OpenAI is building its own job board. Inside ChatGPT, where hundreds of millions of people now ask their questions every week. The launch is scheduled for mid-2026, and the first names participating are no small fry: Walmart, John Deere, Accenture and Boston Consulting Group.
The promise is bare and clear. AI matches what a company is looking for with what a candidate can do. No search bar, no 200 results, no afternoon of scrolling. One conversation, and the match rolls out.
The same company as both trainer and matchmaker
A second track is being added to that. OpenAI wants to certify 10 million Americans in AI proficiency by 2030, ranging from basic use in the workplace to prompt engineering. In this way, the same company becomes both the educator and the matchmaker. The model gets to know the candidate, and the model decides which employer is the right fit.
Candidates are already asking, now OpenAI is building that behaviour into the product
That is the absolute core of the matter for employers. The party that will soon decide whether you get put forward is a language model that chooses which names to mention. No recruiter, no job board.
This is part of a trend that has been ongoing for months. In the first four months of 2026, 68% of Google searches in the US ended without a single click, according to research by SparkToro based on Similarweb data. Two years earlier, the figure was 60%. AI Overviews now appear above more than 20% of searches, and where they appear, the click-through rate drops by almost 60%.
People no longer search in the way you are used to. They ask. And they get an answer back, not a list.
Candidates were already doing that on their own with questions like “which employer suits someone like me”. Now OpenAI is building that behaviour directly into the product. The question “where can I work” and the answer “at these three companies” will soon be in the same chat.
Two types of employers: those who know the model and those who can't place it
The employer that knows the model. Clear, consistent information about who you are, what you do and what you stand for, in the places where AI gets its answers.
And the employer who cannot place the model. Fragmented texts, an outdated careers page, conflicting signals.
The first is mentioned. The second does not exist in the conversation.
What is still a fun experiment today will be a channel in six months. What is still a channel today will be the first place a candidate looks in a year's time. And what is still optional today will be where the shortlist is made in eighteen months.
Those certificates make that picture even sharper. Soon, a generation of candidates will come in who have been trained and assessed by OpenAI itself, with a quality mark that the model recognises. If you as an employer want those people to see you, the model has to recognise you just as well as it recognises that candidate. Otherwise, you'll fall outside the match before anyone has even looked at your vacancy.
This starts in the US, but the Dutch candidate uses the same ChatGPT
This starts in the US, with American names first. But the behaviour doesn't stop at the border. The Dutch candidate uses the same ChatGPT, asks the same question and will soon expect the same answer. Waiting until it officially lands here means you start at the moment your competitor has already been visible for a year.
LinkedIn has organised the labour market around profiles and connections for 15 years. OpenAI organises it around a conversation. The difference lies in the outcome: a conversation only returns a few names. On a job board, you are on page 4 and still get found. In an AI answer, you are either in it, or you are not in it.
This is going to happen. The only open question is whether the model will soon be able to tell a story about you that is accurate.
You don't build that story with an advertisement and not with a slick campaign film. You build it with information that an AI can retrieve and dares to quote. A careers site that machines can read. Vacancies with a structure that a model understands instead of PDFs that it skips. Consistent signals about your culture, your work and your people on the sources where ChatGPT, Gemini and Perplexity pull their answers from.
That is substantive work, not a marketing layer. A model does not quote an atmosphere. It quotes what is concrete and verifiable and appears identically in multiple places.
what being quotable means in practice
Herein lies the reason why “we already do SEO” isn't enough. Search engine optimisation moves you up a list. Generative Engine Optimization ensures you end up in the answer the candidate actually reads at all. The job board shows ten employers. The language model names three. That is a different playing field, with different rules.
This is what we at AI Rebels are working on every day. Being visible and citable in AI answers, not just findable in Google. Because the channel that OpenAI is now building a jobs platform on is precisely the channel where most employers are still completely invisible today.
Mid-2026 is close. And the companies that are already in those answers are building an advantage that you cannot catch up with through a single campaign. The model remembers who is consistent and reliable. You cannot buy that trust; you earn it over time.
Kim, and everyone like Kim, will soon be asking their question to a model. The employers who are mentioned then will have already done their homework last year.
My name is Bryan Peereboom, founder of AI Rebels. We make employers visible in language models.