Written by AI
When the words give it away.
Nextmail reads the sender’s own words as mail arrives, and marks the message when they look generated rather than written.
In short
What makes an email look AI-written?
Language models have habits, like stock greetings and sentences of near-identical length. People leave typos and shorthand. A detector weighs one against the other, but no detector can prove who wrote a message.
60
out of 100 earns the label
20
tells count towards the score
7
count against it, like typos and emoji
How it works
Read from the words they wrote
Quoted history and the signature are cut first. Stock phrasing adds to the score, and a typo or a quick btw takes points off.
A second reading that leans human
From 30, the assistant reads the words too, unless the mail came from a template. It can take off 45 points but add only 35.
Declined rather than guessed at
Anything under 40 words, or encrypted end to end, is not scored, and Details says which. Bulk and template mail are scored but never get the label.
Tells it looks for
Weighted and summed from the sender’s own words.
What you get
In the product today
Labels, never filters
Nothing is moved or hidden on the strength of a guess.
Maybes stay in Details
Between 35 and 60, the verdict stays in the Details tab.
Encrypted mail is skipped
An encrypted message is declined, not scored, and Details says so.
Good practice
Getting the most out of it
- 01
Open the Details tab
It lists the tells that fired and the score behind the label.
- 02
Judge what it says
A label describes the writing, not whether the message matters.
- 03
Read the score plainly
It measures how much stood out, not the odds of being right.
Questions
Asked often
Keep going
Works well with
Start
Your domain,
your mail.
Point a domain at Nextmail and read it in a mailbox built around it. The free plan covers one domain.