A colleague sends you two lines. You ask an assistant to reply. You get back five sentences of correct English that nobody would type into a chat window. The facts are right. The register is wrong.
This is a different problem from AI slop. Slop inflates importance with stock phrases. Polish is the opposite failure. The words are plain and the sentences are clean, and the draft still reads as a letter.
The tells
Read a stiff draft and you find the same eight habits:
- No contractions. The draft writes "do not", "I am" and "it is" where you would type "don't", "I'm" and "it's".
- Every comma correct. A human reply drops commas. A model places all of them.
- Semicolons. Almost nobody types a semicolon in a reply.
- Uniform sentence length. Four sentences of 18 words each. Your own writing mixes 4 words and 24.
- A formal opener. "I hope this email finds you well." "Thank you for reaching out."
- A formal closer. "Please do not hesitate to reach out." "Best regards."
- Written connectors. "However," "Additionally," "Furthermore," "Therefore,".
- Softened asks. "I would like to confirm", "at your earliest convenience", "Kindly".
No single habit is wrong. Density is the problem. Six of them in one paragraph make the reply feel produced rather than typed.
Why it happens
The model answers in the average register of its training data for "professional email". That average is a formal letter. A letter contracts nothing and punctuates everything, so the model does the same. It also never drops punctuation on purpose, because dropping a correct comma looks like an error. Your two-line question gets a letter because letters are what the label describes.
Three before and after examples
Each "After" below comes from the rule set alone. The rules only delete letter furniture, contract verbs, split semicolons and swap a formal word for a plain one. No sentence was rewritten and no fact changed.
Before · 74 words · stiffness 49 I hope this email finds you well. Thank you for reaching out. I would like to confirm that moving our review to Thursday does not present any difficulty on my end; I have reviewed my calendar and the afternoon slot remains available. Additionally, I will update the invitation and notify the wider group in advance. Please do not hesitate to reach out should you require any further information at your earliest convenience. Best regards
After · 47 words · stiffness 6 I want to confirm that moving our review to Thursday doesn't present any difficulty on my end. I've reviewed my calendar and the afternoon slot remains available. Also, I'll update the invitation and notify the wider group in advance. Just let me know when you can. Thanks,
Eleven rules fired: two deleted openers, one semicolon split, three contractions, one connector, three softeners, one sign-off.Before · 72 words · stiffness 48 Thank you for reaching out. I am writing to advise you that scheduled exports do not currently exist within the product. However, our engineering team is actively working on the capability, and I would like to confirm that I will notify you as soon as it is released. In the meantime, you are able to export manually. Should you have any further questions, please do not hesitate to contact me. Best regards
After · 51 words · stiffness 9 I'm writing to tell you that scheduled exports don't currently exist within the product. But our engineering team is actively working on the capability, and I want to confirm that I'll notify you as soon as it's released. In the meantime, you're able to export manually. Any questions, just reply. Thanks,
The answer arrives in the first sentence instead of the third.Before · 59 words · stiffness 51 I hope you are doing well. I wanted to reach out to confirm the status of this morning's deployment. I am pleased to advise you that the release did not encounter any issues; it is now live in production. Additionally, I have verified the health checks. Please do not hesitate to reach out if you require any further detail.
After · 44 words · stiffness 10 I'm reaching out to confirm the status of this morning's deployment. I'm pleased to tell you that the release didn't encounter any issues. It's now live in production. Also, I've verified the health checks. Just let me know if you need any further detail.
A chat reply still does not need 44 words. The rules shrink; you cut the rest.The rules
These are the groups the tool applies. Level 1 rules are safe to apply in one pass. Level 2 and level 3 rules change punctuation or delete a whole clause, so you apply those one at a time.
| Rule | Before | After |
|---|---|---|
| Contract the verb | I am, do not, it is, we will, cannot | I'm, don't, it's, we'll, can't |
| Split the semicolon | …on my end; I have reviewed… | …on my end. I have reviewed… |
| Plain connector | However, Additionally, Therefore, | But, Also, So |
| Plain word | utilize, assist, regarding, prior to | use, help, about, before |
| Cut the padding | in order to, as per, a number of | to, per, several |
| Direct the ask | I would like to confirm | I want to confirm |
| Give a real time | at your earliest convenience | when you can |
| Delete the opener | I hope this email finds you well. | (deleted) |
| Delete the thanks | Thank you for reaching out. | (deleted) |
| Shorten the closer | Please do not hesitate to reach out | Just let me know |
| Shorten the sign-off | Best regards, | Thanks, |
| Drop the serial comma | the file, the log, and the summary | the file, the log and the summary |
Score your own draft
Paste the reply, read the stiffness score, and apply one edit at a time. Your draft stays in the browser. Nothing is uploaded and nothing is rewritten.
Open the Loosen tool →What not to do
Do not ask the model to "make it casual". It adds words you did not write. You get invented slang, forced warmth and an exclamation mark you would never type. That is a second layer of generic writing on top of the first.
Do not lowercase everything. Lowercase does not read as relaxed. It reads as a style choice, and in a work reply it reads as careless.
Do not add typos on purpose. A deliberate mistake is still a mistake. The reader either fixes it for you or wonders whether the numbers are wrong too.
Loosening is subtraction. Cut the furniture, contract the verbs, break the long sentences, and keep every fact you started with. If the draft still says nothing after that, the problem was never the polish.
Questions
Why does my AI reply sound too polished?
The model writes at the register of a formal letter. It contracts nothing, punctuates every clause correctly, and wraps the answer in an opener and a closer. Each habit is correct on its own. Together they read as a document, not a reply.
Should I ask the model to make it casual instead?
No. A casual rewrite adds words you did not write, usually fake slang and forced warmth. Loosening removes and shrinks instead. Cut the opener, contract the verbs, split the semicolons, and keep every fact you started with.
Does a polished reply prove someone used AI?
No. Careful human writers produce polished prose, and many people write formal email by habit. Polish is a reason to reread a draft, never proof of who wrote it.