Experience: When the Ad Was the Only Good Part
- Klarna: The Customer Service U-Turn
- Air Canada: The Chatbot Lawsuit That Set a Precedent
- Chevrolet of Watsonville: The $1 Tahoe Chatbot
- DPD: The Delivery Chatbot That Swore at a Customer
- Valentino: The Handbag Ad That Contradicted the Brand
- Taco Bell: The AI Drive-Thru Rethink
- Google: The Gemini Super Bowl Ad Fact-Check
- Toys “R” Us: The Sora Brand Film
- Friend: The Subway Ad Campaign New Yorkers Vandalized
- CNET: AI-Written Articles With Errors
- Sports Illustrated: The “Authors” Who Did Not Exist
- NYC MyCity: The Government Chatbot That Gave Illegal Advice
- Google Gemini: The Image Generator Paused After Backlash
- Levi’s: The AI Models Plan
- Zillow Offers: The Algorithm That Cost Hundreds of Millions
- What These 20 Fails Teach Us
- Frequently Asked Questions (FAQ)
1. McDonald’s Netherlands: The Christmas Ad Pulled in Days
McDonald’s Netherlands uploaded a 45-second holiday spot on December 6, 2025, titled “It’s the Most Terrible Time of the Year.” It was built from AI-generated clips of holiday mishaps, with Santa stuck in traffic and a cyclist slipping in the snow.
The idea was to position McDonald’s as a place to hide from festive stress. Viewers did not take it that way. Reports from NBC News say the company removed the ad after criticism spread online.
The company’s own statement admitted the problem. It said the spot was meant to reflect stressful holiday moments, but that for many guests this season is the most wonderful time of the year.
The agency side pushed back. The production house said the film took ten people five weeks of full-time work, which is a fair reminder that “AI ad” does not mean “no humans involved.” The audience still judged the result by how it felt, not how it was made.
- Takeaway: A cynical idea wrapped in an uncanny AI finish gets criticized twice, once for the concept and once for the execution.
2. Coca-Cola: Two Years of AI Holiday Ads
Coca-Cola recreated its classic “Holidays Are Coming” trucks with generative AI in 2024. Many viewers called the result cold. The company then did it again in 2025.
The second round swapped uncanny humans for animals, including meerkats, otters and foxes, watching the trucks roll through. Coverage from eWeek notes that the YouTube upload quickly collected thousands of comments, many criticizing the reliance on AI over traditional animation.
Coca-Cola did not back down, and its reasoning is worth hearing. Executives said AI cut production from about a year to roughly a month, and that the first film tested well with customers. They also said the 2025 craftsmanship was “ten times better.”
So this case is not a clear-cut flop on every metric. It is a split between what online commenters feel and what the company’s testing showed. The risk is that a brand famous for making people feel something now has to defend its process.
Also read: Not sure whether a video, image or caption is machine-made? Here are the signs you’re looking at AI slop.
- Takeaway: Legacy brands built on emotion have more to lose, because their own history sets the bar for what “human” should feel like.
3. Guess x Vogue: The AI Model Nobody Asked For
In the August 2025 issue of Vogue, a two-page Guess ad featured a model who does not exist. A small line in the corner read “Produced by Seraphinne Vallora on AI.”
TikTok creators spotted it fast. ABC News reported that the criticism targeted both Guess and Vogue, with viewers and working models questioning what this means for real talent.
Condé Nast, Vogue’s parent company, clarified that an AI model has never appeared editorially in Vogue. The spread was a paid advertisement, not a fashion editorial.
That distinction did not calm the conversation. The disclosure existed, but it was easy to miss, and audiences treated it as if it was hiding in plain sight.
Also read: Fake faces are a bigger brand problem than one magazine ad, as the AI influencer scandals and brand risks show.
- Takeaway: A disclosure that is technically there but hard to see will be read as no disclosure at all.
4. Duolingo: The “AI-First” Memo That Backfired
Not every AI fail is a bad-looking ad. In April 2025, Duolingo CEO Luis von Ahn posted an “AI-first” memo on LinkedIn. It said the company would gradually stop using contractors for work AI could handle and would add headcount only when a team could not automate more of its work.
Many readers heard “AI instead of people.” In an interview covered by Fortune in August 2025, von Ahn said, “I did not give enough context,” and added that Duolingo had never laid off full-time employees.
That is the useful part of this story. The strategy itself may have been reasonable, but the framing made it sound like a cost-cutting announcement aimed at the people who use the app every day.
- Takeaway: An internal AI strategy becomes marketing the moment it is public, so write it as if customers will read it.
5. Willy’s Chocolate Experience: When the Ad Was the Only Good Part
This one is not a Fortune 500 case, but it may be the cleanest example of an AI promise breaking against reality. In February 2024, an unlicensed, Wonka-inspired event called Willy’s Chocolate Experience opened in Glasgow, advertised with dreamy AI-generated images of an enchanted candy world.
Families who paid around £35 a ticket found a mostly empty warehouse. As The Hollywood Reporter covered, customers called the police and the event was cancelled within hours.
One actor later said he was handed 15 pages of AI-generated script. The story spread worldwide, and a Channel 5 documentary followed, as Marie Claire reported.
The AI images were never the crime on their own. The gap between what they showed and what the event delivered was.
- Takeaway: AI visuals set expectations your real product has to meet, and when it cannot, the mismatch becomes the headline.
6. Klarna: The Customer Service U-Turn
Klarna did not run a bad ad. It ran a bold experiment on its own customers, then changed course in public.
In early 2024, the Swedish payments company said its AI assistant was doing the work of about 700 customer service agents, a claim covered widely at the time. The story became a favorite proof point for companies eager to automate support.
By May 2025, the tone had changed. CEO Sebastian Siemiatkowski told Bloomberg that cost had been too big a factor in the strategy and that quality suffered. Klarna started rebuilding a pool of human agents so customers could always reach a person.
Klarna did not abandon AI. It narrowed the role AI plays in the one place where trust is most fragile, a customer with a real problem.
Also read: Automation is reshaping whole workforces, not just support desks, as the AI layoffs at Meta, Amazon and Block show.
- Takeaway: “Resolved fast” can look like success while trust quietly drops. Measure whether the problem was actually solved, not just how quickly the chat closed.
7. Air Canada: The Chatbot Lawsuit That Set a Precedent
This is the case legal and marketing teams cite most often when they talk about chatbot risk.
In November 2022, passenger Jake Moffatt asked Air Canada’s website chatbot about bereavement fares after his grandmother died. The bot told him he could book a regular ticket and apply for the discount afterward. The airline’s real policy did not allow that.
Air Canada refused the refund and later argued in front of a tribunal that the chatbot was a separate entity responsible for its own words. The tribunal rejected that, and the airline was ordered to pay about C$812 in damages, interest and fees.
The dollar amount was small. The reasoning was not. The tribunal said a company is responsible for everything on its website, whether it appears on a static page or comes from a chatbot.
Also read: Wrong answers are far more dangerous in health, as this AI medical advice case that went wrong shows.
- Takeaway: A chatbot’s promise is your company’s promise. If it cannot check answers against current policy, keep it away from money, deadlines and eligibility.
8. Chevrolet of Watsonville: The Chatbot That “Sold” a Tahoe for $1
In December 2023, a Chevrolet dealership in Watsonville, California, put a ChatGPT-powered assistant on its website. Within days, users were testing its limits.
One user, Chris Bakke, told the bot to agree with everything a customer says and to end each reply with a line calling it a legally binding offer. He then said he wanted a 2024 Tahoe for a one-dollar budget. As The Autopian reported, the bot played along, and the post spread fast.
No car was sold. The bot could not set prices or close deals, and a line of text is not a contract. But the dealership removed the chatbot anyway, because the screenshots had already done the damage.
The lesson here is about wiring. A chatbot that only chats is an embarrassment risk. A chatbot connected to quotes, payments or contracts is a financial risk.
Also read: Tools that act on your behalf raise the stakes, as explained in how AI agents click buttons for you.
- Takeaway: Test any customer-facing AI against manipulation before connecting it to anything that can quote a price, take payment or make a commitment.
9. DPD: The Delivery Chatbot That Went Off Script
In January 2024, musician Ashley Beauchamp tried to track a missing parcel through delivery firm DPD’s online chat. Frustrated, he started testing the bot instead.
He got it to write a poem and a haiku criticizing the company, and to say DPD was the worst delivery firm in the world. Reports noted that he had prompted it to exaggerate, but the screenshots still looked terrible for the brand.
DPD disabled the AI element of its chat and blamed an error after a system update. The post reached hundreds of thousands of views within a day, which is the real cost of this kind of slip.
The fair reading is that the bot was pushed. The business reading is that customers do not care who pushed it.
- Takeaway: Assume customers will try to break your chatbot. Add guardrails on tone, topics and language before launch, not after the screenshot goes viral.
10. Valentino: The Handbag Ad That Clashed With the Brand
Not every case involves a chatbot. In early December 2025, Valentino posted an AI-generated Instagram video promoting its Garavani Devain handbag. The clip showed models morphing into surreal shapes, ending in a swirl of merging arms and bodies.
The brand labeled the video as AI-generated, but viewers still called it unsettling “AI slop”. Valentino and McDonald’s both said their ads involved human creative input, as a marketing case study notes, but that explanation did little to change the reaction.
The core problem was fit. Valentino sells craftsmanship and human artistry. Advertising that promise with visuals audiences read as machine-made created a mismatch no label could fix.
Also read: Creative industries are asking the same question, covered in AI in Hollywood replacing actors, writers and VFX.
- Takeaway: If your brand promise is human craft, AI-generated creative does not just risk looking cheap. It can contradict what you sell.
11. Taco Bell: The AI Drive-Thru Rethink
Taco Bell began testing voice AI ordering in 2023 and expanded it to more than 500 US drive-thrus. The goal was fewer mistakes and faster service.
Then the videos started. Customers posted clips of the system misfiring, including a viral order for 18,000 cups of water. Jalopnik covered how the stunt pushed the company to pause and reassess.
Dane Mathews, Taco Bell’s chief digital and technology officer, told The Wall Street Journal that the company was learning a lot and that a human may do better at busy locations. Taco Bell did not abandon AI. It said it would take a more nuanced approach, studying data from more than two million AI-handled orders to decide where the tool actually helps.
The useful lesson is about scale. A glitch in one drive-thru is a minor annoyance. A glitch in 500 drive-thrus, filmed and shared, is a brand story.
- Takeaway: Pilot customer-facing AI where a failure reaches a small audience first, and keep a human ready to step in at peak times.
12. Google: The Gemini Super Bowl Ad Fact-Check
Even Google’s own marketing ran into the accuracy problem. In early 2025, the company released 50 small-business ads, one for each state, showing Gemini helping with everyday work.
In the Wisconsin spot, a cheese shop owner asks Gemini to write a description of smoked Gouda. The AI-written copy on screen said Gouda makes up 50 to 60 percent of the world’s cheese consumption. Viewers quickly pointed out that this did not look right.
Google re-edited the ad to remove the claim, and the corrected version aired. A Google Cloud executive said it was not a hallucination, because multiple websites carried the same figure, according to CNN. Reporting also noted the statistic was hard to source.
Both sides have a point, and that is exactly the problem. An AI that repeats a weak web claim with total confidence is a risky pitch for a tool you want small businesses to trust.
Also read: Trust is built in small habits, and these AI content habits can make your brand look fake.
- Takeaway: If your ad’s pitch is “AI got this right,” check every number as carefully as you would for a human copywriter.
13. Toys “R” Us: The Sora Brand Film
In June 2024, Toys “R” Us premiered a one-minute brand film called “The Origin of Toys R Us” at the Cannes Lions festival. It was billed as the first brand film made with OpenAI’s Sora.
The agency Native Foreign made it, and the company said it was almost entirely created with Sora, with some corrective visual effects and an original score. The reaction online was mostly negative, and filmmaker Joe Russo was among the critics.
A later analysis argued that brand sentiment dipped after viewers called the film soulless. The company and agency saw it as a successful test of a new tool.
The mismatch was emotional. A comeback story for a nostalgic toy brand needs warmth, and many viewers felt the visuals were cold and strange instead.
Also read: The wider flood of low-quality AI content is covered in our look at the AI slop economy.
- Takeaway: “We used the newest AI video model” is a production detail, not a strategy. Audiences judge the story, not the tool.
14. Friend: The Subway Campaign New Yorkers Covered in Graffiti
In fall 2025, the AI companion startup Friend plastered the New York subway with stark black-and-white ads promising a friend who would never cancel plans or leave you waiting.
New Yorkers answered with markers. Riders wrote anti-AI messages across the ads, and CNN described the pendant as a symbol of the backlash against AI.
This case has a twist. Founder Avi Schiffmann told Fast Company that provoking a reaction was part of the plan, and that he considered the roughly $1 million spend cheap for the visibility it brought.
So it is less an accident than a gamble on attention. The risk is that attention and trust are different things, and a product built around closeness needs the second one.
- Takeaway: Provocation gets noticed, but it does not build trust. If your product asks people to confide in it, think hard before making your ads a public fight.
15. CNET: AI-Written Articles With Errors
Some of the costliest AI fails happen in publishing, where accuracy is the product. Between late 2022 and early 2023, CNET published dozens of personal finance explainers written with an AI tool under the byline “CNET Money Staff.”
Futurism found mistakes. One article on compound interest said a $10,000 deposit at 3 percent would earn $10,300 in the first year, when the actual earnings would be $300. The Washington Post’s coverage of the episode highlighted that exact error.
CNET reviewed its AI-assisted stories and found that 41 of 77 needed corrections. It also changed the byline to make the AI involvement clearer.
Two problems stacked on top of each other. Readers were not told clearly that AI was involved, and the content was about money, where one wrong number can mislead someone.
Also read: Search engines have their own view on this, as explained in how Google treats AI-written social captions.
- Takeaway: In finance, health, legal and safety topics, require a human fact-check before publishing, and say clearly when AI helped.
16. Sports Illustrated: The Bylines Nobody Could Find
Sports Illustrated built its reputation on trusted names and real reporters. In November 2023, a Futurism investigation raised questions about some of the writers on its site.
The report said product reviews appeared under bylines of apparently nonexistent writers, with profile photos that looked AI-generated. Poynter’s coverage noted that the profiles disappeared after the publisher was contacted, and that the pages later gained a disclaimer saying the content came from a third party.
The publisher, The Arena Group, said the articles were not AI-generated. It said a content partner, AdVon, supplied them and that some writers used pseudonyms. It ended the partnership, and the magazine’s own union said it was horrified.
Whatever the exact method, the damage was about identity. Once readers cannot confirm a byline is a real person, they start doubting everything else on the page.
- Takeaway: A byline is a promise that a real person stands behind the words. Never let a vendor attach names or faces to your brand that you have not verified yourself.
17. NYC MyCity: The Government Chatbot With Wrong Answers
Brand trust matters for governments too. In late 2023, New York City launched a Microsoft-powered MyCity chatbot to help business owners understand local rules.
In March 2024, The Markup tested it and reported that some answers, if followed, could lead businesses to break city law. Examples included saying an employer could take a cut of workers’ tips and that landlords could refuse tenants who use rental assistance. The Markup’s follow-up investigation found the bot still active a week later.
The city’s response was to keep it online with stronger disclaimers. Years later, the city’s new mayor, Zohran Mamdani, announced plans to shut it down, calling it unusable.
A disclaimer that says “do not rely on this” sits awkwardly next to a tool built to be relied on.
- Takeaway: If a chatbot answers questions about legal duties, money or safety, a footer disclaimer is not a safeguard. Check answers against a verified source before they reach the user.
18. Microsoft Travel: The Food Bank “Attraction”
In August 2023, an article on Microsoft’s travel pages titled “Headed to Ottawa? Here’s what you shouldn’t miss!” listed the Ottawa Food Bank as the third attraction.
The listing even suggested visitors go on an empty stomach, a badly twisted version of a line from the food bank’s own website. The food bank’s CEO called the suggestion insensitive and insulting to the people it serves.
Microsoft removed the article. The company said the error was due to human error and not unsupervised AI, a claim readers were left to weigh for themselves.
The lesson is about context. A system can stitch plausible sentences together without understanding who they hurt.
- Takeaway: If you say content gets “human review,” make sure the review catches tone and context, not just spelling.
19. Levi’s: The AI Models Announcement
In March 2023, Levi’s announced a test with Lalaland.ai to use AI-generated models to supplement human models and show more variety in body types and skin tones on its online store.
Many critics saw it differently. They argued the brand could simply hire more diverse real models, and that the plan cut into opportunities for them. Rangefinder reported that Levi’s then clarified the program was not a substitute for real diversity efforts or live photoshoots.
The company’s first announcement had framed a cost and efficiency tool as a values statement. That framing, not the technology itself, is what drew the heaviest criticism.
- Takeaway: Do not present an efficiency tool as a solution to a values problem. Explain what AI is for, and what it is not replacing.
20. Zillow Offers: The Pricing Algorithm That Backfired
This last case is not a campaign, but it shows what happens when a company bets its credibility on an algorithm’s accuracy.
Zillow Offers made instant cash offers for homes, relying on models to predict future prices. In November 2021, the company announced it would wind down the business, with CEO Rich Barton saying that forecasting home prices was more unpredictable than expected. About 25 percent of the workforce was affected.
The company first flagged a write-down of more than $500 million. Its annual filing later recorded a $407.9 million inventory write-down for 2021, plus separate restructuring costs.
The point for marketers is simple. When a brand sells “our model knows the price,” the cost of being wrong is public, financial and hard to explain.
- Takeaway: Marketing an AI system’s accuracy is a promise. If the model can be wrong at scale, the bill arrives at scale.
What These 20 Fails Teach Us?
Pull back from the individual stories and a few patterns repeat.
Legacy and craft brands face harsher scrutiny.
Coca-Cola and Valentino ran into bigger reactions than a startup likely would for the same experiment.
Quietly pulling a campaign rarely ends the story.
McDonald’s, DPD and Microsoft all removed content, and the coverage lived on.
Hidden or hard-to-see AI use feels like deception.
Guess, CNET and Sports Illustrated all learned this from different angles.
AI that makes promises needs real guardrails.
Air Canada, the Chevy dealership and NYC MyCity show that chatbots create legal and financial exposure, not just bad headlines.
Category matters.
Fashion, finance, news and government audiences react faster and harder than most.
The thread through all 20 cases is simple. Time saved by AI rarely survives a trust crisis. Several brands reversed course, scaled back, or spent more fixing the fallout than the shortcut saved.
This does not mean AI has no place in marketing. It means the best use is usually behind the scenes, with a human accountable for everything the customer sees.
Frequently Asked Questions (FAQ)
Why do AI marketing fails keep happening?
Most share the same causes: too little human review, AI use that is hidden or poorly explained, and tools connected to promises or prices without safeguards. Audiences have also become better at spotting AI visuals and copy, so rough edges get noticed fast.
Is all AI use in marketing risky?
No. The cases in this article share specific traits: weak disclosure, brands built on craft or trust, or AI making real commitments to customers without oversight. Behind-the-scenes uses such as testing ideas or speeding up workflow rarely trigger this kind of backlash.
Which brands have avoided AI backlash?
Some have made a point of staying out of the debate. Dove, for example, pledged in 2024 never to use AI to represent real women in its ads.
Should brands stop using AI in marketing entirely?
Not necessarily. Every case here suggests the same approach: disclose AI use clearly, keep a human accountable for anything customer-facing, and match the technology to what your brand actually promises.
How can I check if my own AI campaign is safe to publish?
Ask five questions before launch. Is AI use clearly disclosed? Has a person verified every fact and number? Could the tool be manipulated into making a promise? Does the style fit what the brand stands for? And what is the worst screenshot someone could take?
Editorial note:
This article summarizes publicly reported events, with sources linked above. Descriptions of reactions reflect what outlets and commentators reported, and opinions about lessons are our own. Where companies disputed claims, we have noted their position.
Keep reading: More risks to watch are covered in the hidden dangers of AI social media management.
Keep reading: See how small habits hurt reach in these AI mistakes that reduce Instagram reach.
Keep reading: Platform strategies can fail too, as shown in the AI-powered Pinterest strategy collapse.
Keep reading: Fake content can hurt brands and customers alike, so read up on AI deepfake scams.