This is not an AI detector. Paste things you have written, then a draft, and see how far the draft has drifted from you.
Runs in your browser. Nothing leaves this tab.
Example: my draft measured against the five articles at /writing. Build a profile from your pieces to compare your own draft.
Paste at least five pieces, separated by a line of three dashes. A thousand words is a useful starting point, not a calibrated threshold.
0 pieces · 0 words
Saving and deleting your profile
What gets saved is a frequency table: up to one hundred marker words that passed the document and variation filters, a number beside each, and the rates. Your own words, then, but single ones, in frequency order, never in the order you wrote them. No sentence goes in and none can be got back out.
Paste the draft you want to compare. The conservative minimum is 150 words; shorter drafts get counts, but no distance.
167 words · 150 needed for a distance
Example report
Distance
1.87
This compares with the variation in your own samples. It is a guide to review, not a quality score or an AI verdict.
Reference and measurement details
The distance is Burrows's Delta, and a Delta is measured in standard deviations, so it needs a population whose standard deviations they are. That population is your pieces: your own commonest words, and your own variation from one piece to the next. Which is why the number reads in units of your writing and not mine, and why the spread of your own pieces is printed beside it.
Built from: 5 pieces, 3502 words, 100 marker words.
Your own spread: 0.94 to 1.12, median 1.07, across 5 of your own pieces. This draft is at 1.87.
Words these samples do not use
- These samples never use "demonstrate". They use "show", 1 times. This draft uses "demonstrate" 2 times.
- These samples never use "commence". They use "start", 5 times. This draft uses "commence" 1 times.
- These samples never use "empower". They use "let", 2 times. This draft uses "empower" 1 times.
- These samples never use "furthermore". They use "and", 117 times. This draft uses "furthermore" 1 times.
- These samples never use "in order to". They use "to", 53 times. This draft uses "in order to" 1 times.
- These samples never use "obtain". They use "get", 18 times. This draft uses "obtain" 1 times.
- These samples never use "prior to". They use "before", 7 times. This draft uses "prior to" 1 times.
- These samples never use "seamless". They use "smooth", 1 times. This draft uses "seamless" 1 times.
- These samples never use "utilise". They use "use", 3 times. This draft uses "utilise" 1 times.
The near-synonyms come from a fixed table of 22 pairs written into this tool. It is not a thesaurus and it cannot find a pair that is not on the list. What makes a row worth printing is your own frequency: the word is in your draft, never in your samples, and the plain one is.
Habits, side by side
Habits, side by side
| Habits, side by side | You | This draft |
|---|---|---|
| Words per sentence, mean | 14.90 | 23.86 |
| Words per sentence, spread | 8.48 | 9.91 |
| Sentences of 8 words or fewer | 0.28 | 0.00 |
| Sentences over 32 words | 0.03 | 0.14 |
| Em dashes per 1,000 words | 0.00 | 11.98 |
| En dashes per 1,000 words | 0.00 | 0.00 |
| Semicolons per 1,000 words | 0.00 | 0.00 |
| Exclamation marks per 1,000 words | 0.00 | 0.00 |
| Question marks per 1,000 words | 5.43 | 0.00 |
| Bracketed asides per 1,000 words | 0.00 | 0.00 |
| Contractions per 1,000 words | 7.14 | 5.99 |
| Sentences opening with and | 0.02 | 0.00 |
| Sentences opening with but | 0.00 | 0.00 |
| Sentences opening with so | 0.01 | 0.00 |
Sentence lengths
| Sentence lengths | You | This draft |
|---|---|---|
| 1-8 | 0.28 | 0.00 |
| 9-16 | 0.34 | 0.29 |
| 17-24 | 0.22 | 0.14 |
| 25-32 | 0.13 | 0.43 |
| 33+ | 0.03 | 0.14 |
Sentences are split on full stops, question marks and exclamation marks. It does not know abbreviations, so Dr. Byrne counts as two.
The sentences pulling hardest
- In today's fast-paced software landscape, it is essential to leverage robust testing methodologies in order to ensure that concurrency defects are surfaced prior to deployment.a word your samples never use
- By pinning the identifiers such that the two rooms order their overlapping windows in opposition, the same harness immediately demonstrated six deadlocks across eight rounds, thereby facilitating a targeted remediation and empowering the team to obtain confidence in the fix.a word your samples never use, longer than your usual
- Our team commenced an investigation into a deadlock condition affecting the group booking pathway — a critical revenue surface — and utilised a comprehensive suite of concurrent test rounds to validate the behaviour.an em dash, a word your samples never use
- It is therefore imperative that engineering organisations delve into the assumptions embedded within their test fixtures, rather than assuming that a passing suite is equivalent to correct behaviour.a word your samples never use
- Furthermore, the underlying issue was not the locking strategy itself but the manner in which the fixture generated its identifiers.a word your samples never use
Can't see
Method and limitations
- Meaning. Every number here counts how often words and marks turn up, and none of them knows what any of it says.
- Register shifts inside one writer. A note to a friend and a note to a bank are two voices from the same person, and this would call the second one drift.
- Anything under 150 words. The tool uses that conservative floor because short drafts make marker counts sparse; it refuses to print a distance and says why.
- Anything from fewer than five pieces. Five is a conservative, uncalibrated floor for estimating variation, so the tool refuses a distance below it and says why.
- Whether the writing is any good. A low distance means your commonest words turn up at similar rates. That is not praise, and it is not a verdict on the draft.
- A substitution that is not on its list. The near-synonyms come from a fixed table written into this page, not from a dictionary and not from a model.