You spend a lot of effort on how buyers see you in AI search. There is a second audience running the exact same play against you, and almost nobody has looked at it: the people you are trying to hire.
A strong candidate does now what a buyer does. Before they read your careers page, before they talk to a recruiter, sometimes before they decide whether to apply at all, they open ChatGPT and ask what it is like to work at your company. Are the layoffs over. Is the remote policy real or is that just the job post. Does it actually pay market. Is the CEO someone people respect. They get a confident, synthesized answer in a few seconds, they form an opinion, and you never find out the question was asked.
Everything you have learned about buyers in AI search applies here, with one difference that makes it sharper. Your buyers are mostly asking about your product. Your candidates are asking about your people, your stability, and your reputation as an employer, and the honest answers to those questions are often things you would rather not have summarized for a stranger.
The talent funnel moved into the chat too, and you have less control over this one
You already know the buyer decides a lot before the demo, inside a conversation you cannot watch. The hiring version is the same shape and, if anything, worse for you, because the source layer for employer questions is even less under your influence than the one for product questions.
Think about where an engine goes to answer “what is it like to work at this company.” Glassdoor and its reviews. Blind and its anonymous threads. Layoff trackers and the news coverage around them. LinkedIn posts from people who left. Reddit. Old press about a bad quarter. Your careers page is in the mix, and it is the least-trusted item in it, for the same reason your homepage is the weakest input on product questions: it is you describing yourself, and the engine has learned to discount that. A polished culture page does not outweigh a consistent Glassdoor theme, because reviews in aggregate read as corroborated and the vendor’s self-description does not.
So the candidate is getting a picture assembled mostly from sources you did not write and cannot edit, about topics more sensitive than anything in the buyer conversation. And unlike a lost deal, you get no signal at all. A buyer who crosses you off at least sometimes tells a rep why. A candidate who reads a discouraging answer and never applies is simply someone you never hear from. Your pipeline of applicants quietly thins and nothing in your recruiting dashboard tells you the reason.
The failure modes are the same two you already know, aimed at a rawer target
Everything you have seen on the product side has an employer-brand twin.
The wrong-fact version is the easy one to underestimate. A candidate asks what you pay, or whether you are still remote, or whether the layoffs are ongoing, and the engine states something outdated as current fact. The pay band it quotes is three years old. The remote policy it describes was reversed, or reinstated, and it has the wrong version. The layoffs it mentions concluded eighteen months ago and it presents them in the present tense. Each of these is checkable and fixable, and each one costs you candidates silently until you fix it, because a person deciding whether to uproot their career reads a stale negative as current and moves on.
The true-but-incomplete version is the harder one, and it is the negative-but-true problem in its most acute form. You did have layoffs. There was a rough stretch. A visible person did leave under a cloud. You cannot correct any of that, because it happened, and you should not try to bury it. But the same dynamic holds: the event is thoroughly on the record and the recovery usually is not. The engine describes the hard quarter accurately and says nothing about the eighteen good months since, because the hard quarter was news and the recovery was not. The candidate hears the worst true thing about you with none of the context that would let them weigh it.
Neither of these is a careers-page copywriting problem, and treating them as one is why employer-brand teams pour effort into the one surface that changes nothing.
Your recruiters and your exit interviews already hold the data
The good news is that you are almost certainly already collecting the signal and throwing it away.
Candidates say this stuff out loud. In screening calls they mention what they read. In the ones who decline an offer, the reason sometimes traces straight back to something an engine told them. Your recruiters hear it and, like your sales reps hearing AI-sourced product objections, they log none of it in a form anyone can act on. Start there, because it is nearly free: one field on the recruiter’s notes for when a candidate cites something about the company they did not get from you, captured in their words. Repeats are your work queue, ranked by how much talent they are costing you.
Then find out what the engine actually says, on purpose rather than by accident. Run the real candidate questions yourself, across engines, the same way you would run buyer queries. What is it like to work here. Do they pay well. Are they stable. Is leadership any good. Then audit which sources those answers are being pulled from, because that tells you whether you are looking at a wrong fact you can correct at the source or a true theme you have to answer with a documented recovery.
What I got wrong
I treated this as a marketing problem and handed it to the wrong team, and it cost a client a hiring quarter before I understood what I was looking at.
They came to me because applications had dropped and the recruiting team could not explain it. Nothing had changed about the roles or the comp. My first instinct was the obvious one: the employer brand needs work, so let us make the careers page better, sharpen the culture messaging, publish some employee stories. We did, and it did nothing, because I had aimed the entire effort at the surface with the least influence over the answer, exactly the mistake I had already learned not to make on the product side and somehow made again the moment the topic was hiring.
When I finally ran the actual candidate queries, the problem was plainly not the careers page. The engine was leading with a layoff that had happened and concluded well over a year earlier, describing it as though it were current, with nothing about the sustained hiring and stability since. Every strong candidate researching them was being handed a frightening and out-of-date picture before they ever reached the page we had spent a month polishing. The recovery was real and it was nowhere on the record.
The fix was the same source-layer work as everything else. The client published a plain, dated account of where the company actually stood on headcount and stability. We got a piece of independent coverage that reflected the current state. Within a refresh cycle the engines started describing the layoff as a past event with a recovery attached, and applications recovered over the following months. The careers-page work had been effort spent on the wrong layer, again, and the thing that moved it was correcting what the outside record said.
The lesson I kept is that your employer brand in AI search is not made of your employer-brand content. It is made of the same third-party sources everything else is made of, and it has a second, invisible way of costing you: not a lost deal, but a great person who read the answer and quietly decided not to apply.
The short version
You have two audiences in AI search, not one, and you have optimized for exactly half of them. Candidates are running the same play buyers run, against sources even less under your control, on questions more personal than anything a buyer asks, and the cost lands as applications that never arrive rather than deals that visibly die.
Capture what candidates tell your recruiters. Run the real hiring questions across the engines yourself. Fix wrong facts at the source, answer true-but-incomplete themes with a documented recovery, and stop pouring the budget into the careers page, which is the one surface the engine trusts least.
If you want to find out what AI is telling your candidates and get the record corrected where it actually lives, that is the work we do and the method we use.