What happens when a recruiter asks ChatGPT about you
AI builds an answer about you from a few public pages or from thin training memory, and that answer is often the first impression a recruiter gets.
When a recruiter asks ChatGPT about you, the AI either searches a handful of public pages and summarises them, or falls back on thin training memory. It checks nothing against a private record, so the answer can be wrong, out of date or generic. Asking the same question yourself, with and without web search, shows what a recruiter sees.
A recruiter has a name on a shortlist and ten minutes. One of the quickest things they can do is ask an AI: who is this person, and what is their background? The answer arrives in seconds, in confident, tidy paragraphs. It reads like a fact sheet. It may not be one.
This post explains how AIs put together an answer about a named person, where it tends to go wrong, and what you can and cannot do about it.
How an AI builds an answer about you
There are two quite different situations, and the AI you are asking determines which one you are in.
With web search switched on, the AI behaves like a very fast researcher. It searches for your name, opens a handful of public pages, and summarises what they say. The usual suspects are your employer's leadership or team page, your LinkedIn profile, interviews and articles in the press, conference speaker bios, and official company registers.
Without web search, the AI relies on what it absorbed during training. For a well-known public figure there may be plenty. For most senior professionals the material is thin, out of date, or mixed with information about other people. The AI still produces a fluent answer, which is exactly why a thin source can look like a solid one.
In both cases, the AI does not verify anything against a private record. It reads, weighs and condenses. Whatever is easy to find and consistent across sources tends to win.
Four failure modes we see in testing
In our early testing, the same patterns come up again and again.
1. Mixing you up with a namesake
If your name is shared with someone in a different field or country, AIs sometimes blend the two. You may be credited with a book, a board seat or a career you never had. Common names and names with several spellings are the most exposed.
2. Stretching dates or merging roles
Two consecutive roles can be turned into one long one, or a title can be attached to the wrong employer. A four-year role can quietly become six. These errors are small in the AI's eyes and large in a recruiter's.
3. Outdated titles
If the most visible page about you was written three jobs ago, that is the version you may be presented as. AIs lean on whatever is most prominent, not whatever is most recent.
4. Leaving out what you want to be known for
This is the quietest failure and often the most costly. The answer is accurate but generic. The turnaround you led, the market you opened, the specialism you spent years on: none of it appears, because no public page says it clearly.
What AIs cannot know
It helps to be clear about limits, because they spare you some worry. AIs cannot see internal numbers: revenue against plan, headcount, budgets, reporting lines, or how a project really went. If a figure is not public, it is not in the answer, and an AI that offers one is guessing. There is little point in worrying about what is not published. The useful question is what is published, and whether it tells your story correctly.
Why the same question gets different answers
Ask the same AI twice and you may get two slightly different summaries. Ask two different AIs and the gap can be wider. Each one searches differently, trusts different sources, and phrases things in its own way. A recruiter may use any of them, so the useful picture is the range of answers, not one screenshot of the best or worst.
It also depends on the question. "Tell me about this person" produces a general biography. "Would this person suit a senior operations role?" produces an assessment, built from whatever evidence the AI can find for that role. We come back to this in the third post of this series.
Why this matters more than it seems
Nobody announces that they used an AI to prepare for a call or an interview. They simply arrive with a first impression already formed. If the summary is accurate and reflects your strengths, you start ahead. If it is wrong, vague or dated, you may spend the first minutes of a conversation correcting the record, or never find out why the conversation went cold.
This is not about gaming a system. It is about making sure the public record is accurate, current and findable, which is also what a careful human researcher would want.
What a good answer looks like
A good answer is not flattering. It is accurate, current and specific. It gives your present title and employer, a sensible timeline, and one or two things you are actually known for. It does not invent anything, and it does not confuse you with someone else. If an AI gets close to that, you are in good shape. If it does not, the cause is usually fixable, because it traces back to public pages that can be updated.
What you can do this week
- Ask the question yourself. Open two or three AIs, and ask what they know about you, as a recruiter would: name, current employer, and the kind of role. Note what is wrong, missing or out of date.
- Try with and without search. The two modes give different answers, and you want to know both.
- Check the sources it cites. When an error appears, it usually traces back to one page. Fix or update that page first.
- Repeat later. Answers change over time as AIs and sources change. A single check is a snapshot, not a verdict.
The next post in this series looks at the five kinds of page that carry the most weight, and what to do with each.