LinkedIn lets an AI agent filter out 80 per cent of candidates before a recruiter opens the first profile.
That figure comes from LinkedIn itself, as it rolls out its Hiring Assistant widely. Prashanthi Padmanabhan, responsible for the technology behind LinkedIn Talent Solutions, says that recruiters using the agent need to review 80 per cent fewer profiles. The acceptance rate of their messages rose by 66 per cent.
Here is how it works. A recruiter types an assignment in plain language, or pastes the job vacancy. The agent searches for candidates on LinkedIn and in the systems linked to it, hundreds of ATSs such as Greenhouse and Workday. It learns which profiles the recruiter approves and which they don't, and hones its selection. The recruiter only looks at what the agent has already pre-sorted.
For recruiters, this is a breath of fresh air. For the rest of the job market, it's a shift.
Two sides, one mechanism
Look at what is happening. I wrote here earlier that candidates already form their judgement about you as an employer in a chat window, before they open your job vacancy. The candidate outsources the research about you to a model.
Now the recruiter is doing exactly the same on their side. They are outsourcing the search for the candidate to a model.
On both sides of the table, there is now a system that makes the initial selection. And that system works in only one way: it reads what is there, and it skips what is not there.
That strikes at the heart of how the labour market works now.
What is readable counts. The rest doesn't.
LinkedIn says it outright. A candidate without a profile, without a digital footprint, does not exist for the agent. Such a person simply never features in the selection. The same mechanism applies on your side of the table. An employer that cannot be clearly read by a model does not feature as an option within that model.
What is legible is included. What is vague is omitted. And what is not structured anywhere does not appear in any selection.
The tool is with the recruiter. The consequences are with you.
Most employers think this is a recruiter's tool and therefore not their problem. That is the costly mistake. The tool is with the recruiter, the consequences lie with you. The roles that the agent manages to match well are the roles with a clear, consistent, machine-readable description. The employers that a model can summarise with confidence are the employers about whom the model dares to say something. The rest are quietly skipped, in a process that nobody at your place will ever see.
And LinkedIn isn't the only one moving in this direction. OpenAI is building its own jobs platform, announced by Fidji Simo, which aims to connect companies and workers directly based on what a company needs and what someone can do. Yet another layer where a model makes the match. Yet another place where your employer brand needs to be machine-readable, otherwise it doesn't count.
Why buying more visibility doesn't fix it
I worked in recruitment for fifteen years. The old reflex during a labour shortage was to buy more visibility. A bigger campaign, a more expensive placement, a more eye-catching post. That reflex doesn't work here. An agent isn't impressed by a flashy video. It reads the text, the structure and the mutually corroborating sources, and draws its conclusion in milliseconds.
What does count is whether your story is the same everywhere. Whether your vacancies are written in such a way that a model can read and return them. Whether what external sources say about you aligns with what you put out yourself. Whether you are recognisable as a single employer, with one clear profile, in the places where the models look.
That is the job. We call it GEO: making sure that a generative model can read and cite you correctly. Until recently, that was about the candidate who asked ChatGPT about you. From now on, it is just as much about the recruiter agent searching on the other side, and about every job board that now has a model make the match.
Start measuring
Look at what the models are now returning about you as an employer, in the questions candidates are actually asking, and see how your vacancies look to a system that processes them automatically. Only when you have that in black and white will you know whether you have a visibility problem or a reputation problem. Two different fixes, and most employers guess the wrong one.
This is accelerating, not slowing down. What is still a manual search today will be an agent doing the pre-selection in six months' time. What is still pre-selection today will be the place where the decision is actually made in a year's time. And what still feels optional today will be the prerequisite for even being considered in eighteen months' time.
A model will soon be sitting on both sides of the table. The only thing you can still determine is whether it can read you.
My name is Bryan Peereboom, founder of AI Rebels. We make employers visible in language models.