AI has already decided what kind of employer you are. It did so without ever opening your careers site.
That is the result of the largest measurement to date of how language models arrive at their answers. In May 2026, the agency 5W released its AI Platform Citation Source Index: 680 million citations from ChatGPT, Google AI Overviews, Perplexity, Gemini and Claude, measured from August 2024 to April 2026.
One source tops the list for all major models: Reddit. It accounts for around 40% of the citations. Wikipedia leads the way within ChatGPT, accounting for between 26 and 48% of the top ten sources. YouTube dominates Google AI Overviews with a lead that the researchers describe as 200 times greater than that of any other video source.
And then there’s the figure that matters most to you: fifteen domains account for 68% of all citations. Fifteen. Your careers page doesn't belong there.
That hits employers harder than marketers. Your careers page was the one channel you had completely in your own hands. Exactly that channel has just been relegated to a supporting role.
What does this mean for your organisation?
You have probably invested heavily in employer branding over the last few years. An EVP, campaigns, a fresh careers site, maybe even an extra recruitment marketer. That investment only lives on channels that the model hardly reads. The work is good. It is in the wrong place for the people who are asking the questions now.
This summer, Built In published its 2026 State of Employer Reputation and Visibility report. 91% of talent leaders say that AI tools determine how candidates discover and research them. 62% describe this influence as significant. Three in four candidates use AI searches during their job search, including those who are simply browsing.
Where does that image come from? 48% of the quotes come from platforms such as Reddit and YouTube. 85% come from content that isn’t yours.
Then put these three figures side by side. One third of employers believe their brand appears correctly in AI answers. Three-quarters are afraid that candidates are getting outdated or incorrect information. One in three feels ready to make content suitable for AI search engines.
One third trusts it. Three quarters are worried. One in three is prepared. That is a blind spot the size of your entire employer brand.
That is how it works in practice
A candidate asks ChatGPT: “Is company X a good employer?” The model doesn't write an answer based on your carefully built EVP page. It grabs a two-year-old Reddit thread, a few reviews, a line on Wikipedia and a YouTube video someone made once. Out rolls a judgement. And that judgement is the shortlist, even before the candidate has seen a single vacancy of yours.
The old funnel had awareness, interest and consideration as separate steps. Those three have been collapsed into a single AI answer. That answer has been written on your behalf.
Things aren’t standing still either. In the 5W data, Reddit’s share within ChatGPT fell from around 60% to 10% in six weeks, following a single adjustment to a Google parameter at the end of 2025. Forbes, Medium and news outlets picked up on the difference. What is your source of truth today may be replaced in a month and a half. This isn’t something you sort out once and then forget about.
Let's make the difference clear. SEO was about ranking in Google and getting clicks to your site. GEO is about being cited in the answer itself, so the candidate gets your story without having to click anywhere. Almost everyone plays the first game. Candidates are already playing the second, and most employers aren't yet.
To a language model, there are now only two types of employers. The employer with consistent, quotable information in the places where the model looks. And the employer the model encounters nowhere, and therefore does not mention either. The former makes the shortlist. The second does not exist in that conversation.
What you can do about this won't take six months.
Start measuring what AI says about you today; you cannot manage what you do not see. Then look at the places where the model gathers its judgement: the Reddit and professional forums where people talk about you, the review sites that candidates still read, and the Wikipedia and video mentions that others made about you. And keep track of it, because the sources shift faster than your job vacancy text ages.
What is still an advantage today will be the baseline in a year's time. What is still invisible today will cost you the best candidates in eighteen months. And what is still called “we'll do that later” today, your competitors will already have done by then.
The candidate is asking AI. The question is whether you are the answer.
My name is Bryan Peereboom, founder of AI Rebels. We make employers visible in language models.