AI Writing Tells Analyzer AI & Writing
Measure the stylistic markers associated with AI writing — dash density, sentence uniformity, overused phrases. Not a detector.
AI Writing Tells Analyzer is a free online tool that measure the stylistic markers associated with AI writing — dash density, sentence uniformity, overused phrases. Not a detector. It runs entirely in your web browser using plain JavaScript, so files are processed on your own device and never uploaded to a server. There is no sign-up, no file size limit imposed by the site, no watermark and no paid tier.
- Price
- Free — no account, no quota, no watermark
- Category
- AI & Writing
- Where it runs
- In your browser, on your device
- Files uploaded
- None
- Technology
- plain JavaScript
- Settings
- 1
- Works offline
- Yes, after the first visit
About AI Writing Tells Analyzer
This is not a detector, and the distinction matters enough to lead with. AI detection does not work: OpenAI retired its own classifier for low accuracy, and peer-reviewed work since has found both high false-positive rates and a bias that penalizes non-native English writers. What is measurable is style. Em dash density, sentence-length variance and the frequency of a known set of overused phrases are all countable facts about a passage — and the score here reports those and nothing more.
How to use it
- Paste or type your text. Every figure recalculates on each keystroke.
- Use the sample button to see what the numbers look like on real prose.
The one setting
| Setting | What it does | Default |
|---|---|---|
| List every phrase found | On or off. On by default. | On |
Under the hood
| Runs on | plain JavaScript — runs the whole thing |
| Controls | List every phrase found |
Questions
Can I use this to check whether a student used AI?
No. Please do not. The score measures writing habits, not authorship, and careful human writing — technical documentation, corporate prose, anything edited for consistency — scores high for the same reasons model output does. Using a number like this as evidence about a person is exactly the failure mode that got AI detectors withdrawn.
What is sentence length variance?
The coefficient of variation of sentence lengths — the spread relative to the average. Human prose typically lands between 0.4 and 0.7 because people mix short punchy sentences with long ones. Generated text tends to cluster tighter, having settled into a rhythm. It is a weak signal on its own, which is why it is capped at a fifth of the score.
Why do these particular words show up?
They are the ones that recur across model families rather than being specific to one product — delve, tapestry, realm, testament, seamless, robust, myriad, and the connective openers like "Furthermore" and "Moreover". Nothing is wrong with any of them individually; the tell is density.
My own writing scored high. Should I change it?
Only if the specific findings point at something worth changing. Four em dashes per thousand words is a habit worth noticing; a sentence variance of 0.3 means your sentences are all the same length, which is worth fixing regardless of who wrote them. The overall number is a prompt to look, not a verdict.