See the numbers a detector is actually looking at.
Most tools hand you a percentage and no reasoning. This does the opposite: no score at all, just the measurable statistics — sentence-length variance, burstiness, vocabulary diversity, repeated openers — so you can judge the text yourself. Everything runs in your browser.
Runs entirely in your browser. Nothing you paste is uploaded, logged or stored.
Paste some text above and the statistics will appear here.
What these numbers mean
Burstiness
The variation in sentence length, expressed relative to the average so it does not change if you write in long or short sentences overall. Human writing tends to lurch: a winding sentence, then three words, then an aside. Generated text tends to hold a steadier pace.
Vocabulary diversity
The share of words in your text that are distinct. Low diversity means the same words recur, which happens naturally in technical writing and also in output from smaller models. It rises and falls with text length, so compare like with like.
Repeated openers
How often sentences begin the same way. Readers notice this as monotony long before any tool flags it, and it is one of the easiest things to fix in your own drafting.
Read this before drawing a conclusion
What these figures are
- +Exact. They are arithmetic on your text, not an estimate — the same numbers however many times you run it.
- +Half the picture. Detectors combine variance measures like these with model-based predictability, which is the part this page cannot compute.
- +Useful for editing. Very even pacing and repeated openers are worth fixing whether or not anything generated the text.
What they are not
- –Not a detection result. There is deliberately no score on this page, because these statistics alone cannot support one.
- –Not evidence about a person. Low variance is normal in abstracts, legal drafting, instructions and translated text, and non-native writers often write with more uniform sentence lengths.
- –Not a target to optimise. Rewriting purely to raise the number is what humanizer tools do, and it tends to make writing worse while proving nothing.
- –Not comparable across languages. Sentence length behaves differently in German, Japanese or Arabic, so a value that looks low in English may be ordinary elsewhere.
Frequently asked questions
Why is there no AI score?
Because these statistics on their own do not justify one. Burstiness is a real input to detection, but a detector combines it with model-based predictability and a classifier trained on many samples. Showing a score derived from variance alone would be exactly the kind of confident-looking number this page exists to argue against. If you want a scored result, use the detector — it shows the sentences behind the number.
What is a good burstiness value?
There is no target, and treating one as a goal is a mistake. As a rough orientation, tightly uniform prose sits below about 0.25 and conversational writing often lands above 0.6, but a technical abstract legitimately sits at the low end. The distribution chart is more informative than the single figure.
Why do you need five sentences?
Variance across three or four sentences is dominated by chance. One long sentence in a short sample swings the figure enormously. Rather than show a number that would move wildly on trivial edits, we withhold it — the same reason we refuse to score very short social captions.
Is my text uploaded?
No. The analysis is plain JavaScript running on your device. There is no request to our servers, nothing is logged, and it works with the network disconnected.
My own writing shows very even sentences. Is that bad?
Not inherently. It may mean your subject calls for it, or that you edit thoroughly — both are fine. If you want more variety, the fastest fix is to break one long sentence in each paragraph and let another run on. But do not chase the number for its own sake; write for the reader.
Want the scored version?
The detector adds model-based analysis and shows which sentences drove the result.
Try the AI detector