How Naturable finds AI writing
Naturable's free checker uses no AI at all. It relies on deterministic rules — the same kind of pattern-matching a spell-checker uses — so it's instant, private, and costs nothing to run. Here's exactly what happens when you hit Check.
1. Pattern detection (33 tells)
The engine scans your text for patterns that repeatedly show up in AI-generated writing, catalogued in Wikipedia's Signs of AI writing guide. Each match is highlighted, explained, and paired with a fix. Categories include:
- Vocabulary tells — over-used LLM words like delve, testament, landscape, leverage, robust, seamless.
- Cadence tells — "It's not just X, it's Y", the rule of three, manufactured punchlines.
- Filler & hedging — "in order to", "due to the fact that", stacked hedges.
- Structure tells — em-dash overuse, emoji, bold overuse, Title Case headings.
- Chatbot leftovers — "Great question!", "I hope this helps", cutoff disclaimers.
About two-thirds of these patterns can be detected and fixed with pure rules — which is exactly why the free tier can offer them unlimited.
2. Sentence-rhythm analysis (burstiness)
AI writing tends to be uniform: sentences hover around the same length with the same structure. Human writing has burstiness — it alternates short, punchy sentences with longer, winding ones. Naturable measures this as the coefficient of variation of your sentence lengths:
burstiness = standard deviation of sentence length ÷ mean sentence length
This is pure arithmetic — no model required. Higher variation reads more human, and we surface it as your Sentence rhythm score.
3. The scores
Two headline numbers, both heuristic:
- Human score (0–100) — the inverse of an "AI-tell index" built from tell density, the count of strong tells, and a penalty for low burstiness. Higher is cleaner.
- Sentence rhythm — your burstiness, scaled to 0–100.
These are guidance, not a verdict. Naturable is a writing-quality tool, not a detector, and no tool can promise a specific outcome against any third-party checker.
4. Fixing: rules first, AI when needed
Fixes come in two tiers:
- Quick fix (free) — deterministic replacements the rules can make safely: cutting em-dashes, removing emoji and chatbot chatter, "in order to" → "to", straight quotes, and so on.
- AI rewrite (Pro) — the semantic tells rules can't safely rewrite — significance inflation, synonym cycling, generic conclusions — where a model has to understand meaning. This is the paid tier.
Privacy
The free checker runs entirely in your browser. Your text is never uploaded. Only the Pro AI rewrite sends text to a server, and only when you opt in. See our Privacy Policy.