Only 8.3% of AI Employer Descriptions Matched What Employees Were Actually Saying

Only 8.3% of AI engine answer descriptions of employers closely reflected what employees were actually saying on Glassdoor and Indeed.

This is one of the most surprising findings from our client data on BrandScore. And it raises some serious questions about what candidates are being told when they use AI to research potential employers.

We see what AI is telling candidates about employers every day through BrandScore. But we wanted to understand how closely those descriptions reflected what employees were actually saying in reviews.

So we took 2,970 Glassdoor and Indeed reviews, all published within the previous two years, and compared the collective sentiment for each employer and experience theme against 386 AI answers from ChatGPT, Perplexity, Gemini and Claude.

The result? Just 8.3% of the AI employer descriptions in the answers were classified as closely representative of the employee review sentiment.

As part of our work with employers, we also examine the sources AI cites, checking what those pages actually say and what evidence may have been used by AI.

Glassdoor regularly appears among the most cited sources, alongside Indeed. But when we examined 1,717 genuine Glassdoor citations, only 3.3% linked directly to an individual employee review.

Most pointed to employer review landing pages and aggregated content. The 3.3%? They were the two or three reviews that anyone can see without logging in. AI can't log in.

Now, that doesn't mean AI isn't accessing some employee reviews through other means. Our separate testing showed that it sometimes can, although access varied considerably between AI engines.

But there is a big difference between citing Glassdoor and accurately representing what employees are saying. Either way, candidates believe AI answers reflect the actual experience of working there. AI says things like "based on Glassdoor reviews", for example.

The 3.3% finding tells us where citations pointed. The 8.3% finding tells us how closely the AI narrative matched the employee feedback we independently collected.

We can't say one caused the other.

What we can say is that AI may be drawing on ratings, summaries, search results, employer content and other sources to construct its answers. And the resulting picture can be quite different from the experiences employees are reporting.

You could have excellent employee feedback about flexibility, career development or culture, but AI may barely communicate those strengths. AI answers could even contradict that feedback.

This is just one reason why so many employers are using BrandScore.

We measure what six leading AI experiences, including Google AI Overviews and AI Mode, are saying about employers. We provide automated discovery of whether AI names you to candidates, how it describes your employer brand, which sources shape its answers and how you compare with your talent competitors. Google AI Overviews now frequently appear above traditional Google Search results.

Uniquely, we don't just list citations. We test every single one to check the evidence AI may have used from that cited domain.

That gives employer brand teams something they haven't had before: the ability to see the difference between the employee experience, the brand they're communicating and the version AI is presenting to candidates.

We run 1,400+ candidate-style prompts across six AI experiences, presenting a dashboard of results, trends and recommendations.

Do you know how closely AI's description of your organisation reflects what your employees are actually saying?

Join employers such as the BBC, Rio Tinto, Westpac and more using us. Check us out and get your first set of results in a matter of hours.