Can recruiters tell a cover letter was written by AI? 6 tells

Four polls that we could trace to their source asked people who hire whether AI use in an application can be recognised. One question named cover letters and one asked about the job application as a whole. The other two asked about resumes or CVs only. Samples and years for every figure are in Cover letter statistics.

Read the verbs: "can tell", "can often tell", "possible to tell", "easy to detect". Each one reports confidence. What is being told differs too: AI used "to help with" an application in one question, a resume "written by AI" in another. An answer to a pollster is a belief, not a detection rate: nobody gave these respondents letters of known origin to sort. A person who says they can tell is also counting only the letters they noticed. The ones that passed are missing from the count.

Most polls on this subject, these included, come from companies with something to sell to job seekers or employers. So does this guide.

A measurement shows people texts whose origin is known and counts how often they sort them correctly. We found no such test of recruiters reading cover letters. The nearest we have is a peer-reviewed set of experiments on a neighbouring kind of writing.

The participants were members of the public, not recruiters, and they judged profile-style texts, not cover letters. Someone who reads applications every working day might do better; that is the test we could not find. Confidence has been polled several times. We found no measurement of the skill itself.

Software that labels text as AI-written has been tested in public. We have three sources, all from 2023.

The same classifier labelled human-written text as AI-written 9% of the time. OpenAI withdrew it in July 2023 "due to its low rate of accuracy".

In that peer-reviewed paper, 89 of the 91 essays by non-native speakers were flagged by at least one detector. Separately, a test of 14 detection tools, 12 public and 2 commercial, concluded that they are "neither accurate nor reliable" and that they lean towards classifying text as human-written.

If you are worried about being wrongly flagged

These tests found errors in both directions: AI text labelled as human, and human text labelled as AI. They are evidence about those tools in 2023; we could trace nothing newer. We have no figure on whether employers run letters through such software, so we make no claim about it.

• Treat a detector's score as a guess. Do not rewrite your own letter until a tool approves of it, and do not take a clean score as protection.
• If English is not your first language, the result on non-native writers is the one to know: the wrong flags fell on people who had written their own essays.
• Keep your working: the listing, your notes, your earlier drafts. If you are ever asked how a letter was written, answer honestly and show them.

This list is our own judgement of what makes a letter read as generated. It describes the text, not what recruiters or software notice, and a tired person writing alone can produce all six. The examples are invented: Ayşe Demir, a procurement officer at Havelby Foods, is applying to be Procurement Manager at Orrinmoor Leisure.

1. It could be sent for any job

2. A stock opening

3. Adjectives with no evidence

4. A company paragraph that names nothing

5. Every sentence the same shape

6. Facts that are not in your resume

Being recognised matters only if something follows from it. On that the polls disagree, and the wording of each question explains much of the gap.

Some rows count plainly against a tool like ours. The UK result is the harshest, and in the 2026 Resume Genius poll only job-hopping, at 65%, was named as a red flag more often.

Now read the wording. "AI-generated" suggests a document handed over as it came out. "Heavily" builds a judgement into the question. "Drafting support" and "for cover letter writing" describe assistance, and those two questions drew majority approval. The answer moves with the words even among the same respondents: compare rows one and two, then three and four. None of these numbers counts rejections. Each is what someone told a pollster they would do.

Being undetectable is the wrong aim, and nobody can promise it. Aim for a letter that is true, written for one job, and yours to defend.

1. Follow the instruction if there is one. If a listing or an employer says not to use AI, do not use it. If a form asks whether you did, answer truthfully.
2. Draft from your own material. Give the tool your resume and the real listing, not a job title. ChatGPT cover letter prompts shows how with a general chatbot.
3. Check every fact against your resume, then go through the six tells above.
4. Add what no draft can know: why this employer, and the detail behind your best result. The ten-minute pass takes one letter through the whole edit.

One working paper bears on the editing step. After a large online labour platform gave applicants an AI cover letter tool, the link between how closely a letter's text matched the job post and whether the applicant got a callback fell by 51%. Our reading, which does not flatter a company like ours: once anyone can produce a well-matched letter, a well-matched letter says less. The same abstract reports that access to the tool raised callback rates, and that time spent editing the AI draft was positively correlated with hiring success. The paper covers freelance platform jobs and is not yet peer reviewed, and a correlation is not proof that editing caused the success.