How to read a score
What the verdicts and colours mean, why a score is an estimate, and where it is weakest.
A score is an estimate, not proof
DropSlop's detector gives each text a score from 0 to 1: its estimate of the chance that a machine wrote it. It is a signal to help you edit, not proof of who wrote something. Like every detector, it sometimes marks writing a person wrote and misses writing a machine produced. Never use a score on its own to judge a person or their work. That is why DropSlop always shows you what it marked, not only a number.
The three verdicts
The score is shown as one of three bands:
| Verdict | Score |
|---|---|
| Reads like a person | under 0.15 |
| Partly machine-made | 0.15 to under 0.45 |
| Reads machine-made | 0.45 and over |
Press Show the numbers on any result to see the scores themselves.
The whole draft and single paragraphs
- The verdict at the top is for the whole draft. It is the most reliable reading DropSlop gives, because the detector is built to read whole articles.
- Paragraph tones are rough. Each paragraph is coloured on its own, but one short paragraph gives the detector very little to go on, so its tone can swing either way. Use paragraph colours to see where to look, and the whole-draft verdict to judge the draft.
- Very short text isn't scored. Paragraphs under 12 words, headings, list items, quotes and code are left out and marked "not scored".
Long drafts: the first 3,500 characters
On a long draft, the part of the detector that reads your wording looks at about the first 3,500 characters (roughly the first 600 words). Past that, it only counts things such as sentence length and stock words. So:
- an edit far down a long draft can move the whole-draft score less than you expect, and the X-Ray tells you when an edit sits past that point;
- to judge the later part of a long piece, check it on its own as well.
What the marks mean
- Underlined words are tells: stock phrases, openers like "Moreover," and other habits machine writing shares. Most come with a suggested cut.
- A dotted underline marks a sentence that is very predictable. It is a pointer only, with no suggested edit.
- The bars and cells show one bar per sentence and one cell per word, so you can see the rhythm: sentences that are all about the same length are a common machine habit.
Where it is weakest
DropSlop works on English prose. It is least reliable on:
- short or very formulaic text, such as product blurbs and boilerplate;
- business pages written by people, which can read machine-made to it;
- text from the newest AI models, which it can miss.
If a result surprises you, read the marks and decide for yourself.