AI Detection in Hiring: Is It Fair to Screen Cover Letters for AI?

A growing number of recruiters now paste cover letters into an AI detector before they ever read them. The pitch sounds reasonable: filter out lazy, machine-written applications and keep the people who actually tried. But look at how these tools work — and at who they tend to flag — and the fairness of the whole idea falls apart fast.

This is worth taking seriously, because the stakes for a job seeker are higher than a grade on one essay. A bad detector call can quietly cost someone an interview they earned.

Key takeaways

• AI detectors output a probability, not proof — a high score on a cover letter is a guess, not evidence of cheating.
• Detectors misfire on short text and non-native English writers, so hiring screens can filter out exactly the candidates many employers want.
• Most companies have no stated policy banning AI tools, so penalizing for AI use enforces a rule that was never communicated.
• A fairer screen evaluates the substance of an application — relevance, specifics, fit — not whether a tool helped draft it.
• For candidates, the safe move is to use AI as a drafting aid and then make the writing genuinely yours.

Why employers started screening for AI

The motive is easy to understand. Since late 2022, application volume has exploded. A single LinkedIn posting can pull hundreds of submissions in a day, and a noticeable share read like they came off the same template — because they did. Recruiters are tired, and an AI detector promises a quick way to thin the pile.

There’s also a values argument. Hiring managers say a cover letter is supposed to show genuine interest and communication ability. If a candidate just typed “write me a cover letter for this job” into a chatbot, the thinking goes, the letter tells you nothing real about them.

That instinct isn’t wrong. The problem is the tool they’re reaching for to act on it.

What an AI detector actually measures (and doesn’t)

AI detectors don’t read for sincerity. They run statistics. Most of them score text on how predictable it is — the detectors measure perplexity and burstiness, roughly “how surprising is each word” and “how much does sentence length vary.” AI models tend to produce smooth, low-surprise, evenly paced text, so detectors learn to flag that pattern.

Notice what’s missing. The detector doesn’t know who wrote the letter, whether it was edited, or whether a human just happens to write in clean, even prose. It produces a number — say, 87% likely AI — and that number is a statistical estimate, not a verdict.

Two facts make this worse for hiring specifically:

• Cover letters are short. Detectors need a decent amount of text to be confident. A 250-word letter gives them little to work with, and confidence drops fast on short samples.
• Cover letters are formulaic by design. “I’m excited to apply for the role of…” “I believe my background in… makes me a strong fit.” Those phrases are common because they work — and that commonness reads as low-perplexity, AI-like text even when a human wrote every word.

So the very genre you’re screening is one detectors are bad at judging. That’s a structural mismatch, not a tuning problem you can fix with a better tool.

The fairness problem nobody mentions in the job posting

Here’s the part that turns an awkward tool into an ethics issue.

In 2023, a Stanford research group (Liang et al.) tested several commercial detectors on essays written by non-native English speakers. The detectors flagged those essays as AI-generated far more often than essays by native speakers — even though humans wrote all of them. The reason tracks with the statistics above: writers working in a second language often lean on safer, more predictable phrasing, which is exactly what detectors penalize.

Now connect that to hiring. If a company screens cover letters with a detector, it’s likely flagging a higher share of applications from non-native English speakers, immigrants, and international candidates. Many of those same companies say they want a diverse, global workforce. A detector-based screen can quietly work against that goal, and nobody in the loop necessarily realizes it’s happening.

There’s a second fairness gap. Most job postings say nothing about AI. There’s no rule against using ChatGPT to draft a cover letter — and yet a candidate can be silently rejected for it. Penalizing people for breaking a rule you never told them about is hard to defend, whether you’re a school or an employer. (Schools are wrestling with the same thing; see how the debate over AI detection and writing plays out across different settings.)

A mini-scenario

Picture two applicants for a marketing coordinator role.

Maya is a native English speaker. She wrote her cover letter herself in fifteen minutes, used a couple of standard openers, and kept it tidy. The detector scores it 72% AI. She gets cut.

Daniel used ChatGPT to produce a first draft, then rewrote it heavily — added a specific story about a campaign he ran, changed the structure, cut the filler. The detector scores it 30%. He moves forward.

Maya wrote every word and lost. Daniel leaned on AI and won. The screen didn’t measure honesty or effort. It measured how closely the prose matched a statistical pattern. That’s the core problem with using detection as a gate: it isn’t aligned with the thing the employer actually cares about.

Is using AI on a cover letter even cheating?

For most jobs, no — and treating it that way runs two different things together.

Using AI to brainstorm, outline, fix grammar, or tighten phrasing is closer to using spell-check or asking a friend to proofread than to faking a credential. What an employer should care about is whether the application is accurate, relevant, and genuinely reflective of the candidate. A letter that lies about experience is a problem no matter who or what typed it. A letter that’s honest and specific is fine no matter what helped draft it.

The real failure mode isn’t “AI was involved.” It’s “this submission is generic and tells me nothing.” Those often overlap — unedited AI output tends to be vague and interchangeable — but they’re not the same thing, and a detector can’t tell them apart.

If your company genuinely wants to set expectations, the answer is a policy, not a detector. Tell candidates what’s allowed. State whether AI assistance is fine, discouraged, or off-limits for a writing sample. That’s transparent and enforceable in a way a probability score never will be.

A fairer way to screen

If the goal is to find people who fit the role, evaluate the things that actually signal fit:

• Specificity. Does the letter reference the actual company, team, or problem — or could it be sent to any employer? Specifics are hard to fake and easy to read for.
• Relevant substance. Does the candidate connect their experience to what the job needs? That’s a judgment a human reviewer makes well and a detector can’t make at all.
• Work samples and interviews. If you want to know whether someone can think and communicate, ask them to. A short prompt, a portfolio, or a 20-minute conversation tells you more than any score.

Used this way, AI detection becomes at most a soft, internal signal — never a silent reject button. The gap between “one input a human reviews” and “an automated gate” is the gap between a tool and a liability.

Frequently asked questions

Can recruiters really tell if my cover letter was written by AI?

They can run it through a detector and get a probability score, but they can’t tell for certain. Detectors flag statistical patterns common in AI text — they don’t see your draft history. A high score is a guess that can be wrong, especially on short documents like cover letters.

Is using ChatGPT to write a cover letter considered cheating?

That depends entirely on the employer, and most don’t have a stated policy. Using AI to brainstorm, outline, or polish is widely treated like using a grammar checker. The risk is submitting something generic or inaccurate — not the tool itself. If a posting explicitly bans AI assistance, follow that rule.

Why would an AI detector flag a cover letter I wrote myself?

Detectors look for predictable phrasing, and cover letters are naturally formulaic. Short text gives the detector little to work with, so confidence drops. A 2023 Stanford study also found detectors disproportionately flag non-native English writers, whose phrasing can read as more predictable to these tools.

Should companies use AI detection to screen applicants at all?

It’s hard to defend as a sole filter. Because output is probabilistic and prone to false positives, using a score to reject candidates risks discarding qualified people and may disadvantage non-native speakers. Judging content — specificity, relevance, fit — plus interviews and work samples is far more defensible.

The bottom line

AI detection in job applications answers the wrong question. It tells you how a text patterns statistically, not whether a candidate is honest, capable, or right for the role — and it stumbles hardest on short, formulaic writing by non-native speakers, which describes a lot of perfectly good applicants. If you hire, screen for substance. If you apply, use AI as a drafting aid, then make the writing unmistakably yours. (Curious what a tool like this costs? Here’s what PaperBleach costs.)

Want to see how your draft reads to a detector before you hit submit? You can check and humanize a draft and watch which lines look machine-made — then fix them in your own voice.