The Glassdoor Problem: When AI Uses Outdated Reviews to Define Your Employer Brand
The Glassdoor Problem: When AI Uses Outdated Reviews to Define Your Employer Brand
A UK fintech company spent 18 months transforming its culture. New leadership, 4-day work week, doubled parental leave, rebuilt management training from scratch. Employee NPS went from -12 to +41. Glassdoor rating climbed from 2.9 to 4.2.
Then we ran an AI audit.
When we asked ChatGPT "What's the culture like at [Company]?", it responded:
"Reviews suggest a demanding work environment with long hours and limited work-life balance. Some employees have noted concerns about management transparency."
That description was accurate — in 2023. In 2026, it's a ghost of a company that no longer exists.
Welcome to the Glassdoor Problem: AI is defining your employer brand using data that may be years out of date.
How AI Uses Glassdoor (And Why It's Dangerous)
When AI answers questions about employers, it draws from whatever sources its training data contains. For most mid-market companies, that's a thin pool: your website, maybe some press coverage, LinkedIn, and — disproportionately — Glassdoor.
Glassdoor reviews carry outsized weight for a simple reason: they're detailed, opinion-rich, and structured. They contain exactly the kind of content AI needs to answer questions about culture, management, compensation, and work-life balance.
The problem is freshness.
Large language models are trained on data snapshots. GPT-4's training data has a knowledge cutoff; Claude's has a similar boundary. Even when models have web access (like Perplexity), they still synthesise from cached and crawled content that may be weeks or months old.
Glassdoor compounds this. The platform's review history goes back years. A company with 50 reviews might have 30 from 2021-2023 and only 20 from 2024-2026. When AI summarises sentiment, it's averaging across that entire history — giving equal weight to a frustrated review from three years ago and a glowing one from last month.
The Numbers
We analysed AI responses about 200 UK employers through OpenRole audits. When AI did provide culture or sentiment information:
- 62% of responses referenced themes that matched reviews older than 18 months
- 34% cited specific concerns that the company had demonstrably addressed
- Only 23% of AI culture descriptions reflected the company's current state accurately
- In 41% of cases, AI's tone was measurably more negative than the company's recent reviews
That last stat is particularly damning. AI isn't just outdated — it's biased towards older, more negative data because dissatisfied employees historically leave more detailed reviews, giving AI more material to work with.
Why "Just Get Better Reviews" Doesn't Work
The obvious response is: encourage more positive Glassdoor reviews to dilute the old ones. Some companies run campaigns asking current employees to leave reviews, hoping to shift the average.
This helps on Glassdoor itself. It barely moves the needle with AI.
Three reasons:
1. AI Doesn't Recalculate in Real Time
Even if your Glassdoor rating jumps from 3.1 to 4.3 this month, AI models trained on older data still carry the old sentiment. Models are retrained periodically, not continuously. Your improved Glassdoor score won't propagate to AI responses for weeks — or months.
2. Review Volume Is Cumulative
A company with 200 Glassdoor reviews accumulated over 6 years has a deep history AI can draw from. Even 50 new positive reviews still leave 150 older ones in the pool. AI doesn't weight by recency the way a human reader would.
3. Glassdoor Isn't the Only (or Best) Source
Here's the counterintuitive truth: the best way to fix what AI says about you isn't to fix Glassdoor. It's to give AI better sources entirely.
AI models prioritise first-party content — content published on your own domain — over third-party review sites. A detailed blog post about your culture, published on your company website, carries more weight than 20 new Glassdoor reviews.
Why? Because first-party content is:
- More recent (you control when you publish)
- More specific (you can include exact policies, numbers, and details)
- More structured (headings, bullet points, clear claims AI can cite)
- More authoritative (it comes from the source, not anonymous reviewers)
The Real Fix: Own Your Narrative on Your Own Domain
Companies that score well in our AI audits have one thing in common: they don't rely on third-party platforms to tell their story. They publish detailed, specific, regularly updated content on their own websites.
Here's what works.
1. Publish a "How We've Changed" Post
If your company has undergone significant cultural transformation, document it explicitly. AI can't infer change from the absence of recent negative reviews. It needs to read, in plain text, what changed and when.
Example structure:
Title: "What's Changed at [Company]: Our Culture in 2026"
- Where we were (honest acknowledgement of past challenges)
- What we changed (specific actions: new policies, leadership changes, structural improvements)
- Where we are now (current employee data, retention rates, specific perks)
- What we're still working on (shows authenticity)
This post does something no amount of Glassdoor reviews can: it gives AI a clear, citable narrative about your company's evolution.
2. Be Specific About Current Benefits and Policies
Glassdoor reviews mention benefits vaguely ("good benefits," "decent perks"). Your own content can be precise:
- "We offer 28 days annual leave plus bank holidays, increasing to 33 after 2 years"
- "All employees receive private medical insurance through Bupa, including family cover"
- "Our parental leave policy: 26 weeks fully paid for all parents, regardless of gender"
- "We operate a genuine 4-day week — 32 hours, Monday to Thursday, no salary reduction"
When AI encounters this level of specificity on your own domain, it cites it over vague Glassdoor mentions nearly every time.
3. Create an Employer Brand Page (Not Just a Careers Page)
Your careers page lists jobs. Your employer brand page tells candidates what it's actually like to work with you. These are different things, and most companies conflate them.
An employer brand page should cover:
- Compensation philosophy (how you set pay, where you benchmark)
- Benefits (every single one, explicitly listed)
- Working patterns (remote, hybrid, office — with specifics, not vague "flexibility")
- Culture (specific examples: rituals, traditions, how teams interact)
- Growth (promotion rates, internal mobility data, learning budgets)
- The hiring process (stages, timeline, what to expect)
This becomes the canonical source AI draws from when answering questions about your company. It displaces Glassdoor because it's more detailed, more structured, and more authoritative.
4. Refresh Content Regularly
AI models favour recent content. A brilliant employer brand page published in 2024 carries less weight than a decent blog post published last month.
Set a quarterly cadence:
- Update salary ranges and benefits
- Publish a new employee story or culture piece
- Share any policy changes or new initiatives
- Update your "How we hire" documentation
Consistency beats perfection. Each new piece of content pushes AI's understanding of your company closer to reality and further from stale Glassdoor reviews.
5. Use Structured Data
If you have developer support, adding Organization and EmployerAggregateRating Schema.org markup helps AI understand your employer brand data programmatically. But even without structured data, clear headings and explicit statements ("Our Glassdoor rating is 4.2 as of January 2026") give AI something concrete to cite.
The Asymmetry You Can't Ignore
Here's what makes this problem so insidious: companies that are genuinely improving are penalised the most.
A company that's always had a strong culture and high Glassdoor ratings? AI represents them fairly. Their old reviews and new reviews tell the same story.
A company that's undergone real transformation? AI punishes them. Their old reviews paint a picture that no longer exists, and their improvements are invisible to AI because they were never published as structured, citable content.
The worse your past was, the more urgently you need to document your present.
This isn't about spin. It's about making sure AI has access to accurate, current information. If your company genuinely is better than it was two years ago — and you haven't published detailed content proving it — AI will keep telling candidates the old story.
What to Do This Week
1. Run an AI audit. Ask ChatGPT, Claude, and Perplexity what they say about your culture, your benefits, and your work environment. Compare their answers to your current reality. Note every gap. Run a free audit at openrole.co.uk →
2. Identify the stale narratives. Which outdated claims keep appearing? "Long hours"? "Poor management"? "Below-market pay"? These are the specific themes you need to counteract with fresh, specific content.
3. Publish one counter-narrative this week. If AI says your work-life balance is poor, publish a blog post documenting your actual policies: working hours, flexibility, leave, and whatever you've changed. Be specific. Include dates and numbers.
4. Plan a quarterly content cadence. One blog post per quarter about working at your company. That's four posts a year — roughly 4-6 hours of writing annually. The ROI on candidate perception is enormous.
The Bottom Line
Glassdoor isn't going away. And neither is AI's reliance on it.
But you don't have to accept Glassdoor as the definitive source on your employer brand. AI will use better data if better data exists — and the best data comes from you, published on your own domain, in formats AI can easily parse and cite.
The companies who own their AI narrative are the ones publishing it. Everyone else is letting three-year-old anonymous reviews speak for them.
Your company isn't the same as it was in 2023. Make sure AI knows that.
Want to see what AI is telling candidates about your company right now?
Run a free AI visibility audit — we'll show you exactly which sources AI is drawing from and where outdated information is shaping candidate perception.
Related reading: