# Copylint

> Copylint is a deterministic quality gate for user-facing copy. It reports each finding with line, rule, and excerpt, suggests a fix where the catalog has one, and changes the text only when you accept a fix.

Copylint checks product copy against 148 rules for AI writing tells, shows the rule and line behind every flag, and suggests fixes you can accept or skip. Not a humanizer, not a detector.

2 free scans, no card. Plans from $19 a month.

- It is not a humanizer. It never rewrites a whole draft; every fix is a visible suggestion for one finding that you accept or skip.
- It does not detect AI authorship. A low score says the text avoids a known catalog, not who wrote it.
- It will not help anyone pass GPTZero or another detector.

## Pages
- [Live demo](https://copylint.xyz#demo): Paste a draft on the home page.
- [How a draft gets stopped](https://copylint.xyz#how): Score weights, threshold 40, exit 0 or 1.
- [Pricing](https://copylint.xyz/pricing.md): Open source free, Solo $19 a month, Team $79 a month.
- [Scan API](https://copylint.xyz/docs.md): POST /api/v1/scan.
- [Rules reference](https://copylint.xyz/rules.md): Catalog, weights, and how a score is calculated.

## Straight answers
### Is Copylint a humanizer?
No. A humanizer rewrites a whole draft so it reads as human or slips past a detector. Copylint flags specific tells and suggests fixes for them. Nothing changes until you accept it, and the score never uses a model.

### Does it fix the text for me?
It suggests, you decide. Word-level fixes, such as trimming a wordy connective to one word or cutting a filler adverb, come from the catalog and apply in one click, even in the demo on this page. For a flagged phrase with no plain replacement, the app can rewrite that one sentence, or fix the whole draft in one pass: every safe catalog fix, then a rewrite of each flagged sentence, which costs one rewrite from your monthly allowance per sentence. Every rewrite is scanned again before you see it, the model is told never to add facts your draft does not contain, and you accept or discard the result.

### Does it detect AI authorship?
No. A low score means the text avoids a known catalog of tells. It does not say who or what wrote it, and we do not claim otherwise.

### Will it help me pass GPTZero or another detector?
No. Evasion is out of scope and stays out. The rules exist so readers get concrete sentences, not so a classifier gets fooled.

### Why is the threshold 40?
One critical finding plus two majors trips it. Set it lower per project with a flag. A pass is a floor for review, not a substitute for it.

### What about false positives?
A catalog word can be ordinary in your domain, as a garden-design blog knows. Findings carry line and excerpt so a reviewer can dismiss one in seconds. Per-project allowlists are on the roadmap.

### What happens to the text I paste?
The open-source scanner makes no network requests. The hosted tier stores a scan for 90 days on your account and never trains anything on it. An AI rewrite sends only that sentence and its paragraph to OpenAI, which does not train on it, and we do not store the rewrite.

### What can it scan?
Plain text and Markdown from anywhere: paste it in the app, run the CLI on files, send it to the API, or let the GitHub Action scan the Markdown a pull request changed.

### Can I cancel?
Yes. Plans are monthly and cancel from the billing page at any time, and access lasts until the end of the period you paid for. Every account starts with two free scans, no card needed.

HTML: https://copylint.xyz. Last updated 2026-09-17.