A road bike with a second set of handlebars growing out of the saddle. Frame lettering nobody could read. That was the Instagram ad REI’s audience scrolled past in June 2026, and it became the internet’s favorite example of what happens when automated tools get the final say.
The laughs were easy. The business lesson is harder. Business Insider’s reporting, based on interviews with eight advertisers and agency executives, described Meta’s AI ad tools as chaotic, producing absurdities and misrepresentations.
That is the pattern behind this whole list. In each case, a machine produced or changed something a customer was going to see, and a human check either came too late or never happened.
Some of these cases hit Facebook and Instagram directly. A few ran on TV, YouTube or in print. I kept them because the failure is the same one your brand could repeat tomorrow.
Every case comes from public reporting. Where a company disputes how events were described, I say so. By the end you will have 15 real examples and a plain checklist for reviewing AI-made ads before they reach your audience.
Table of Contents
- REI’s Two-Handlebar Bike
- The Pajama Dress That Became a Shirt and Pants (and the Women’s Group That Got Men)
- The Valentine’s Day Campaign Meta’s AI Changed Without Permission
- True Classic’s Top Ad and the AI “Granny” Problem
- The Meta AI Ad That Advertised a 100% Off Deal
- The AI Settings That Switch Themselves Back On
- The Queensland Symphony Orchestra’s Facebook Ad
- Google’s “Dear Sydney” Gemini Ad
- McDonald’s Netherlands Pulls Its AI Christmas Ad After Three Days
- Vogue, Guess and the AI Model Nobody Asked For
- Coca-Cola’s AI Holiday Ads
- Meta’s AI Image Detector and the Muse Image Rollback
- When Advantage+ Spent Budgets in Hours
- Refunds That Arrive as Ad Credits
- Meta Says Review It Yourself: The Accountability Gap
1. REI’s Two-Handlebar Bike: When an Instagram Ad Rewrote a Real Photo
In June 2026, REI ran an Instagram ad for a Van Rysel road bike. Viewers spotted a second set of handlebars near the saddle, along with extra chains and blurry lettering on the frame.
The ad did not start as an AI invention. The picture came from a real photo shoot, and Amity Rockwell publicly confirmed she had been hired for that shoot. A good photo went in. A distorted one came out.
REI pointed at the platform. The retailer said Meta had auto-enrolled it in an AI personalization tool that altered a vendor-provided image. After the criticism, REI opted out of the feature.
Whoever is right about the toggle, customers saw REI’s name on the ad. That is the real cost of this kind of failure. People do not blame the tool. They blame the brand.
The lesson:
Treat every AI-edited version of your creative as a new ad that needs its own approval. Open the live ad preview, not just the original upload, and check it again after launch.
Also worth reading: See how small automation habits quietly hurt your reach in these AI mistakes that kill Instagram reach.
2. The Pajama Dress That Became a Shirt and Pants (and the Women’s Group That Got Men)
Ads consultant Jessica Gleim works with female-founded brands, and she told Business Insider the suggestions she saw were often unusable. In one case, Meta proposed turning a client’s pajama dress into a shirt and trousers. For a women’s networking group in Montana, the suggestion was to add men to the ad.
These are different kinds of mistakes, and both matter. A pajama dress and a shirt-and-pants set are different products. A shopper who clicks expecting one and receives the other is a return, a refund request and a lost customer.
Adding men to a women’s group ad changes who the group appears to be for. The audience the group exists to serve may feel it was not seen at all.
AI tools are good at patterns. They are weak at identity: what exactly this product is, and exactly who this brand speaks to.
Run an identity check on every AI-suggested variation before it goes live:
- ✅ Does the product shape, colour and cut match what you actually sell?
- ✅ Are the people shown the audience you serve?
- ✅ Is every word of on-image text readable and correct?
- ✅ Do the price, logo and offer match your store?
Related read: Brands that skip human review often meet the dangers laid out in these 11 hidden dangers of AI social media management.
3. The Valentine’s Day Campaign Meta’s AI Changed Without Permission
Seasonal campaigns leave no room for do-overs. A small business called Quite Literally Books had its Valentine’s Day ad altered without permission, in the same wave of complaints that followed the REI incident. The business belongs to photographer and marketer Abigail Hogue, who had shot the creative herself.
That detail matters. Hogue was not a big brand handing work to an agency. She made the images, uploaded them and then watched a system change her work.
A small shop has no buffer. There is no agency team to catch a warped product photo at midnight, and Valentine’s Day will not wait for a support ticket to clear.
If you run ads yourself, build a short routine around launch day:
- ✅ Confirm which AI creative enhancements are on, at the ad level.
- ✅ Preview every placement, including Stories and Reels.
- ✅ Check the live ad again a few hours after it starts running.
- ✅ Keep screenshots, in case you need to ask the platform for a refund.
The lesson:
Your original photo is not the ad your customer sees. Always judge the version that is actually running.
4. True Classic’s Top Ad and the AI “Granny” Problem
In October 2025, Business Insider reported that advertisers were seeing bizarre ads, including an AI granny created by Meta’s tools. The head of marketing at clothing brand True Classic posted on X about what Meta’s AI had done to the brand’s top-performing ad.
The same report said the strange results continued even after some brands switched off AI-related settings. One agency CEO said the feature kept turning itself back on, even for ads where he had disabled it twice.
Meta’s position, according to the same report, is that millions of advertisers see value and better performance from its Advantage+ creative tools. Both things can be true. Plenty of advertisers do get results, and plenty also get surprises.
The danger here is the top performer. When a winning ad gets altered, you may not notice for days, because the numbers still look fine while the creative drifts away from your brand.
The lesson:
A winning ad deserves more checking, not less. Save a copy of the approved version, and compare it against what is live every week.
Don’t miss: This is only one chapter of a bigger story. Read how 20 AI marketing fails destroyed brand trust for even more real examples.
5. The Meta AI Ad That Advertised a 100% Off Deal
The 2025 holiday season brought a different kind of surprise. AdExchanger spoke with ten social advertisers who said Meta’s AI creative feature had been producing strange, off-brand ads for about a month. David Herrmann, who runs the boutique agency Herrmann Digital, posted on X that Meta’s AI had even advertised a 100% off deal. His account contained discount codes created for sending free gifts, and the AI appeared to pick up on them.
Compare that with the two-handlebar bike. A distorted image is embarrassing, but a fake offer is a promise a customer can screenshot. Whether a brand has to honor a mistaken discount is a question for a lawyer. The customer service headache is real either way.
The same report added a second problem. Three agency buyers said even large spenders often cannot get a human response when something goes badly wrong. That leaves the advertiser to catch the error alone, often after the ad has been running for days.
Here is how to protect your offers:
- ✅ Keep promo codes and internal gift codes out of anything an AI tool can read.
- ✅ Read the headline and offer in every ad preview, not just the picture.
- ✅ Set spend and discount alerts so a runaway offer gets noticed fast.
- ✅ Save screenshots of anything wrong, in case you need to dispute it.
The lesson:
An AI tool can only misuse what it can see. Limit what it can see, then check what it published.
6. The AI Settings That Switch Themselves Back On
This case is less dramatic than a mutant bike, and it may be the most important one on the list. Several advertisers told Business Insider they had hit a problem where Meta switched on its automatic adjustment settings even after they had turned them off.
Meta’s position is that many AI features are off by default and that advertisers can review outputs before publishing. Advertisers answer that the review step is a lot of work when you run hundreds of ads, and that a setting you disabled can quietly come back.
By August 2026, the pushback had reached agency boardrooms. Agency executives told AdNews that they were switching off Meta’s AI creative tools, arguing that an aggressive rollout put the platform’s adoption goals ahead of brand control.
The danger is not one bad ad. It is the silent change, where the ad that goes live is not the ad you approved and nobody tells you.
Build a simple habit around it:
- ✅ Check your creative settings before you publish, then again right after.
- ✅ Re-check running campaigns on a weekly schedule.
- ✅ Keep a screenshot of the approved version to compare against the live ad.
- ✅ Ask your agency in writing which AI features are active on your account.
The lesson:
Turning something off once is not a control. Verifying it stayed off is.
Keep reading: Automation promises save time, but it also creates blind spots. See the AI automation traps that hurt TikTok creators for the same pattern on another platform.
7. The Queensland Symphony Orchestra’s Facebook Ad
Not every failure comes from a platform tool. Sometimes the brand makes the mistake itself.
In early 2024, the Queensland Symphony Orchestra ran a Facebook ad inviting people to try “something different” on a Saturday night. As Creative Bloq reported, the image showed musicians with too many fingers or no hands at all, and the figures had a waxy sheen that gave the AI origin away.
The backlash came from inside the arts world. Artnet reported that the Media, Entertainment and Arts Alliance publicly criticized the orchestra on Facebook, calling the artwork some of the worst AI-generated images it had seen, and objecting that an arts organization had used it instead of paying artists.
The orchestra’s side of the story, as DIYPhotography relayed it, is that the image was a sponsored ad that reportedly ran for about 24 hours before a Mahler performance. A short run did not stop the reaction. The people who cared most were its own musicians and patrons.
This case teaches something the REI example does not. The audience you most want to win over may also be the audience most offended by AI imagery. An orchestra sells live human skill. A fake, waxy audience of malformed musicians works against its own message.
The lesson:
Match the tool to the brand promise. If your product is human craft, show human craft, even in a small ad.
8. Google’s “Dear Sydney” Gemini Ad
This one was not an AI mistake in the visual sense. It was a message mistake, and it ran on TV rather than on Facebook or Instagram.
During the 2024 Paris Olympics, Google aired a 60-second ad for its Gemini assistant. A father asks Gemini to help his young daughter write a fan letter to hurdler Sydney McLaughlin-Levrone. The backlash centered on the idea of a parent handing a child’s personal letter to a chatbot. Tom’s Guide covered Google’s decision to pull it from Olympic coverage.
The details show how fast a brand can retreat. According to afaqs, the ad aired widely on NBCU’s Olympic broadcasts, with its last broadcast at midnight on July 30. It stayed on YouTube, but with comments turned off.
Google’s defense was that the goal was to celebrate Team USA and that the ad had tested well beforehand, NewsNation reported. That is the uncomfortable part. Pre-launch testing missed something the public noticed immediately.
The takeaway applies to any brand. A tool demo can make AI look helpful, but a story about replacing a child’s own words with a machine’s makes people feel something is being taken from them.
The lesson:
Before you show AI in your ads, ask what the story says about the people in it. If the answer is “a machine does the human part,” expect pushback.
One more link: Authenticity is the thread running through all of these cases. Our breakdown of AI content habits that make a brand look fake shows how to avoid it.
9. McDonald’s Netherlands Pulls Its AI Christmas Ad Within Days
This case ran on YouTube, not Facebook or Instagram, but it shows how quickly an AI ad can turn on its own brand.
McDonald’s Netherlands posted a 45-second Christmas film on its YouTube channel on December 6, 2025. It was titled “It’s the Most Terrible Time of the Year” and was made from AI-generated clips. NBC News reported that McDonald’s Netherlands confirmed it had removed the ad after the social media backlash.
The idea was a dark joke about holiday stress, with the chain offered as a place to hide until January. Viewers did not enjoy it. Mediaweek’s coverage described a sharp consumer reaction that reignited the industry debate over how far brands should lean on AI.
The company’s explanation was honest and short. It said the ad was meant to reflect stressful moments in the Netherlands, but that for many guests this time of year is “the most wonderful time of the year.” The production side pushed back on the idea that AI meant a shortcut. Fox Business reported the production company said the job took ten people five weeks of full-time work, far more hours than a traditional shoot.
That detail matters. The ad failed even though a lot of human effort went into it. The tool was not the only problem. The tone was.
The lesson:
AI cannot rescue an idea that does not fit the season. Before you spend time polishing AI clips, test whether the concept itself will land with real people.
10. Vogue, Guess and the AI Model Nobody Asked For
Print can fail too. In August 2025, Vogue ran a two-page Guess ad featuring an AI-generated model. The only disclosure was fine print reading “Produced by Seraphinne Vallora on AI,” as USA Today reported.
Some readers said they cancelled subscriptions, and models and creatives voiced worry about their own work. The agency behind the images, Seraphinne Vallora, created two digital models for the campaign, named Vivienne and Anastasia. According to DesignRush, many readers missed the tiny disclosure entirely.
The blame was passed around. Malay Mail reported that Vogue said the placement was purely commercial and not an editorial decision, and that Guess’s co-founder had approached the AI agency directly.
Three things went wrong here. First, disclosure was easy to miss, which made the ad feel sneaky. Second, the audience for a fashion magazine tends to care about real bodies and real careers. Third, neither company took ownership early, which let the story run.
Compare this with the REI case. REI’s problem was a tool that changed a real photo. Guess’s problem was a choice to replace the photo entirely. Different mistakes, same loss of trust.
The lesson:
If you use AI-generated people, say so clearly, not in the corner. Decide in advance who answers questions when the criticism arrives.
Related read: Fake people carry their own risks. See the full story behind an AI influencer’s rise and fall.
11. Coca-Cola’s AI Holiday Ads: Backlash With a Twist
Coca-Cola is the most interesting case on this list because the story is not one-sided.
Coca-Cola released its first AI-made holiday ads in 2024, and many viewers disliked them. Undeterred, it came back in November 2025. NBC Chicago, via TODAY, reported that the new “Holidays Are Coming” ads arrived on YouTube on November 3, with personified animals this time, and that comments were again largely negative.
The company defended the work. Best Media Info reported that five specialists refined 70,000 clips in 30 days using tools such as Sora, Veo 3 and Luma AI. Coca-Cola’s head of generative AI argued that the craftsmanship was far better than the year before, according to VnExpress.
Here is the twist. eWeek noted that research from System1 suggested audiences were more forgiving in private testing than the loud online reaction implied.
So was it a failure? It depends on what you count. Online sentiment was harsh. Private testing was kinder. That gap is useful for any brand. The loudest voices on social media are not always your whole audience, but they shape headlines.
The lesson:
Measure both. Track public reaction and private performance, and do not assume one tells the whole story.
Here is a fair way to judge an AI campaign before you launch it:
- ✅ Would the ad work without the AI label?
- ✅ Does it feel warm, or just polished?
- ✅ Can you explain, in one sentence, why AI was the right tool?
- ✅ Do you have a plan for public criticism?
12. Meta’s AI Image Detector and the Muse Image Rollback
This case is not an ad, but it matters to every advertiser on Meta’s platforms.
Meta launched its Muse Image generator in July 2026. Campaign reported that Meta pulled a feature, which let users generate images from public Instagram accounts, just days after launch following privacy and copyright criticism. Meta said the feature had missed the mark.
Then came the detection problem. A Reuters analysis, as summarized by TechStory, found that Meta’s detector flagged all 40 original Muse Image test pictures as AI-generated, but missed 55% of them once they were cropped to roughly a third or half of their original size.
What does this mean for advertisers? Platforms are building both the generator and the detector, and neither is perfect. If you rely on a label or a watermark to tell you what is real, you may be disappointed.
For brands, the practical risks are clear. A cropped, reposted version of your AI-made ad may lose its label. A fake ad using your brand’s look may slip past detection. Neither Meta nor any other platform can promise full protection.
Build your own record:
- ✅ Keep the original files for every ad, including the prompts you used.
- ✅ Note which tools touched each creative, and when.
- ✅ Search for copies of your ads that appear outside your account.
The lesson:
A watermark is a helpful signal, not a guarantee. Your own paper trail is the safer protection.
Worth a look: Unlabelled AI content is also reshaping politics. Read how AI-made campaign ads can slip through without labels.
13. When Advantage+ Spent Budgets in Hours
Some failures cost money rather than reputation. In February 2024, advertisers reported that Meta’s Advantage+ shopping campaigns were behaving unpredictably. NewsBytes reported that RC Williams, co-founder of the agency 1-800-D2C, found that Meta had used about 75% of the daily ad budgets for two clients within a few hours on Valentine’s Day. The same report said the tools had been disregarding cost caps since February 14.
The complaints went beyond one agency. eMarketer summarized reports from advertisers who said the tool inflated their costs by as much as ten times. It also noted that advertisers struggled to get issues addressed because of layoffs in Meta’s customer service department.
This is not an image error, but it is the same root problem. When automation makes decisions you cannot easily see, mistakes grow quietly until the bill arrives. Think of it as the money version of the two-handlebar bike.
Put guardrails around spending:
- ✅ Set daily and lifetime spend alerts on every campaign.
- ✅ Check spend mid-morning on launch days, not only at the end of the day.
- ✅ Keep a named contact at your agency for emergencies.
- ✅ Save screenshots of spend and cost-per-result at the first sign of trouble.
The lesson:
Automation without visibility is a risk, whether the output is a picture or a budget.
14. Refunds That Arrive Late and Inconsistently
Even when a platform admits a mistake, getting money back can take weeks. This case is older, and it was a delivery bug rather than an AI creative error. It belongs here because it shows what advertisers face when automated systems fail.
On April 23, 2023, a glitch caused Meta to spend advertisers’ daily budgets within hours. Search Engine Land reported that costs per acquisition tripled, and one ecommerce advertiser’s $13,000 budget was spent in three hours with no results. Meta later said it had identified and resolved a technical issue in its automated systems and apologized.
The refund process was the second problem. The same report said some advertisers started getting refunds on May 12, but the amounts were uneven and it was unclear why some businesses received more than others. Smaller advertisers said the process was much slower for them. Bloomberg, quoted in that article, noted that refunds for ad glitches are typically cents on the dollar, and that Meta does not tell advertisers how it calculates them.
For a small business, that is painful. Cash flow does not wait for a platform’s review.
The lesson:
Assume that compensation, if it comes, will be slow and partial. Protect your budget first. Document everything, and file the claim as early as you can.
15. Meta Says Review It Yourself: The Accountability Gap
The final case ties the whole list together, and it is still unfolding.
Adweek reported on October 22, 2026 that online bookseller Bookshop.org and the digital agency Exverus found ad changes they had not meant to publish. ContentGrip’s summary adds detail. Bookshop.org said an automated account update enrolled it in a generative AI ad feature that inserted an AI-generated illustration into an ad originally made by people. It then disabled Meta’s AI ad features.
Exverus described a wider range of changes. Its team saw added sitelinks, translated copy, generated pictures and carousel variations based on client imagery, and said buyers had not chosen some of them. The agency pulled live ads to remove changes that slipped past its review.
The timing matters. Meta’s preview of its Commercial Terms says a revision takes effect on October 30, 2026. According to the same ContentGrip report, the updated AI section acknowledges that automated outputs can be inaccurate or unsuitable, and says businesses take responsibility for assessing and using the resulting content. Meta told Adweek that the update consolidates its existing AI advertising terms and that individual creative options can be disabled.
That is Meta’s contractual position. It does not settle how any dispute would be decided under the law, and this article is not legal advice. If you run ads for a regulated product or use licensed creative, have your legal team read the updated terms.
The accountability gap is clear. Brands say a platform changed their ads. Platforms point to terms and toggles. Customers only see the brand’s name.
The lesson:
Do not wait for the platform to protect you. Treat every AI-enhanced version as something you must inspect, approve and be ready to defend.
Your 10-Point AI Ad Review Checklist
Use this before every launch and again after every major account change.
- ✅ Open the ad-level creative settings and confirm exactly which AI enhancements are on.
- ✅ Turn off anything you did not choose, then check again right before publishing.
- ✅ Preview every placement: Feed, Stories and Reels on both Facebook and Instagram.
- ✅ Compare the preview against your approved original: product shape, color, logo and text.
- ✅ Read every headline, price and offer out loud.
- ✅ Check who appears in the image, and whether they match your audience.
- ✅ Confirm AI-generated people are clearly disclosed to viewers.
- ✅ Recheck live ads a few hours after launch, then weekly.
- ✅ Keep original files, prompts and screenshots of the approved version.
- ✅ Set spend alerts and know who to contact if something goes wrong.
Frequently Asked Questions (FAQ)
Is Meta responsible when its AI changes my ad?
It depends on the platform’s terms and the law, and the answer is still being tested. Meta’s terms place broad responsibility for AI feature output on the business using the tools. Brands can disagree with that, but you should assume you will be the one answering to customers. Ask a lawyer if the stakes are high.
How do I check whether Advantage+ Creative is changing my ads?
Work at the ad level. A practical guide explains how to turn off Advantage+ Creative enhancements before you publish, then check preview placements. After the ad goes live, look at it as a customer would and compare it with your approved version.
Did all of these cases happen on Facebook and Instagram?
No. The REI ad ran on Instagram. The pajama dress, women’s group, Valentine’s Day, True Classic and “100% off” cases involved Meta’s ad tools. The Queensland Symphony Orchestra ad ran on Facebook. Google’s “Dear Sydney” was a TV ad, McDonald’s Netherlands posted on YouTube, and the Vogue and Guess ad appeared in print. I included the non-Meta cases because the lessons apply to any brand.
Should brands stop using AI in ads altogether?
Not necessarily. The Coca-Cola case shows that reactions can be mixed, with loud online criticism and kinder private testing. The safer approach is to use AI as a draft tool, keep a human reviewer in charge, and be open when AI is used.
What should I do if an AI-altered ad is already live?
Pause it right away. Take screenshots, restore your approved creative, and check similar ads in the same account. Contact the platform’s support with your evidence, and consider telling customers if the ad misrepresented a product or an offer.
Final Thoughts
Looking across all 15 cases, the pattern is simple. The AI was rarely the only cause. The failures happened where a human check was skipped, rushed, or assumed to exist somewhere else.
The tools will keep improving, and so will the settings that control them. The brands that avoid headlines will be the ones that look at what is actually running under their name.
Keep reading:
- Learn how AI influencer scandals create real brand risks.
- See why Pinterest flagged and banned some AI pins.
- Find out how Google treats AI-written social captions.