Your Employer Brand Has a Ghost

Jul 23, 2026

The short version: your best candidates research you through AI before they apply. If the model describes a company you stopped being three years ago, they self-select out before your recruiter ever sees them. It shows up as a thinner pipeline and softer offer acceptance, for reasons that never appear in an exit interview.

Ask a strong candidate how they research a company now and they will not start with your careers page. They open ChatGPT and ask what it is like to work there, whether the company is stable, what it pays, and whether it is going anywhere. The model answers with confidence. That answer forms their first impression, and it does it before they have read a single word you wrote about yourself.

Here's the trap: The story the machine tells about you as an employer is assembled from whatever is circulating: old Glassdoor threads, a layoff headline from a different era, a funding round that is no longer your latest, a tech stack you have since replaced. The company that ghost describes is not the one you run today. But it is the one the candidate meets first.

 

Why the machine's version lags reality

Employer reputation is unusually prone to legacy signals. The internet keeps a long memory of how a company used to be, and that memory skews negative and sticky. A rough patch three years ago generates content that outlives the fix. A reorg leaves a trail that a turnaround does not erase.

Machine interpreters treat all of it as current. They do not know your culture changed, your leadership changed, or your trajectory changed unless the signals they read reflect it, and those signals almost always lag. This is the failure mode we call legacy interpretation, and it is the most common and most invisible of them all. We cover it in depth in What Is Legacy Interpretation and Why Does It Cost You?.

 

What the ghost costs you


The no you never hear

This is the expensive part. When a candidate rules you out based on an AI answer, they do not send an email explaining why. They simply never apply, or they take the other offer. The rejection is invisible. You cannot fix a pipeline problem you cannot see, and a misaligned machine interpreter produces exactly that: attrition at the top of the funnel with no fingerprints on it.


A quality gap, not just a volume gap

The candidates most likely to do this research are the ones you want most. Strong people have options and do their homework. They are precisely the ones who will quietly cross you off on the strength of a stale description. So the damage is not spread evenly. It concentrates at the top of your target list, widening the gap between the talent you attract and the talent you need.


Softer offer acceptance

Even candidates who make it through arrive shaped by what they read. If the machine seeded doubt about stability, growth, or culture, your recruiters and hiring managers spend interview time correcting context instead of selling the role. Offers get declined for reasons nobody can name, and the loss gets written off as compensation or timing when the real cause was set before the first call.


Why internal comms cannot outrun it

Most leaders respond to a culture problem by talking more internally: all-hands, values decks, manager briefings. That work matters, but it cannot reach the person who has not applied yet. The candidate forming an impression through AI is outside your internal channels entirely.

There is an internal cost too. New hires now onboard by asking a model about their own employer, and current employees do the same when they want to understand strategy. When the machine's version does not match leadership's, you get quiet internal confusion that no amount of internal communication fully resolves while the external interpreter network stays misaligned.

What talent teams can do, and where it needs leadership

Some of this is squarely in recruiting's reach. Keep the sources candidates actually check current: your careers content, your leadership bios, your presence on the profiles that feed the models. Retire or correct the specific legacy items, the old headline, the outdated review context, that anchor the wrong story. Make the true version of the culture specific and provable rather than a page of adjectives.

But employer brand is only one slice of how you are interpreted, and the machines do not keep it in a separate box from how you are described to buyers and investors. A candidate and a customer often get the same drifted answer. Fixing it durably means aligning the whole interpreter network, not just the careers page, which is why it usually needs an owner above any single function. That larger discipline is Model Response Optimization.

See your employer ghost in 15 minutes

Before you plan a fix, look at the problem. Open ChatGPT, Gemini, Claude, and Perplexity and ask each, as a candidate would:

  1. “What is it like to work at [company]?”
  2. “Is [company] a good place to work? Is it stable and growing?”
  3. “What do employees say about [company]'s culture and leadership?”

Read the answers as a candidate weighing your offer against another. Mark what is wrong, what is missing, and what is years out of date. If it stings, that is the point. For the full version of this exercise across your whole brand, not just the employer angle, run the 30-minute mini-BIA.


Frequently asked questions


Do job candidates really use AI to research employers?

Yes, and increasingly it is their first step, ahead of your careers page. Strong candidates in particular ask AI what a company is like to work at, whether it is stable, and what it pays before they decide to apply, so the model's answer shapes the impression your own content never gets to make first.


Why does AI describe my company as an employer inaccurately?

Because it assembles the answer from signals that skew old and sticky: past reviews, dated headlines, a reorg or layoff from a previous era, an outdated tech stack. Machine interpreters treat those legacy signals as current unless newer, consistent signals overtake them.


How does this actually hurt hiring?

In three ways: candidates rule you out silently so you never see the loss, the effect concentrates among the high-quality candidates most likely to do research, and even those who apply arrive with doubts that soften offer acceptance.


Can't we fix this with better internal communications?

Internal comms cannot reach a candidate who has not applied yet. It also cannot outrun a misaligned external interpreter, since your own new hires and employees increasingly form impressions by asking AI. The fix has to address the external network, not just internal channels.


What is the first thing we should do?

Spend 15 minutes reading what the major models say about working at your company, mark what is wrong or outdated, and correct the sources you control. Then decide whether the pattern warrants mapping the full interpreter network.


The bottom line

You compete for talent against a version of your company that no longer exists, and it gets to speak first. The candidates you want most are the ones most likely to hear it. Until you read what the machines are telling them, you are recruiting against a ghost you have never met.

Start by seeing the network the way a candidate does. Request your Brand Interpretation Assessment, or read the Interpreter Architecture whitepaper for the full thesis.