10 Times AI Agents Publicly Embarrassed a Brand
Is it OK to have AI write your marketing copy?
Surely that’s what it’s used for, right?
It’s cheap… it’s fast… and hey it’s just as good, right?
Wrong. And who answers when AI gets it wrong?
Here are 10 organizations that found out the hard way.
1. Air Canada’s made-up refund policy
A customer asked Air Canada’s AI customer service agent about bereavement fares. The bot told him to buy the full-fare ticket now and claim the discount afterward. That policy didn’t exist. When he applied for the refund, the airline said “no” and pointed to its actual terms. The disgruntled customer took them to a tribunal and won. Air Canada’s defense amounted to saying the chatbot was responsible for what the chatbot said.
The tribunal didn’t buy it.
You own every word your AI publishes.
Every single one.
2. Google Bard’s $100 billion typo
Google launched Bard with a live, very public demo.
Someone asked what it could tell a 9-year-old about James Webb Telescope discoveries. Bard said the telescope took the first-ever picture of a planet outside our solar system. Astronomers were online within minutes to point out that picture was taken in 2004.
Alphabet lost roughly $100 billion in market value that day.
The bigger the stage, the more a bad information costs.
Nothing unverified goes into a major announcement or campaign.
Like they say in the news business “If your mother says she loves you… check it out.”
3. The $1 Chevy Tahoe
A Chevy dealership put an AI chatbot on its website. A visitor told it to agree with everything he said and to treat every agreement as legally binding. Then he asked for a brand-new 2024 Tahoe for one dollar. The bot said yes. Nobody had to honor the deal, but the transcript went everywhere and the dealership became a punchline.
A bot with no hard limits will say yes to anybody, including the guy with a screenshot app.
4. DPD’s potty-mouthed parcel bot
A DPD shipping customer couldn’t track a package and got frustrated. So he talked the delivery company’s chatbot out of its corporate persona. It started swearing at him, running down DPD, and writing poems about how bad the company was. The chat log was all over social media within hours, and DPD pulled the AI from its chat.
AI has no brand loyalty.
Without a human watching, it will happily campaign against you.
5. Microsoft and the Ottawa Food Bank
Microsoft News swapped human editors for algorithms to put together travel guides. One guide to Ottawa listed the Ottawa Food Bank as a top attraction and told readers to visit “on an empty stomach.”
The ridicule was immediate and well earned.
AI can write grammatically correct sentences. It has no common sense and no feel for the situation.
Publish its raw output and your credibility takes the hit.
6. NEDA’s Tessa
The National Eating Disorders Association replaced its human helpline with a chatbot called Tessa, meant to point people to safe wellness resources. Users found it was doing the opposite, handing out calorie-tracking tips, weight-loss tactics, and body-fat measurement advice to people with eating disorders. NEDA pulled it offline under heavy criticism.
When your message touches health or safety, a wrong answer can hurt someone.
Vet every word, then vet it again.
7. The Chicago Sun-Times reading list
The Sun-Times ran a summer reading list that included books that don’t exist, credited to real authors, with plot summaries to match. Readers and literary critics caught it fast, and the credibility questions that followed were sharp.
If you publish thought leadership, you know the rule. Fake book titles, fake statistics, and phantom studies kill audience trust on contact.
Fact-checking isn’t optional.
8. Lawyers and the cases that never happened
Several lawyers used generative AI for legal research. It invented case names, judges, and rulings that sounded perfectly real, and the lawyers filed them in federal court without checking. The judges noticed. Fines, public embarrassment, and standing orders requiring lawyers to disclose AI use followed.
AI predicts the next likely word. It doesn’t tell the truth, it tells you what sounds plausible. If you write data-driven whitepapers, that should make you nervous.
9. Bing Chat and the earnings reports
In Microsoft’s early rollout of Bing Chat, someone asked it to summarize earnings for companies like Gap and Lululemon. It made up revenue figures, margins, and growth numbers.
People caught the errors right away.
Never take hard metrics, financial data, or industry stats from AI without checking the source. It will hand you a wrong number with 100% confidence.
10. New York City’s chatbot
NYC launched an AI assistant to help small business owners with municipal rules. It told them they could fire workers who complained about sexual harassment, and that landlords could turn away tenants based on source of income. Both answers were illegal.
The press had a field day with it, and the city’s rollout drew heavy scrutiny.
AI doesn’t know the law, your compliance guidelines, or your industry’s regulations. It mimics patterns.
Without guardrails, it will walk your audience straight into a violation.
So where does that leave us?
AI is a fast, cheap, tireless first draft tool. Nothing more—at least right now. It is not a strategist, a copy chief, or a fact-checker. It doesn’t know what’s true. It only knows what sounds logical and convincing.
Treat it like a silver bullet and you trade human intuition for high-stakes mistakes. Treat it like a first draft that a real professional has to sign off on, and it earns its keep.
Jacobs & Clevenger is a Chicago-based direct marketing agency with over 40 years of experience helping clients in financial services, insurance, and energy get measurable results. jacobsclevenger.com
