Candidates Ask ChatGPT About Your Company. What Does It Say?

You found the perfect candidate, sent a well-written offer, and got silence. Before replying, they did what most Polish developers now do with an unfamiliar foreign company: asked an AI chat about you. If the answer was empty, vague, or wrong, your offer went to the "probably not" pile without you ever knowing.

For a company hiring remotely in Poland without a local office, this is the new first impression. There is no building to visit and no friend-of-a-friend who works there. There is only what the public record says, and increasingly it is an AI model that reads that record aloud.

Why candidates ask AI before they reply

Because a chat answers the actual question. A candidate does not want ten links; they want to know whether your company is real, funded, and worth leaving a stable contract for. So they ask directly: "what is Acme GmbH, is it a legitimate employer, what do people say about working there?"

The answer comes back as a few sentences with a handful of facts. There is no second page of results. Whatever the model says becomes the candidate's starting point, and verification cuts both ways: you screen candidates, candidates screen you.

This hits foreign employers hardest. A well-known Polish brand can survive a thin AI answer, because candidates already know it. A company with no Polish presence cannot. The AI answer is often the only thing standing between "interesting offer" and "looks like a scam".

Where ChatGPT gets its answer about your company

AI chats do not keep a secret file on your company. They combine what the model learned in training with what web search returns at the moment of the question. For questions about a specific company, the search path does most of the work.

Two practical consequences follow.

First, the model can only say what someone has published. Your website, your careers page, press mentions, conference talks, listings in industry directories, developer reviews. Thin public record, thin answer.

Second, a detail most companies miss: ChatGPT's web search runs on Bing's index, not Google's. Why Bing quietly decides your ChatGPT visibility is a story of its own, but the short version is blunt: if your site is absent from Bing, ChatGPT cannot cite it, no matter how well it ranks on Google.

How to check what AI says about your company

A fifteen-minute test, no tools required:

1. Write down the questions a candidate would ask. Not your brand name alone — the real questions: "is [company] a legitimate employer", "what does [company] do", "[company] remote work reviews".
2. Ask them in ChatGPT, Gemini and Perplexity. Each uses different sources and answers differently.
3. Use a logged-out session. Your own account has context about you; the candidate's does not.
4. Write down what comes back: facts, wrong facts, and competitors the model names alongside you.

One caveat: chat answers vary between runs. A single check is a snapshot, so repeat the same questions after a week or two before drawing conclusions.

How to fix a blank or wrong answer

The fixes are unglamorous and they work:

• Say plainly what you do, on your own site. A clear "we build X for Y, remote-first, hiring in Poland" page gives the model exactly the sentence it needs. Marketing poetry does not.
• Make the hiring intent visible. A careers page that mentions Poland and remote work is quotable evidence that a Polish candidate's offer is real. It also helps where you post the role — job listings are public sources too.
• Get into Bing. Register the site in Bing Webmaster Tools and submit it. Ten minutes of work, and it is the entry ticket to ChatGPT's answers.
• Collect independent mentions. A founder interview, a listing in a SaaS directory, a conference appearance. The model trusts a fact repeated across unrelated sources far more than your own homepage.
• Measure instead of guessing. The manual test above is enough to start. For ongoing tracking there are tools for it; I built CiteLyzer, which checks weekly whether AI chats mention your brand when your audience asks.

The mistakes that keep employers invisible

Often the problem is not a missing public record but one the model cannot use. The same patterns come up again and again with foreign employers:

The careers page lives only inside an ATS. A jobs.lever.co/acme or Greenhouse board is convenient, but it is a page on someone else's domain, weakly connected to your company in the index. Keep a plain careers page on your own site that lists the roles and links out to the ATS, not the other way around.

The site says everything except what you do. "We empower digital transformation" gives the model nothing to repeat. One quotable sentence — "Acme builds inventory software for mid-size retailers, 40 people, remote-first, hiring in Poland" — outperforms a page of vision statements.

Different names in different places. The legal entity is "Acme Software GmbH", LinkedIn says "Acme", the website brands itself "acme.io". A human connects the dots; a model may treat them as three thin entities instead of one solid company. Pick one public name and use it everywhere, with the legal name stated once on the site.

Nothing about Poland or remote work anywhere. If a candidate asks "does Acme hire remotely in Poland?" and no public page has ever said so, the honest model answer is "unclear" — which reads as "no". A single sentence on the careers page settles it.

All news lives on your own blog. The model weighs a fact repeated by unrelated sources far above your own claims. If the only place that ever mentioned your funding round or your product launch is your own domain, you have one source, not several. Even a modest external footprint — a directory listing, a podcast appearance, a local press mention — changes how confidently the model talks about you. The onboarding practices you already document are exactly the kind of material worth publishing where others can cite it.

None of these are expensive to fix. What they share is the failure mode: they starve the model of material it could quote with confidence.

Bottom line

Your employer brand in Poland is whatever an AI chat says it is, because that is what candidates ask first. For a foreign company without local presence, that answer is nearly the whole first impression — and it is built from public sources you control more than you think.

Run the fifteen-minute test. The worst outcome is learning that the model describes you wrongly or not at all. That is also the cheapest problem to fix before your next offer goes out.