LinkedIn dropped a figure this month that puts the recruitment side on edge. Recruiters using its Hiring Assistant look at 80 percent fewer profiles before deciding who to approach.
Announced on 3 September. An AI agent that does the initial screening. It reads the recruiter’s brief in plain language, searches LinkedIn and the connected systems, and returns a shortlist. Greenhouse, Workday, hundreds of ATSs are integrated with it. TCS ran the pilot and, by its own account, is halfway through it.
“An 80 per cent reduction in the number of profiles a recruiter has to look at before deciding who to approach,” says Prashanthi Padmanabhan, VP at LinkedIn Talent Solutions. InMail acceptance rates for users went up by 66 per cent.
Read that as an employer, not a recruiter
On one side of the table there is now a model that reads candidates. On the other side there has long been a model that reads employers. Both sift through applications before a human sees anything.
We already know the candidate side. Someone asks ChatGPT which employer in their field suits them, and gets three names back. The recruiter side is new on this scale. Now an agent does the first 80 per cent of the search work, and a human only looks at what is left over.
Two filters, both from a language model. And on both you either appear, or you do not.
In the small print of the LinkedIn news is the sentence that really matters. The agent can only assess candidates who have a digital footprint. No footprint, no assessment. Anyone who is invisible online does not make the list, no matter how good they are.
Turn that sentence around and you have precisely the employer's problem. A model can only recommend an employer that it can read. No readable story, no recommendation. You could be the best place in your sector; if the model finds nothing about you that it trusts, you drop out of the answer.
This is where it pinops for most management boards. They measure their campaigns, their events, their careers site. What a language model says about them, nobody measures. It is the only channel in the recruitment mix without a baseline measurement and without an owner.
Recruitment is becoming a conversation between machines on both sides.
The recruiter's AI weighs the candidate. The candidate's AI weighs the employer. Two models do the pre-selection; the two people only meet once both filters give the green light.
And those two filters talk to each other. A recruiter will soon approach more sharply, because their agent delivers better matches. But whether that candidate says yes depends on what her model has already told her about you. An invitation from an employer she just saw passing by positively in ChatGPT is more likely to be opened than a cold email from a name she didn't encounter anywhere. That 66 per cent higher acceptance rate says enough. An email lands better if the brand behind it already meant something in the conversation that the candidate had themselves.
Your employer brand convinces a machine first
That means something awkward. Your employer brand will soon convince a machine first and only then a human. And that machine determines whether that human gets to see you at all.
What a model cannot read, does not exist for that model. An empty robots.txt, no structure, thin or slow pages, a story that differs across five places. Then you miss out on the answer, even if your brand is strong.
As far as AI is concerned, there are only two types of employers left. Those who are readable and citable, and are mentioned. And those who are not, and simply never come up.
From 7.5% to over 90%
In the audits that we run, we see just how big that gap is. An employer scored 7.5 per cent visibility on a exploratory question in their niche; on almost every variant of the question, their name simply didn't come up. Basics in order, technology, structure, quotable content, and that ran to upwards of 90 per cent. Same brand, same offering, a different answer.
The knee-jerk reaction is to wait until it has crystallised. Until LinkedIn, OpenAI and Google have finished their agents. That is the wrong reaction. The agents do not need to be finished to determine who makes the list right now. They are running today, on the data that can be found about you today.
You do not see the pre-selection happening yourself.
There is no rejection, no drop in your click count. The candidate you never saw is simply not there. The recruiter who was looking for your type of people received a list you weren't on.
What seems like an experiment today will be the norm in six months' time. What is still an advantage today will be the minimum requirement in a year's time. And what still feels optional today will be the baseline in eighteen months' time.
Give the machine something good to read
Employers who realise this are already measuring what AI says about them. Which questions they appear in and which they don't, alongside which competitors, with which narrative. They aren't waiting for the machine to choose their side; they are giving the machine something good to read.
My name is Bryan Peereboom, founder of AI Rebels. We make employers visible in language models.