There's a new anxiety in cover-letter writing: that an AI-drafted letter will get flagged, the application will be silently dropped, and you'll never know why. The anxiety is partially justified and mostly overstated. AI detection in hiring is messier than both candidates and vendors claim.

This post is what recruiters are actually doing in 2025-2026 when they suspect a cover letter is AI-written, what they catch, and how to use AI as a drafting tool without triggering the flag.

What recruiters actually do

The first myth to dismantle: most recruiters do not run cover letters through AI-detection tools. The SHRM 2024 survey on AI in hiring found that fewer than 15% of recruiting teams use any kind of AI-detection software on candidate writing, and among those, fewer than half trust the output enough to act on it.

The reason is that the tools — GPTZero, Originality.ai, Turnitin AI, and similar — have unreliable accuracy. Multiple academic studies through 2024 documented false-positive rates of 10-30% on legitimate human writing, with non-native English writing flagged at higher rates than native. Recruiters who experimented with these tools quickly noticed they were rejecting strong candidates for false-positive reasons, and pulled back.

What is happening: recruiters detect "AI-feeling" letters the same way they always detected templated letters. They read the first sentence and the last paragraph. If both read like every other cover letter that week, the candidate's letter loses signal — not because it's AI, but because it's generic. AI letters get caught because they're generic by default, not because there's a magic "AI tell" in the prose.

What actually triggers the flag

The failure modes that get a letter categorized as "AI-written" by recruiters:

Same letter, three different companies. When a recruiter has multiple applications from the same candidate across roles at the same company, they sometimes see identical letters with just the company name and role swapped. This is the strongest tell — not because of AI specifically, but because the candidate clearly didn't write a fresh letter. Identical letters across companies were happening long before ChatGPT; they're easier to catch now because AI makes mass-producing them faster.

Buzzword stacking. "Innovative, dynamic, results-driven, passionate professional" — phrases like this scream AI in 2025 because AI defaults to them. They also screamed "lazy human writer" before AI existed. The fix is the same either way: cut adjective stacks; replace with one specific accomplishment.

No specific company reference. A letter that could be sent to any company in the industry has no specific anchor — no mention of a recent product launch, a specific team, a unique aspect of the role. AI defaults to this when prompted with "write a cover letter for this job"; humans default to it when they're tired. The fix is one specific sentence about the company.

Detection-tool flag. Real but rare — a small subset of recruiters use detection tools and act on the output. About 11% of "AI flagging" comes from this path, and it's the path most subject to false positives.

For broader cover-letter mechanics, see cover-letter-opening-lines-that-work and cover-letters-when-they-matter.

How to use AI as a drafting tool

AI is a strong drafting tool for cover letters. The mistake candidates make is treating the draft as the final product. The draft is a starting point that needs human editing — and the edit isn't long.

The pattern that works:

1. Use AI for the structural draft. Feed in the JD and your resume; ask for a four-paragraph cover letter. You'll get something competent and generic — that's the starting point.
2. Replace one paragraph with a specific anecdote. A real story with names, numbers, and dates. This is the paragraph the recruiter will remember and the part AI can't write without your input.
3. Rewrite the opening sentence. AI defaults to "I am writing to express..." or "I was thrilled to come across..." Both are dead. Rewrite the first sentence in your voice. See cover-letter-opening-lines-that-work for patterns.
4. Add one company-specific reference. A recent launch, a blog post, a public talk, a known engineering challenge. One sentence is enough.
5. Vary sentence length. AI tends to produce paragraphs of similar-length sentences. Read your edit out loud and break up the rhythm — one short sentence, one longer one, repeat.

A 10-minute edit of an AI draft produces a stronger letter than 45 minutes of writing from scratch. The recruiter cannot tell which path produced it. The edit isn't optional, though — the unedited AI draft is what gets categorized as generic and skipped.

What about the "rewrite this in my voice" prompt?

Asking an AI to rewrite a letter in your voice doesn't really work unless you've given it real samples of your writing. Even then, the result usually sounds like an AI imitating a voice — close but uncanny. The honest workflow is: AI drafts structure, you write the specific parts. Trying to fully automate the voice produces letters that read as off in ways the recruiter notices without knowing why.

For broader cover-letter structure, see cover-letter-four-paragraph-structure.

What about the "AI-aware" companies?

A small but growing set of companies — particularly in tech, hiring tooling, and some financial-services firms — have explicit policies. Some forbid AI in applications. Others embrace it ("use whatever tools help you do your best work"). Read the application instructions; some companies state their policy explicitly.

When a company forbids AI explicitly, the honest answer is to write your own letter. Submitting an AI-drafted letter against an explicit policy isn't an AI-detection question — it's a trust question, and the trust consequences are real if discovered.

What about the ATS itself running detection?

A common worry: that the ATS quietly runs an AI-detection model and filters candidates before any human sees them. This is, as of 2026, rare. A handful of large enterprise ATS systems offer the feature; most don't use it by default. The candidates who get screened by AI-detection at ATS-level represent a small minority of applications.

For the broader ATS landscape, see how-applicant-tracking-systems-work and ai-resume-screeners-how-to-think.

What this isn't

A few clarifications:

• It's not "AI is fine, write nothing yourself." AI without human editing produces generic letters. Generic letters underperform whether the cause is AI or laziness.
• It's not "AI detection is solved." New tools and techniques emerge regularly. The advice for now is to focus on writing quality, not on outsmarting detectors.
• It's not a recommendation to lie about AI use. If a company asks directly, answer honestly. The deception cost is bigger than the AI cost.

The short version: use AI for the structural draft, then spend 10 minutes editing in specifics — one anecdote, one company reference, a rewritten opening, varied sentence rhythm. Recruiters detect generic letters, not AI letters per se. The fix is specificity. The unedited AI draft is what gets caught; the edited one usually doesn't.