A note before the tells. Models change, but the giveaways below come from how these tools write by default rather than from any particular version. The list ages slowly and the fix underneath it does not age at all.
Somewhere in your organisation, and quite possibly in your own outbound history, an email is being deleted unread because the recipient could tell no human wrote it. People have started saying this out loud, that they can spot the AI-drafted message from the first line and have stopped reading them. The uncomfortable part is that the senders have no idea.
The email went out, no reply came back, and the silence got filed under "busy" rather than "detected". This is worth taking seriously, and not because using AI for email is wrong. Most professionals now do.
The issue is that being caught doing it lazily carries a real social cost, because the recipient is not offended by the tool. They are offended by receiving something no human attended to, and attention is precisely what they are detecting.
The short answer: AI-drafted emails get detected through a handful of consistent tells, the padded warmth, the restating of the recipient's own message, the perfectly balanced structure, the enthusiasm of the sign-off. They get fixed with a two-minute human pass that cuts the padding, adds the one specific detail only you would know, and keeps your own phrasing where it matters. AI supplies the shape. You supply the evidence that a person was here.
What actually gives a draft away
| The tell | Why it reads as AI | The fix |
|---|---|---|
| The warm-up padding, "I hope this finds you well" | Default politeness inserted for nobody in particular | Delete it, or replace with something true, like "good to see you Thursday" |
| Restating their email back at them, "Thank you for reaching out regarding..." | The model summarising its input, which no human does to a colleague | Answer the question. The reply itself proves you read it |
| Perfectly balanced everything: three points of equal weight, uniform paragraphs | Trained symmetry, where real emails are lopsided because real priorities are | Let the thing that matters take the space, and shrink the rest |
| Formal connectives mid-email: "furthermore", "I wanted to touch base" | Register borrowed from a business-writing textbook nobody owns | Swap for how you would say it across a desk |
| The enthusiastic sign-off about being thrilled | Emotional temperature the situation never asked for | Match the sign-off to the stakes, which for most emails is mild |
The two-minute pass that fixes it
Since the recipient is detecting an absence of attention, the fix is a visible unit of attention, and it takes about two minutes.
- Cut the padding. Read the draft once and remove every sentence that could be sent to anyone, because those are the filler.
- Add one specific detail. The reference to what they said on the call, the constraint that actually matters, the honest reason for the chase. A single genuine specific does more to humanise a message than any amount of tone adjustment.
- Move the ask forward. Check it sits where you would put it, which is usually earlier and blunter than the draft has it.
- Keep your own phrasing. Even where the tool's version is technically smoother, hold onto anything that sounds like you, because smoothness is now part of the tell.
It also helps to brief better upstream, since a draft built from real context needs less rescuing, and there is a fuller piece on that handover. Claude tends to hold your phrasing better than ChatGPT when you give it something to hold, which matters for email more than almost any other format. Email is the one place your voice is the product.
The deeper rule
Use the tool for what it is good at here: the shape, the first pass, the unsticking of the reply you have been avoiding. Reserve for yourself the part that was always the point of an email, which is the evidence that a specific person thought about another specific person for a moment.
That evidence takes two minutes to add, and its absence is now legible to everyone. The people deleting AI emails unread are not rejecting the technology. They are enforcing an old standard by a new method, and it is a standard worth passing.
Not sure where to start?
The right tool drafts. You make it yours.
The free finder tells you which AI to open for the task in front of you, in about 60 seconds, and the full "AI, sorted." reference lands in your inbox.
Try the free finder →One question in, one answer out.
Clair helps non-technical professionals know when to trust their AI, when to check it, and when to skip it.