What makes writing sound like AI
Writing sounds like AI when it comes from a model’s defaults: a stock opener, a hype word where a fact should be, a tidy triad, a closing summary of what you just read. Each is fine alone. Together, and with no detail only the writer could know, they read as machine-made.
The 12 AI writing patterns it checks
The copy patterns follow Peter Yang’s no-ai-slop skill. The ones marked “quoted” are matched against your exact words. The rest are judged across the whole text.
- “It’s not just X, it’s Y”quoted
- Fix: Say the one specific thing. Drop the contrast setup.
- Slop vocabularyquoted
- Fix: Replace each hype word with what actually happens.
- Stock openers and fillerquoted
- Fix: Start with the point. Cut the sentence that clears its throat.
- Stock closersquoted
- Fix: End on the last real point. Do not summarise a short text.
- Faux-insight setupsquoted
- Fix: Cut the setup. Make the claim on its own.
- Em dash habitquoted
- Fix: Use a period or a comma.
- Decorative emojiquoted
- Fix: Cut decorative emoji. Keep one only where it carries information.
- Stacked hedgesquoted
- Fix: Commit to the claim, or say what you do not know.
- No specificsjudged
- Fix: Add one detail only you know: a number, a name, what happened.
- Uniform rhythmjudged
- Fix: Break the pattern once: one very short sentence, or one long one.
- Balanced on every pointjudged
- Fix: Pick a side and say why.
- Triads everywherejudged
- Fix: Keep the one item that is true and drop the other two.
How the AI writing score is worked out
60% of the score is Jev’s answer to one question about the whole text: was it most likely written or heavily rewritten by a language model? The other 40% is how strongly the eight strongest patterns fire, each weighted by how much it gives away. A quoted pattern only counts when the exact words are in your text.
| Score | Verdict | Means |
|---|---|---|
| 80–100 | Reads like a model | Almost every sentence is a default. |
| 60–79 | Heavily machine-flavored | A person edited it, lightly. |
| 40–59 | Some AI tells | Mixed: real voice with stock phrases. |
| 20–39 | Mostly human | A few stock phrases slipped in. |
| 0–19 | Reads human | Specific, uneven, and somebody's own. |
AI writing checker questions
- Does this sound like AI?
- The checker answers that by listing what it found. It looks for stock phrases like 'It's not just X, it's Y', hype words like seamless and unlock, filler openers, and for the lack of any specific detail. Each pattern it finds is quoted from your text, so you can see exactly what reads as machine-made.
- Is this an AI text detector?
- No. It does not guess whether a person or a model typed the words. It finds patterns that AI writing tends to use. A human can write all of them, and a model can be prompted to avoid them. Treat the score as how much the text sounds like AI, not as proof of how it was made.
- How is the score worked out?
- Jev answers one question about the whole text, which carries 60% of the score. The other 40% is how strongly the eight strongest patterns fire, each weighted by how much it gives away. A pattern with a quote only counts if the exact words are in your text.
- Is my text saved?
- No. Your text is sent to the models once to be judged, and it is not stored. No report page is made, so there is nothing to share or leak.
- Why only 5 checks a day?
- Each check is a model call that costs real money. The limit keeps the tool free for everyone.
Built on Jev
Many small yes/no judgments, each with a probability, in one fast call. Checking a website instead? Try the AI Slop Detector. Writing a page for search? Check its meta title and description. Or see what Jev is.