A candidate decided not to apply to you last week. You didn't see it. No rejection, no email, no missing application that you could count. She asked ChatGPT a question, got an answer about your organisation, and moved on.
Perhaps that answer was correct. Perhaps not.
In May, PerceptionX surveyed 306 jobseekers across seven countries. 96% of those who use AI do so to research employers. 72% do this before deciding to apply for a job. And 40% discover employers in this way whom they were not previously aware of.
40% helps candidates get to know potential employers through a language model. Not via your ‘work with us’ page. Not via a job advert. Through a conversation you aren’t part of.
Searching became asking, and to a question AI gives a single answer
Search behaviour itself is shifting. Earlier this year, SparkToro found that 68% of Google searches in the US end without a click. Two years ago, that figure was 60%. Where an AI Overview appears, the number of clicks falls by over a third. People get their answer on the screen and carry on.
Searching became asking. And with a question, AI gives one answer, with names in it. No ten blue links for the candidate to choose from themselves.
And the questions are precisely those of someone who is orientating themselves. Which employer in my sector has the best culture. Where can I really progress. Which company suits someone with my experience. Those are not questions about your brand name. Those are questions from people who do not yet know you, in the part of the funnel where most employers have nothing set up. That is precisely where the choice is made.
The bit that hurts
Now for the bit that hurts. In that same survey, 82% said that AI had changed their perception of a company. And 58% had caught AI providing incorrect information about an employer.
More than half. An incorrect salary, a reorganisation that is long past, a cultural image that makes no sense, a merger from five years ago that is talked about as if it happened yesterday.
In July, the Harris Poll added a further layer to this. 72% of people know that AI can sound confident even when its advice is wrong. They know this. And yet they still act on it, because the answer comes fluently and at just the right moment.
AI has become the first recruiter.
It filters candidates before you see a CV. It answers “is this a good employer” before someone clicks on your job vacancy. And it does that based on what it finds about you on the web, not based on what you think you are projecting.
For a language model, there are two types of employers. The employer that can read, understand, and quote it. And the employer that is not clearly stated anywhere, and is therefore filled in or skipped.
What is still unclear on your site today will be summarised by a model tomorrow. What is summarised tomorrow will be repeated next month. And what is repeated often enough passes for truth.
The impression you never see
Here lies the real problem. You cannot correct an impression that you never see. A bad Glassdoor review can be answered, in public, with your name attached to it. A wrong answer in a private conversation between a candidate and Gemini remains beyond your reach. You do not know that it was given, you do not know to whom, and you do not know what was in it.
People are not search results. But to the model talking about you, right now you are precisely that: the snippets of text it can find about your organisation, and the confidence with which it strings them together. Your most expensive employer branding film weighs less than three lines on a forum that the model can actually read.
That affects everyone working in labour market communication. The recruiter wondering why applications for a strong vacancy are disappointing. The employer branding manager spending months building a narrative that the model doesn't pick up. The executive board thinking a nice careers site is enough, while the first impression was formed somewhere else long ago.
What you are doing today
You start by measuring. What does AI say about you today, to which questions, on which platform, with what sentiment. Without that baseline measurement, you are steering by gut feeling, and a gut feeling is precisely what a language model does not read.
After that, you set up the basics. Crawler access, correct technical foundations, content that genuinely answers a question rather than evading it with corporate jargon. A language model quotes what it can easily summarise and trust. Corporate buzzword bingo doesn't make the cut.
And then you keep measuring. AI answers change every month. Today's edge will be an average in a quarter's time, and today's lag will be a habit in a quarter's time.
Last week's candidate isn't coming back to ask if it was correct. She already has her answer. The only question that matters is whether you gave that answer, or if a model filled it in for you.
My name is Bryan Peereboom, founder of AI Rebels. We make employers visible in language models.