How to Use the LLM Burstiness Forensic Stylometer
Detect whether text was written by a human or generated by AI by analyzing "burstiness" (sentence length variance), vocabulary entropy, repetition patterns, and transition word density.
Step 1: Paste the text you want to analyze. The tool works best with 200+ words of continuous prose.
Step 2: Click "Analyze Burstiness" to see a comprehensive forensic breakdown including a human writing probability score, a visual sentence-length chart, and specific forensic notes.
Burstiness: The Statistical Fingerprint of Human Writing
When humans write, they produce sentences of wildly varying lengths. A short punchy sentence. Then a longer, more complex one that builds on the previous thought, adding nuance and context that the reader needs to fully understand the argument being made. Then another short one. This natural rhythm is called "burstiness," and it is one of the most reliable statistical indicators of human authorship.
Why AI Writing Is "Flat"
Large language models generate text one token at a time, choosing each word based on probability distributions. This produces a characteristic "flatness" in sentence structure. AI sentences tend to cluster around a consistent length (15 to 22 words), use transition words at unnaturally high rates ("Furthermore," "Moreover," "In conclusion"), and repeat bigram patterns more frequently than human writers. The result is text that reads smoothly but monotonously, like a news anchor reading from a teleprompter.
What the Tool Measures
The burstiness metric is the coefficient of variation (standard deviation divided by mean) of sentence lengths. Human writing typically has a burstiness value above 0.35. AI writing typically falls below 0.25. The tool also measures vocabulary entropy (how diverse the word choices are), bigram repetition rate (how often two-word combinations repeat), and transition word density (the frequency of formal connectors like "however" and "therefore"). These four metrics are combined into a single Human Writing Probability score.
The Visualization
The sentence-length distribution chart shows each sentence as a vertical bar whose height represents its word count. Human writing produces a jagged, irregular skyline. AI writing produces a more uniform, cityscape-like pattern. This visual representation makes the difference immediately obvious, even without understanding the underlying statistics.
Limitations and Caveats
No statistical method can definitively prove authorship. A human who writes in a very structured style may score as "likely AI." An AI that is prompted to vary its sentence length may score as "likely human." The tool provides a probability estimate, not a verdict. Use it as one signal among many when evaluating content authenticity. For academic integrity investigations, combine statistical analysis with other evidence (writing process documentation, version history, interview assessment).
Frequently Asked Questions
Partially. If a human wrote the original draft and AI polished it, the burstiness will be lower than pure human writing but higher than pure AI writing. The tool will likely classify it as "Mixed Signals." Heavy AI editing (rewriting most sentences) will push the score toward "Likely AI." Light editing (fixing grammar only) will keep it closer to "Likely Human."
This is a heuristic tool, not a forensic laboratory instrument. Its results may support an investigation but would not be considered definitive evidence on their own. For formal academic integrity cases, use this as a screening tool and escalate confirmed concerns to qualified forensic linguists who can perform more rigorous analysis.
Vary your sentence lengths dramatically. Mix 5-word sentences with 30-word sentences. Use contractions. Add personal anecdotes. Reduce transition words. Write in your natural voice and use AI for research and outlining rather than drafting. The more of your own writing rhythm you preserve, the more human the text will score.