AI Content Burstiness Checker

Measure sentence-length variance and lexical diversity to estimate whether text was written by a human or generated by AI. Nothing leaves your browser.

—

—

—

Analysis Breakdown

Statistics

How AI Text Differs from Human Text

No detector is perfect — this tool is heuristic, not definitive. But two statistical properties are genuinely different between the two:

1. Burstiness (Sentence-Length Variance)

Human writers naturally mix short punchy sentences with long complex ones. LLMs tend to produce sentences of similar length — often clustered around 15–25 words. The coefficient of variation (stdev ÷ mean) of sentence length is therefore lower in AI text.

  • Human writing CV: typically 0.45–0.9
  • AI writing CV: typically 0.25–0.4

2. Lexical Diversity (Type-Token Ratio)

The TTR is the number of unique words divided by total words. AI text often has a lower TTR because LLMs default to a narrower vocabulary and recycle phrases. Human writers — especially good ones — use more varied vocabulary and include proper nouns, idioms and unusual verbs.

  • Human TTR (500 words): typically 0.42–0.55
  • AI TTR (500 words): typically 0.32–0.42

Other Signals This Tool Checks

  • Average sentence length — AI often produces 18–24 words per sentence consistently.
  • Contraction usage — human writing uses more contractions (don't, can't, it's).
  • Transition word density — AI overuses "moreover", "furthermore", "additionally", "in conclusion".
  • Paragraph length uniformity — AI tends toward similar-length paragraphs.

Limitations — Read This

This tool will never give a definitive verdict. A human writing in a formal academic register may score as "AI-like". A language model prompted to write in a deliberately varied style may score as "human-like". Use it as one signal among many — combined with your knowledge of the author, the context of publication, and any editorial process.

It is not admissible as evidence for academic misconduct or editorial punishment. False positives are common on short samples under 150 words. For any sample under 200 words, treat the result with suspicion.

How to Humanise AI-Assisted Content

  1. Vary sentence length dramatically — pair a 4-word sentence with a 32-word one.
  2. Add specific details — numbers, names, dates, personal anecdotes.
  3. Use contractions and first-person voice.
  4. Cut generic transitions — replace "moreover" with a plain period.
  5. Break the pattern of "topic sentence + three supporting sentences" paragraph structure.
  6. Add a genuine opinion or nuance an LLM would smooth over.

AI Detection FAQ

Is this tool 100% accurate?

No — and neither is any commercial detector. This tool measures statistical properties (variance, diversity) that differ between human and AI text. It gives a probability estimate, not proof.

What is burstiness exactly?

Burstiness is the coefficient of variation of sentence length — standard deviation divided by mean. High burstiness means sentences vary wildly (human-like). Low burstiness means they're all similar length (AI-like).

Why does short text score poorly?

Statistical measures require enough data. Under 150 words, the sample size is too small to establish reliable variance or diversity figures. Aim for 300+ words.

Can AI be prompted to write with high burstiness?

Yes. A well-crafted prompt can produce bursty, diverse text that scores as "human". This is the fundamental limitation of all statistical detectors.