Everyone Is Using AI to Apply. What Should Hiring Managers Actually Do?

Open a req today and you will get more applications than you did three years ago, and they will read better. Cleaner sentences, tighter paragraphs, fewer typos, and a consistent, oddly familiar cadence across candidates who have nothing else in common. Generative AI got very good at the cover letter, which was always a formulaic document to begin with.

That leaves hiring managers with a genuinely awkward question. If a candidate used ChatGPT to write their application, does that tell you anything useful about whether they can do the job?

Sometimes the answer is no

For most roles, it does not. A cover letter is a low-signal artifact even when a human writes it. It is a genre with a known structure, a known register, and a known set of moves, which is exactly why models produce passable ones. Screening out everyone who used a tool to clear that bar mostly screens for who felt comfortable not using one.

There are exceptions worth naming. If you are hiring a writer, an editor, a content marketer, or anyone whose output is prose, then the writing sample is the work sample, and how it was produced is directly relevant. If you have explicitly asked candidates not to use AI, the answer changes again, because now you are measuring whether they follow instructions.

The point is to decide what you actually care about before you start scanning, rather than letting a score decide for you after the fact.

What a detector can honestly tell you

A detector measures patterns in text. The strongest signals are flattened sentence rhythm, stock phrasing, and predictable structure. That is genuinely useful information, and a sentence-level report will show you which specific passages carry it, which is far more actionable than one number for the whole document.

What it cannot do is establish authorship. No detector, including the one behind this site, returns a verdict. It returns a likelihood, and the honest way to read a high score is "this reads like generated text," not "this person deceived me."

Where false positives will hit you hardest

Detection misfires in patterned ways, and hiring is unusually exposed to the worst of them.

• Non-native English writers. This is the sharpest risk. Someone who learned English through formal instruction tends toward simpler grammar and more common vocabulary, which is precisely the pattern detectors read as machine-generated. Stanford researchers found that more than half of a set of real TOEFL essays written by non-native speakers were wrongly flagged by popular detectors.
• Short documents. A three-paragraph cover letter is close to the minimum length worth checking at all. Detection on short text is noisy, and confidence should drop accordingly.
• Heavily-templated writing. Candidates coached by a career center or working from a university template are following a formula on purpose. So is the model.

Read those three together and a pattern emerges: the applicants most likely to be wrongly flagged are disproportionately the ones with the least room to absorb an unfair rejection. That is a legal exposure as much as an ethical one, and it is worth running past your own legal and HR review before a score touches a hiring decision.

A more useful way to use the signal

The productive move is to treat a flagged application as a prompt for a better question rather than as grounds for a decision.

• Ask about the work, not the writing. If a cover letter claims a result, ask the candidate to walk you through how they got it. Someone who did the work can go three levels deep. A generated paragraph cannot.
• Prefer a longer writing sample. If prose matters for the role, ask for a real piece of work, with more text for a detector to actually reason about.
• Use the passages, not the percentage. A report that marks specific sentences gives you something concrete. If all you have is an unexplained number, weight it less, not more.
• Never make it the sole basis for a rejection. There is no version of this tooling that supports that, and vendors who imply otherwise are overselling.

The honest bottom line

AI-assisted applications are now the baseline, and no detector is going to put that back in the box. What a detector is good for is telling you where a document stopped sounding like a person, so you know which part of the conversation to open with. That is a real and useful thing. It is just considerably smaller than "catching cheaters," which is how this category tends to get sold.

If you want to see what that looks like on a real application, the AI detector for hiring managers page runs a free check with no account, and the report marks the passages rather than just handing you a number. It is also worth reading why detectors flag genuinely human writing before you rely on one.