Of everything a language model says about you as an employer, only 15 per cent comes from your own website. The other 85 per cent comes from Glassdoor, Indeed and Reddit.
That figure is in the white paper that Randstad published on 20 August about employer brand visibility in AI. It comes from the State of AI Search research by Airops. And it turns fifteen years of employer branding on its head.
You've polished your careers site. Your EVP is looking sharp. You've got videos, employee stories, a career blog. All yours. All on the sidelines.
Because the candidate doesn't ask you anymore. They ask ChatGPT. And ChatGPT builds its answer from the places where you aren't holding the pen.
What the candidate is doing today
Kim, Ibrahim, or Fatima, is no longer searching for “best fintech employers Amsterdam”. They type: “I'm a developer, young family, I value psychological safety. Which fintechs in Amsterdam actually live up to their culture promises?”
The model reads that question and searches its sources. Your careers site is one of them. The Glassdoor review from three years ago is another. The Reddit thread about your application process counts just as heavily.
What the model returns is a summary of all those voices combined. Your voice is one of them, and usually not the loudest.
The figures that employers actually feel
Built In deed begin 2026 onderzoek onder 162 TA- en HR-leiders. 91 procent zegt dat AI-tools nu zwaar meewegen in hoe kandidaten werkgevers ontdekken.
33 percent are convinced that their brand is accurately represented in those AI answers. 73 percent are worried that candidates are getting outdated or incorrect information. And 35 percent feel prepared to ready content for AI search.
Read those figures together. Nearly everyone sees that it's happening. One in three thinks what is being said about them is true. One in three is ready for it.
That gap is where the work lies.
There is another layer beneath that. AI can retell something old or incorrect about you with the certainty of a fact. Think of a reorganisation from two years ago, a salary indication that is no longer correct, a cultural narrative that you yourself revised a long time ago. The candidate no longer weighs those sources themselves. The model has already done that and returns the result as the conclusion.
Why your career site isn’t enough
The reflex is predictable. More content on your own site. Another story, another page, another EVP revision.
That helps, but it only accounts for that 15 per cent at most. You’ll have to earn the other 85 per cent elsewhere.
An employee who shares on LinkedIn what the work is really like reaches a network that is often ten times larger than your own follower base. An honest, recent review on Glassdoor carries more weight than three coats of polish on your homepage. A well-answered question in a professional forum remains citable for years.
What the model picks up is consistency between those places. If your site says one thing and Glassdoor says another, the model doesn't take your side. It averages out. And in that average, you disappear.
The two types of employers
For a language model, there are now only two types of employers.
The first one is findable. The information is accurate, is roughly the same everywhere, is up to date, and comes from sources that the model trusts. That employer is mentioned when a candidate asks who is a good fit for her.
The second is invisible. Often a fine employer, but with a story that is fragmented, outdated, or nowhere to be found. That employer never comes up in conversation. And what does not come up in conversation does not exist in the dialogue the candidate is having.
Whoever understands where the conversation is taking place right now is on the right side. That takes insight, not a big budget.
What you can do this month
Start measuring what AI says about you today. Ask the questions that your candidates ask to ChatGPT, to Gemini, to Perplexity, and read what comes out. Usually, that is a confronting half an hour.
Next, look at your external sources. Is your Glassdoor profile still up to date? Are reviews from 2023 at the top? Are you represented well on Indeed? Do your employees speak up on their own, or only your marketing department?
Activate the employees. Ask them for real stories about real work. Every model immediately spots a template post that everyone shares on the same day as noise. Individual, honest stories are the cheapest and most credible way to win over that 85 percent.
And make sure your job data is clean. Google overhauled its search experience into AI Mode, which now has a billion users a month. AI no longer displays job listings without current, structured data as a neat line on a list; it summarises them and filters them out. An expired vacancy that is still online is noise in your narrative.
Many people call this SEO 2.0. That is incorrect. With SEO you fight for a place in a list; with AI you fight to be mentioned in an answer that the candidate reads as truth.
The employers who are tackling this now will be in that answer in a year's time. The rest will be explaining why they are not in it.
My name is Bryan Peereboom, founder of AI Rebels. We make employers visible in language models.