Usually yes, and not because they ran it through a detector. They recognise the register — the same one that arrives in eleven other messages that week. What gets noticed is not that a machine helped. It is that the email says nothing only you could say.
Read this if: you are drafting with AI and want to know what to change before sending.
Read instead: the line-by-line structure, 12 finished examples, and every situation in one table.
What actually gives it away
None of these are exotic. They are the default output of every general-purpose model asked to write a professional email, which is why they show up together.
Fluency with no facts
The sentences are well formed and the paragraph contains no information. "I bring a strong background in cross-functional collaboration and a passion for delivering impactful results" is grammatical and empty. A manager skimming for a reason to reply finds nothing to hold.
Even rhythm
Human email is lumpy. A long sentence, then three words. Generated text keeps a steady clause length from the greeting to the sign-off, and after a hundred of them the flatness becomes recognisable on sight.
The three-item habit
"Scalable, reliable and maintainable." "Strategy, execution and measurement." Groups of three, over and over. Real writing lands on two things, or four, or one.
Vocabulary nobody says out loud
Leverage as a verb. Delve. Robust. Seamless. Landscape. Underscore. Testament. These words are not wrong, they are just far more common in generated text than in anything a person types into a compose window at 9pm.
Praise with nothing behind it
"I have long admired your company's commitment to innovation." A model that has not read their changelog produces admiration in the abstract, and abstract admiration reads as a mail merge.
The mirrored construction
"It is not just a role, it is an opportunity." "Not only did I lead the team, I also..." This shape appears constantly in generated prose and rarely in a hurried email.
What detectors actually do
Almost nobody runs a hiring email through an AI detector. Detection tools are aimed at essays and coursework, they need several hundred words to say anything, and a hundred-word email is below the length where their output means much.
They also produce false positives on plain, well-structured writing by non-native speakers, which is one reason few hiring teams treat a score as evidence of anything. The practical risk is not a detector flagging you. It is a manager recognising the shape and putting the email down.
What managers report noticing
The complaint is rarely "this was written by AI". It is that the volume of applications went up while the amount of information in each one went down. Recruiters process roughly 291 applications per hire, and a generated email that adds another well-written paragraph containing no checkable fact makes the pile worse rather than standing out in it.
Which is the useful frame. The question is not whether a model touched your draft. It is whether the finished email contains something true, specific, and yours.
The edit pass
Five moves, in this order. They take about four minutes and they change the email more than any prompt does.
- Replace the claim with a fact. Find the sentence about your experience and swap it for one result someone could verify. A number, a system you built, a scale you handled. This one move fixes most of the problem.
- Add the detail only you know. Something from the posting, their product, or a decision they made. If you cannot find one in five minutes, delete the compliment rather than keeping a generic one.
- Break a sentence. Cut the longest one in half. Let a short fragment sit on its own. The rhythm stops being flat.
- Say one true limitation. You are junior, you have not worked in this sector, you are relocating in March. Models do not volunteer weakness. Managers reply to it.
- Read it out loud. Any sentence you would not say to a person in a corridor gets rewritten in words you would.
Where a generated draft genuinely helps
The blank page is the real obstacle. Most people do not fail to send this email because they cannot write — they fail because writing it means deciding what to claim about themselves, and that is unpleasant enough to postpone for a week.
A draft built from the actual job posting and your actual profile puts the five parts in the right order, names the right role, and gets the length roughly right. What you are left with is an editing job, and editing is much easier than starting.
It also helps with the parts that are pure convention: the subject line naming the role, the closing that asks for one small thing, the decision to leave the resume unattached on a first message.
What you paste in decides most of the quality
The gap between a usable draft and an unusable one is almost entirely input. Two things change the output more than any prompt wording.
| Given the full posting and your history | Given a job title and a prompt | |
|---|---|---|
| Recipient | Named, so the greeting and the register are settled | [Hiring Manager], which you then have to fix |
| Evidence line | Drawn from something you actually did | Invented, or a bracket for you to fill |
| Relevance line | Points at a problem the posting names | Generic praise for the company |
| Length | Constrained on purpose | Three paragraphs, usually |
| Work left for you | An edit | A rewrite |
Which is the practical argument for pasting the whole posting rather than the job title, and for giving it your real history rather than asking it to write about a “senior marketer with 7 years of experience”. A model writing about a generic person produces generic sentences, and no amount of instruction fixes that — it has nothing else to say.
What it should never do
- Invent a metric. If a number appears in your draft that you did not supply, delete it. Being caught inflating a result at interview costs more than never getting the interview.
- Claim familiarity you do not have. "I have followed your work for years" is checkable in a way that goes badly.
- Guess the manager's name. Verify it. A confidently wrong name ends the conversation in the first word.
- Write the whole thing unread. The email arrives with your name on it and you will be asked about every sentence in it.
A note on where our own drafts come from
Since this site sells a tool that does this, the disclosure is worth making plainly: our drafts are built from the job posting and your LinkedIn profile rather than from a prompt, for the reason in the table above, and the greeting is composed separately per contact so the name is not guessed. The output is held short and plain because that is the register that survives the edit pass with the fewest changes.
None of which removes the five moves. A draft is a draft whoever generated it, and the sentence that earns the reply is still the one only you can write.
Works with LinkedIn, Indeed, Greenhouse, Lever, and company career pages.
FAQ
Is it wrong to use AI to write an email to a hiring manager?
No, and the line most managers draw is about accuracy rather than authorship. Using a model to structure a note about things you genuinely did is ordinary. Using it to manufacture experience is misrepresentation, and it surfaces at the first interview question.
Can a hiring manager detect AI writing in a short email?
Not reliably by tool — a hundred words is below the length where detectors give a meaningful score. By eye, often yes, because the vocabulary and the even rhythm are recognisable after enough exposure.
Do recruiters run AI detectors on applications?
Some screen resumes and written assignments. Emails, rarely — there is no workflow for it and the false-positive rate on short professional text is high enough to make the result useless.
How do I make an AI-written email sound like me?
Replace the general claim with one checkable fact, add a detail specific to that company, break the sentence rhythm, and admit one real limitation. Then read it aloud and rewrite anything you would not say.
Should I tell them I used AI?
No. Nobody discloses their spellchecker either. What matters is that every claim in the email is true and that you can talk about all of it.
Does a generated email get fewer replies?
A generic one does, and generated emails are generic by default. An edited draft carrying one specific, checkable result performs like anything else written by hand, because the specificity is what the manager is responding to. The line-by-line guide covers what that looks like in practice.