WritingInstant, then a few seconds
Writing voice analyserWhat does your writing voice actually look like?
Paste two emails you have sent. Get your writing fingerprint, measured from the text rather than guessed, and one new email in your voice.
Which reply was written with a voice profile?
Two AI replies answer the same email. One was written against a voice profile, a description of how a specific person writes; the other had no profile and fell back on the default register models reach for. You have twelve seconds to pick the one with no profile, and the reveal names the tells. A free game from Mailient.
- Free
- No account
- Twelve seconds a round
- Three lives
What you paste
Three steps, no sign-up
Read the email
Each round opens on one situation a founder might face. Both replies answer it.
Pick the reply that had no profile
Both were written by AI. One was written against a description of how a specific person writes. The other was told nothing. You have twelve seconds to choose.
See why
The reveal names the tells in the default reply and the profile the other one followed. Right picks earn points and build a streak. A wrong pick or a timeout costs a life.
What you get
How the game scores
Everything here is fixed rules, nothing hidden. This is the whole scoring model.
| Part | Where it comes from | What it tells you |
|---|---|---|
| Points for a right pick | Fixed rules | (100 + 10 for every second left) × (1 + 0.25 for every pick in your streak), rounded. Deciding faster and keeping a streak both pay. |
| Lives | Fixed rules | Three. A wrong pick or running out of time costs one, and the run ends when they are gone. |
| Streak | Fixed rules | Resets to zero on a wrong pick or a timeout, and multiplies your next right one. |
| Best score | Fixed rules | Saved in this browser only. There is no leaderboard and nothing is sent anywhere. |
Reading it
How to read the result
After each pick the reveal names the tells in the default reply and the profile the other one followed. A right pick scores points that grow with the seconds left and with your streak; a wrong pick or a timeout costs one of three lives.
The tell is rarely a stock opener. It is a reply with nothing you could check: no number, no date, no name, no version of events you could disagree with. If you finish reading and cannot repeat one fact back, that is the default one.
To sound less like the default yourself, write shorter, be specific, and cut sentences that manage how the message will feel instead of saying something. If you draft with AI, give it how you actually write and who you are writing to.
Background
Why it works this way
Why everyone's email started sounding the same
Two years ago your writing was yours. It had a rhythm, a set of words you overused, a way you opened when you were in a hurry. Now a large share of what arrives in any inbox has passed through a language model, and models have a house style.
That style is not bad writing. It is competent, grammatical, and characterless, because it is the average of everything ever written. It hedges, it balances, it says nothing in more words than nothing requires.
The result is that fluent, polished email has stopped being a signal of effort. If anything it can read as the opposite: the more perfectly balanced the paragraphs, the more likely nobody thought about you while writing them.
What the scarce thing is now
Writing that sounds like a specific person. Not better writing, not more correct writing. Recognisable writing.
That is the whole reason a voice profile exists. Ask a model to write an email and you get the average. Tell it that this person writes eleven-word sentences, never opens with a greeting, is blunt about money and never apologises for their prices, and you get something a reader will not flinch at.
The email you were not hunting in each round was written that way. The one you were hunting is what you get from every AI email tool that does not know who you are.
Limits
What it cannot tell you
Every tool here is a first pass with edges. These are this one's.
- It is not an AI detector. It tests one thing: whether you can tell a default-register reply from one written with a voice profile. Doing well here is not evidence you can spot AI text in general.
- Nothing here was written by a human. Every reply was generated. A version with one genuinely human option would be a better hook, but we could not build it honestly.
- Length is not a tell. The two replies in a round are matched on length before you see them, and a generated round is thrown away if it fails that check.
- A run can end early on material. If the supply of rounds runs out before your lives do, the run ends cleanly instead of repeating one.
Beyond the sample
When this is not enough
A game trains your eye on replies we generated. It does not change what an AI tool writes for you.
Mailient builds a writing-voice profile from a sample of your recent sent mail (up to 40 emails) and uses it when it drafts a reply you ask for on Gmail. It also shows what needs a decision today, and by default sends nothing until you approve it.
Your data
What we keep
Read off the code that does it, not written from memory.
- In your browser
- The whole game. Your best score is saved in this browser and nowhere else.
- Sent to our server
- Only a request for a fresh round, carrying the topics of the last few rounds. Nothing about you.
- Sent on to a model
- Yes, to write new rounds while you play (Gemma through OpenRouter). You type nothing, so it sees nothing of yours.
- What we keep
- Nothing about you. A request counter tied to your IP address (30 rounds an hour) holds no content.
Read from components/SpotTheAI.tsx, app/api/spot-round/route.ts and lib/rate-limit.ts.
Questions
Questions people ask, answered plainly
6 answers, grouped. Each one is written against what the tool actually does.
Playing
How does scoring work?
A right pick scores (100 + 10 for every second left) times (1 + 0.25 for every pick already in your streak), rounded. So a fast, correct pick with a streak of three scores far more than a slow one at the start. A wrong pick or a timeout costs one of your three lives and resets the streak.
About the emails
Were any of these written by a real person?
No, and that matters enough to say plainly. Every email in this game was generated. A version where one option is genuinely human would be a better hook, but we cannot build it honestly, and a site that claims a human wrote something a machine wrote has no business lecturing anyone about AI slop.
So what is the actual difference between the options?
Whether the model was given a voice profile. One of the two was written against a description of how one specific person writes: their sentence length, whether they greet by name, how blunt they are about money, what they never say. The other was written with no profile at all, in the register a model reaches for by default. Same brief, same model, one difference.
What are the tells?
Not the obvious ones. An email that opens by hoping this finds you well is easy, and a game built out of those would teach nothing. The replies here are the harder version: fluent, warm, plausibly human, and completely uncheckable. Nothing in the default one can be verified or argued with: no number, no date, no name, no version of events you could disagree with. If you finish reading and cannot repeat one fact back, that is the one.
Your data
Does the game send anything about me?
Only a request for a fresh round, which carries the topics of the last few rounds so you are not given repeats. You type nothing, so there is nothing of yours to send. We keep a request counter tied to your IP address, limited to 30 rounds an hour, which holds no content.
Mailient
Is this what Mailient does?
It tests the difference a voice profile makes. Mailient builds a writing-voice profile from a sample of your recent sent mail (up to 40 emails) and uses it when it drafts a reply you ask for on your Gmail. Every reply in this game was generated for the game; none was written by Mailient for a real person.
More free tools
Try another one
SecurityAbout two seconds
Email spoofing checkerCan someone send email as this domain?
Type your domain. See whether a stranger can email your customers from your address, what that email would look like, and the exact records that stop it.
InboxInstant, then a second look
Inbox triage testWhich emails in this list actually need you?
Paste the subject lines in your inbox. Get them sorted into what needs you, what can wait and what a machine sent, with the reason for each.
This is the sample. Mailient does this on Gmail.
This game runs on replies we generate. Mailient does the same kind of reading on your Gmail: it shows what needs a decision today, drafts in your voice when you ask, and by default sends nothing until you approve it.