Your first introduction to a candidate is via a language model. You do not listen in, you do not intervene, and yet that is where the judgement is made.
New research by PerceptionX (May 2026, 306 job seekers across seven countries) shows how far advanced this already is. Of the job seekers who use AI, 96 per cent research employers with tools like ChatGPT, Gemini and Claude. 74 per cent do so as standard, at every serious step.
So before anyone opens your vacancy, before anyone sees your careers site, that person has already discussed your company with a model.
And that model has an opinion.
82 per cent of respondents say that AI has changed their perception of a company. The conversation with the machine now does what your recruiter, your campaign and your referrals used to do.
The model has an opinion and that isn't always right
This is where it gets awkward. 58 per cent caught AI providing incorrect information about an employer. The wrong salary, an outdated culture, a reorganisation from three years ago, a merger that never went ahead. The model fills in the gaps, and it often fills them in wrongly.
What is missing will be filled in. What is unclear will be simplified. And what is nowhere stated does not exist.
That is how a language model works. It looks for the most probable answer, not the right one. It reads what is written about you, it weighs the sources it trusts, and it makes a summary. If you do not provide those sources, someone else will: an old Glassdoor thread, a forum post from 2022, a competitor whose narrative is in better shape.
What candidates question the model about
The research also shows what candidates are asking about. Salary (53 per cent), career progression (52 per cent), the application experience itself (51 per cent). Precisely the topics your official communication usually remains vague about. The model does not remain vague. It provides an answer, with or without your input.
Take a candidate who is hesitating between two employers. She asks ChatGPT which of the two is a better fit for someone with her experience. The model names your competitor, with a reason, and not you. That candidate never applies, never calls, never lets you know she has been by. You lose someone you have never seen.
ChatGPT drives this behaviour. 89 per cent of AI-using jobseekers tackle that first. In India and Brazil, Gemini is neck and neck. Google AI Overviews stands at 27 per cent and grows with every month that the search page features more AI and fewer blue links.
Compare that with where the employer branding budget goes. To the careers site that nobody visits first. To the campaign that the candidate doesn't read. To the slick video that ends up in no answer at all.
Two types of employers remain. The employer that can cite the model, with clear, consistent, verifiable information. And the employer that has to guess the model, with scattered signals and outdated sources. The first gets into the conversation. The second does not exist there.
This is not SEO 2.0.
With SEO, you want a click. With a language model, you want to be the answer. Google sends the candidate to your page; ChatGPT summarises you and the candidate never reads your page. This calls for different choices regarding how you write, structure and distribute information.
What works is less complicated than it sounds.
Make sure the basic facts about your employment are correct in one place and the same everywhere: what you do, how you work, what you offer, how the application process goes. Models rely on consistency. Three versions of the same story means doubt three times over.
Write that information so that a model can process it. Short, factual passages that directly answer a direct question. Marketing fluff that looks nice and says nothing is useless to a model.
Be present in the places where the model gets its sources. Your own site, as well as the external sources that AI trusts when it talks about employers.
And measure what is said about you. Ask the question that a candidate also asks: “I have 4 years of experience as a [job title]. Which employers in [region]suit me, and why?”
Run that question ten times, with slight variations, in ChatGPT, Gemini and Perplexity. What the model repeats every time is your real reputation in AI. The answer is often a shock, and it is the first honest signal you have had in months.
The shift is not in the distant future. 65 per cent of respondents expect to lean even more on AI within twelve months. 77 per cent would even leave the searching to an AI agent. Next year's candidate will talk less to you and more to a model that talks on your behalf.
You can get angry about that. You can think it's not going to move that fast. Or you can make sure the model gets it right when it talks about you.
The conversation about your business is already taking place. The only question is whether you feature in it.
My name is Bryan Peereboom, founder of AI Rebels. We make employers visible in language models.