A candidate considering you hasn't opened a jobs page for ages. They open ChatGPT.
She types a question in plain language. “Which healthcare employers have a good culture and real scope for career progression?” Within two seconds, an answer rolls out. Three names, four names, with a brief reason for each.
That answer is the new first impression. And you didn't write most of it yourself.
You didn't write 82% of that answer yourself.
In December 2025, Muck Rack published a statistic that every employer branding department should have hanging above their desk. 82% of the sources cited by AI systems come from earned media. From what journalists, trade publications and your own staff write about you. Your polished ‘careers’ website accounts for only a fraction of that figure.
Ahrefs measured the same pattern from the other side in August 2025. Brand mentions correlate three times more strongly with AI visibility than backlinks do. 0.664 compared to 0.218. The name that crops up in dozens of places outside your own domain beats the neat link structure your agency has spent ten years building.
The scale on which this plays out.
On 27 February 2026, ChatGPT reached 900 million weekly users. Google AI Mode has been running in over 200 countries since last autumn, including Dutch. And 69% of searches end without a single click. The person asking gets their answer and moves on. They never reach your site.
For employer branding, that rewrites the whole funnel.
The orientation phase is the stage in which the choice is made. That was already the case when people went through ten blue links. It is just that this orientation now takes place within the model, before the candidate knows your name, before they click on an advert, before a recruiter notices anything.
Furthermore, the questions are comparative. Candidates ask for the best employer in their region, in their profession, for someone with their profile. The model places you next to your competitors and chooses who it mentions. You are not at that table yourself.
If you are in that answer, then you exist. If you are not in it, you are not even mentioned.
The answer is not yet determined.
In early 2026, AirOps and Kevin Indig showed that only 30% of brands remain consistently visible across different AI sessions. Asking the same question on Monday and Thursday yields a different list, different names, and a different order. One time you are mentioned, the next time you disappear, without you having changed a thing.
For a recruiter, that is a strange thought. You are used to a job posting that stays as you published it. An AI answer doesn't behave like that. It is compiled anew each time from what the model can find and trust about you at that moment.
A sleek career site won't get you far here. AI needs to be able to find enough about you to recommend you with confidence, time and time again.
This takes place on a different playing field to SEO. SEO aims at a position in a list. Here, it is about being present in the answer itself, as a source the model dares to cite.
What we measure at ProRail
What works, we see reflected in the work we do for clients.
For ProRail, we measured across five platforms for 47 days. Approximately 9,300 AI answers. The average AI visibility comes out at 71.8%, with more than 5,000 citations of their own domain and less than 1% negative sentiment. That 71.8% comes from content that AI can easily read, summarise and trust.
Three things determine whether you are in that answer.
Technology First. A crawler that cannot or is not allowed to read your page will never cite you. An empty robots.txt, missing structure, slow or thin pages: what AI cannot read, does not exist for AI.
Content afterwards. Citeable content provides a direct answer, is attributed to a recognisable source and is also accurate independently of the page. The model grabs concrete figures, real examples and a clear sender. Vague brand language misses the mark.
Entity Last but not least – and the toughest of all. That’s precisely the 82%. Your brand must exist as a name in places you don’t control yourself. In trade media, in stories told by employees, in articles you don’t pay for. That’s the hardest thing to control and the most decisive factor in whether AI dares to mention you.
Employers who get this sorted now are building an advantage that compounds every month. The rest will find out in a year's time that a competitor is being routinely recommended in conversations they never even saw.
Start measuring
Therefore, start by measuring. You cannot manage an answer that you have never read. Ask the questions that your candidates are asking the models they are asking them to, and see what comes back. Alongside whom are you mentioned, and who takes your place the moment you are absent.
My name is Bryan Peereboom, founder of AI Rebels. We make employers visible in language models.