An AI reputation crisis starts when a false or outdated statement about a company is screenshotted, shared and put to the company by a reporter before anyone inside has seen it. Two documented cases show the pattern: Air Canada's chatbot error in 2024 and the false claims ChatGPT produced about Brian Hood and Mark Walters.
By Ronn Torossian. Last updated: November 7, 2026. Cases are attributed to named sources. The scenario and plan are my own judgment, and this is communications guidance, not legal advice. The holding statement and drill are hypothetical templates for a fictional company.
How could an AI reputation crisis start inside an AI answer?
A crisis starts inside an AI answer in four distinct steps, each supported by documented public disputes. The table lists each step alongside the corresponding documented example.
| Step | What happens | Documented example |
|---|---|---|
| 1. A wrong answer appears | An engine or chatbot states something false | ChatGPT said Mark Walters was accused of embezzlement, per Loeb & Loeb |
| 2. Someone relies on it | A customer, reporter or investor acts on it | Jake Moffatt booked flights after Air Canada's chatbot misstated its bereavement fares, per Global News |
| 3. It becomes public | A screenshot, complaint or lawsuit spreads | Brian Hood threatened to sue after a false bribery claim, per Paste Magazine |
| 4. The company is asked | A reporter or regulator wants a response | Air Canada argued it could not be liable for its chatbot, and lost, per CBS News |
Why does a wrong answer travel so far?
A wrong answer travels far because readers often stop at the summary and because a screenshot is easy to share. Pew Research Center analyzed March 2025 browsing data from 900 U.S. adults and found that users clicked a traditional search result in 8% of visits when a Google AI summary appeared, against 15% of visits without one, and only 1% clicked a link inside the summary.
The finding describes Google summaries, not every engine, but it shows why a company may be the last to see a wrong answer. Readers take the summary at face value, and no one tells the company until a customer or reporter asks.
What did the Air Canada case show about a company's own chatbot?
The Air Canada case showed that a company can be held responsible for what its own chatbot says. On February 14, 2024, British Columbia's Civil Resolution Tribunal ordered Air Canada to pay $812.02 to Moffatt, according to CBS News. Tribunal member Christopher Rivers wrote that it "makes no difference whether the information comes from a static page or a chatbot," as quoted by MobileSyrup.
What did the Walters case show about a third-party engine?
The Mark Walters case showed that a plaintiff faces a hard road against an AI company. On May 19, 2025, a Georgia judge granted summary judgment to OpenAI, according to Bar & Bench, finding no defamatory meaning as a matter of law and no showing of fault or damages, per Loeb & Loeb. A company hit by a false third-party answer should plan on communications, not a lawsuit, as its first tool.
How should a company respond when a reporter asks about an AI answer?
A company should confirm the facts, check the answer itself and respond with a short, accurate statement. The steps below are editorial judgment and not legal advice.
- Ask the reporter for the exact question, the engine and the date, and save the screenshot.
- Run the same question yourself, twice, in the same engine. Muck Rack notes that asking the same question again can return different citations, according to Unite.AI, so the answer may differ.
- Check the claim against your records, and decide whether it is false, outdated or accurate.
- Send the reporter the correct facts with a source, and say what you are doing to correct the answer.
- Log the exchange, the facts sent and the date.
What would a holding statement for a false AI answer say?
A holding statement says the company is aware of the answer, states the correct fact and explains the next step. The example below is a hypothetical for a fictional company, Northwind Foods.
"Northwind Foods is aware that an AI engine has described our chief executive as having left the company. That is incorrect. Maria Lopez has been chief executive since January 2026. We have asked the engine's operator to correct the answer and updated the source pages it appears to use. We will update this page if anything changes."
What should a company prepare now?
A company should prepare four practical defenses before a wrong AI answer reaches the press.
- A monthly audit of what engines say, scored on a simple one-page grid with accuracy, tone and sources.
- A pre-approved holding statement for a false AI claim, which says the company is aware, is checking and will update.
- An owned fact page with current names, leaders, products and dates.
- A triage rule that sends serious false claims to counsel the same day.
The ChatGPT wrong CEO biography guide sets out how to correct a wrong answer about a leader, and the crisis communication plan template covers the wider plan. The guides to AI hallucinations and reputation and to the first 24 hours of a crisis follow later this week.
Who should own an AI reputation incident?
One named owner should run the incident, with counsel, communications and the web team as support. The table shows a simple split.
| Task | Owner | Support |
|---|---|---|
| Confirm the wrong answer and log it | Communications lead | Web team |
| Trace the source page and fix it | Web team | Communications |
| Decide whether the claim is serious | Counsel | Communications lead |
| Answer reporters | Named spokesperson | Counsel |
| Report the error to the engine's operator | Communications lead | Counsel |
How do you rehearse the response?
You rehearse with a short drill that uses a made-up wrong answer. The drill takes an hour and shows who does what. The scenario below is a hypothetical.
- The facilitator announces that a reporter has sent a screenshot of an AI answer that names the CEO in a false story.
- The team confirms the facts, finds the source page and drafts the holding statement within 30 minutes.
- Counsel reviews the draft, and the spokesperson reads it aloud.
- The team lists what slowed them down, such as missing contacts or an unclear owner, and fixes each item.
Frequently asked questions
Can an AI answer start a PR crisis?
An AI answer can start a PR crisis when a false statement is screenshotted, relied on and reported before a company notices, as the Air Canada and Walters cases show.
Is a company responsible for its own chatbot?
A company can be held responsible for its own chatbot, as British Columbia's Civil Resolution Tribunal held Air Canada responsible in February 2024, according to CBS News. Courts differ across jurisdictions, so organizations should consult counsel.
What is the first step?
The first step is running a monthly audit to find an incorrect AI answer before a reporter does.
Who should speak to the reporter?
A named spokesperson with counsel's approval of the statement. The spokesperson sends the correct fact with a source and does not guess about why the engine said it.
Should a company ask an engine's operator to correct an answer?
Yes, using the operator's feedback channel, with the corrected pages as evidence. It may not work quickly, so correct the source pages and answer the reporter in the meantime.
Write the holding statement for a false AI claim this week and get counsel to approve it.