AI Text Detector

Heuristic pattern matcher for ChatGPT / Claude writing tells — English & Persian. Evidence, not proof.

Tip: paste ≥100 words for a reliable verdict — Persian (فارسی) rules apply automatically.

This is a heuristic scorer — a high score is not proof, and clean human prose can score high while casual human text scores low. Signals are drawn from Wikipedia's “Signs of AI writing”, studies of ChatGPT's overused vocabulary (“delve”, “pivotal”, …), and 2026 corpus studies of Claude-specific tells (“is genuinely”, “less like X, more like Y”). Persian (فارسی) has a dedicated rule set: bureaucratic register («می‌باشد», «لازم به ذکر است», «در راستای», «مورد بررسی قرار»), significance inflation («نقش بسزایی»), marketing/SEO tells, travel-brochure vocabulary, vague attribution, colloquial counter-signals (می‌شه، واسه، اگه) and informal-punctuation detection — RTL rendering included. Markdown and code detected in the paste are excluded from language analysis. Hidden Unicode is scanned as its own evidence family: zero-width spaces, bidi controls, Unicode tag characters, narrow no-break spaces and homoglyphs are highlighted, and long runs raise a flag. Note: Anthropic's official Claude watermark (Aug 2026) is statistical — no characters are added — so a clean Unicode scan is not proof of human authorship.

Signal breakdown

Evidence