Is my writing AI? Check it with the free local AI-tell checker

Paste anything: an email, a chapter, a cover letter. The counts update as you type, and the example button loads a passage built to trip every category so you can see what a heavy result looks like.

This is the measurement that justifies the tool’s existence. Take the built-in example passage, which lands at 96 out of 100. Run it through the mechanical dash remover — every em dash and typed dash gone, spacing repaired, wording untouched. The score falls to 68.

Twenty-eight points is a big movement for deleting one character, and it is exactly why dash-heavy drafts look so machine-written. But 68 is still deep in the “many patterns” band, and here is why: all 14 catalogue terms are still there. Removing punctuation cannot remove vocabulary, sentence uniformity, or the absence of specifics. Those need editing, which is the honest boundary of any mechanical fix — and the same reason no remover will ever move a real detector score.

The score is a weighted sum of the counts, out of 100. It exists so you can compare two drafts at a glance, not so you can quote it at anyone. Bands are deliberately conservative:

• Under 28 — few patterns. Nothing is clustering. Ordinary human prose lands here most of the time.
• 28 to 54 — some patterns. A handful of catalogue items at normal rates. Worth a light edit, not a rewrite.
• Over 55 — many patterns. Multiple items clustering at once. This is the shape of an unedited model draft, and it is also the shape of a style guide written by a committee.

• Who wrote the text. Nothing in the character statistics identifies an author, human or machine.
• Whether a detector will flag it. Commercial detectors score token probabilities against a model. Those scores move when you edit punctuation, which tells you what they are and are not measuring. See how accurate AI detectors are.
• Whether the content is true. Every fact, citation and figure needs checking separately. Style analysis says nothing about accuracy.
• How non-English text compares. The vocabulary lists are English. Other languages are still measured for dash density, punctuation and sentence-length uniformity, and the panel says so.

1. Fix the mechanical items first: run the draft through the AI dash remover so the dash family is gone and the wording is untouched.
2. Address the vocabulary. Open the word and phrase list and replace each hit with something specific, or delete it.
3. Fix the rhythm by hand. This is the part no tool can do: put a four-word sentence next to a thirty-word one, and let one paragraph run to a single line.
4. Add detail. A date, a price, a name, a quote. Specifics are both the best writing advice and the strongest anti-tell there is.