Step 1: Select Transformation Mode
Choose your target mode (Deep Telemetry, Academic Rubric, Editorial Scan) in the toolbar.
Statistical + Neural Telemetry
Zero-Retention RAM Execution
AI Content Detector & Structural Uniformity Analyzer
Inspect the exact statistical signals—sentence length standard deviation, type-token diversity, and structural predictability—rather than relying on opaque black-box guesses.
0 words · 0 / 10,000 chars
Ctrl+Enter to Run
Words: 0
Burstiness Index: —
Readability Score: —
Engine Status: M33 Neural Vault Ready
⚡ Fast Flow Handoff:
How to Use the ZeeML AI Content Detector
Step 2: Paste Source Passage
Enter or paste your draft into the left source pane.
Step 3: Run & Inspect Telemetry
Click "Analyze Text Origin" to generate the output and review burstiness and readability scores.
Quick Formula & Operational Reference Cards
Sentence Length Variance (σ)
Measures how much sentence lengths fluctuate across a passage. Uniform 18-word sentences indicate synthetic generation.
Type-Token Ratio (TTR)
Calculates unique vocabulary diversity relative to total word count.
Transition Density
Detects overuse of formulaic sentence starters typical of instruction-tuned LLMs.
Worked Step-by-Step Benchmarks
Uniform LLM Paragraph
Input: Three consecutive 19-word sentences starting with "Furthermore," "Moreover," and "Additionally."
Result: Flagged for low burstiness (18/100) and high transition predictability.
Human Technical Writing
Input: Mix of a 6-word thesis, a 31-word empirical derivation, and an 11-word conclusion.
Result: Classified as natural human cadence (Burstiness: 76/100).
How Statistical Perplexity & Burstiness Telemetry Works
Autoregressive language models sample tokens from a truncated probability distribution, which minimizes surprising word choices (low perplexity) and keeps sentence lengths clustered around the training mean (low burstiness).
ZeeML’s AI Content Detector computes transparent, reproducible metrics across your draft so you can see exactly which structural features contribute to the uniformity score.
Frequently Asked Questions
Can AI detectors ever produce false positives on human writing?
Yes. Highly formulaic legal boilerplate or rigid five-paragraph essays can exhibit low burstiness. That is why ZeeML displays the underlying C1-C3 predictability, flow, and rhythm signals alongside the summary score.
Is my submitted document stored when I run an AI detection scan?
Never. All scans run ephemerally in memory and are discarded immediately after returning your telemetry report.
What do the C1, C2, and C3 forensic signals measure?
C1 measures word-choice surprisal (token predictability), C2 evaluates phrase transition smoothness across bigrams, and C3 quantifies sentence-length rhythm variance (burstiness).
How do I inspect sentence-by-sentence AI probability highlights?
Run an analysis or click "Diff Highlights" in the output pane to view every sentence color-coded inline (red for AI-flagged, amber for mixed, and emerald for human cadence).