AI Tone Detector

About AI Tone Detector

Paste a passage and this tool lists exactly where AI-flavoured features hit: boilerplate connectives (Firstly/Moreover/In conclusion), em-dash sprinkling, suspected parallelism, uniform sentence rhythm, and buzzword clusters. It deliberately never shows an "AI rate %" — percentages are pseudo-precision and real detectors are controversial enough already. It's a heuristic first-pass: spot the heavy spots, rewrite them yourself, then check a real detector (GPTZero, Zhuque, ZeroGPT) if you need one. Everything runs locally in your browser.

FAQ

Why is there no "X% AI" score?

Because that number would be pseudo-precision. Feature hits are countable facts ("in conclusion" appears 3 times), but there is no reliable formula from hit counts to "probability of AI" — any percentage is invented. False positives make it worse: human officialese, essays and press releases hit these features heavily, while lightly edited AI text can lose them all. So this tool reports facts (what hit, where) plus a coarse band, and leaves the judgement to you. If you want a number, use a real detector — and read its false-positive rate alongside.

Which features are checked, and at what thresholds?

Chinese rule set: boilerplate connectives (2+ hits), em-dashes (2+), suspected parallelism (3+ consecutive near-equal clauses of 6+ chars), AI-favoured buzzwords (2+), hollow openings ("随着…的发展"-style), uniform sentence length (5+ sentences with CV ≤ 0.22), numbered scaffolding (3+ markers), hollow recap closings. The English set covers connectives, em-dashes, buzzwords (delve/tapestry/leverage…), hollow openings and uniform length. Every threshold is an experience-based noise floor, trading false positives against misses — and all of them are visible in the open source.

After this first pass, which real detector should I use?

Depends on language and use case: for Chinese, try Zhuque first (Tencent's free lab tool, strongest on Chinese, covers text/image/video, daily quotas); for English and education, GPTZero (sentence-level highlights, free 10K words/month, limited on Chinese); for a no-sign-up quick check, ZeroGPT (free 15K characters per check, multilingual, ad-supported). We maintain review pages for all three with pricing and trade-offs. And once more: no detector is guaranteed accurate — treat every result as advisory evidence, never as proof.