How to Catch an AI Hallucination About Your Company
To find out whether AI is saying something false about your company, ask it the questions candidates ask. Word each one a few different ways, run them in more than one AI tool, and check every answer against the facts. It takes about an hour. Do it once a quarter.
Why do AI tools get companies wrong?
AI hallucinations rarely look like mistakes. A wrong answer arrives in the same calm, confident tone as a right one, so a candidate has no reason to doubt it. They show up most where the facts about a company are thin, mixed up or out of date. A rename, a new CEO or a merger can leave old and new facts side by side, and an AI tool may blend them.
Which questions should you test?
Start with the questions AI tools get asked about employers. VisiPage, the AI visibility platform from the Best Practice Institute, tracks the most common ones, including: Is your company a good place to work? Who is the CEO? What awards or certifications does it hold? Then add the questions your recruiters hear most often.
How do you run the test?
Pick the AI tools your candidates use. ChatGPT, Perplexity, and Gemini are a sensible start. For each question, ask it three different ways, screenshot every answer with the date, and mark each fact as right, wrong, out of date or borrowed from another company. Keep the screenshots in one folder so you can compare next quarter.
What do you do when you find a wrong answer?
Fix the source, not just the answer. Best Practice Institute’s VisiPage work focuses on four signals AI relies on: structured data, such as Schema.org markup for your CEO, locations and certifications; repetition across sources, so your site, LinkedIn page and verified profiles say the same thing; trusted references, meaning independent sources that confirm your facts; and clear content, with plain sentences that state one fact at a time.
What does a strong record look like?
Mohegan uses its Most Loved Workplace® recognition in the places you’d expect, such as recruitment materials and LinkedIn, and in some you might not, such as investor materials. Every one of those places repeats the same verified fact: Mohegan’s own team members validated the recognition. That’s the kind of consistent, independent fact an AI tool can repeat with confidence.
How often should you check?
Every quarter, and after any big change: a new name, a new CEO, a merger or a new certification. AI models retrain, and a fixed answer doesn’t always stay fixed.
Get your free Most Loved Workplace Employer Brand Report at https://certcheck.mostlovedworkplace.com/
Frequently Asked Questions (FAQ)
What is an AI hallucination about a company?
Which AI tools should I test?
Does a Most Loved Workplace® certification help?
Louis Carter is the founder and CEO of Best Practice Institute, Most Loved Workplace, and Results-Based Culture. Author of In Great Company, Change Champions Field Guide, and Best Practices in Talent Management, as well as a series of Leadership Development books. He is a trusted strategic advisor and coach to CEOs, CHROs, and leaders of mid-sized to F500 companies – enabling change and steering employer brand development together with highly effective teams, leaders, and organizations as a whole.
0 Comments