Originality.ai score: what the percentage means, as far as anyone outside knows
An Originality score is the AI-likelihood reading from Originality.ai, a paid detection product used mostly by publishers, agencies and content teams who buy writing from others. The report shows a percentage for AI and a percentage for original, with sentence-level highlighting, and it sits beside a separate plagiarism scan. Humanize 360 is a different tool with a different scale, and we do not know Originality.ai's internals. This page describes what the product reports in general terms and what remains unknown.
The company has said publicly that it trains classifier models on human and AI samples and has released several model versions over time with different names and trade-offs, including versions tuned for fewer false positives. Which version scans your text and how its threshold is set are product decisions the vendor can change. Any specific number you read about "what counts as AI on Originality" should be treated as that writer's observation on that day, not a fixed rule.
Reading the two percentages
The AI and original figures are a confidence split, not a word count. A reading that says most of the text is likely AI does not mean most words were machine-written; it means the classifier leaned that way for the document as a whole. The sentence highlighting is more useful for editing than the headline number, because it shows which passages carry the texture the model responds to.
Why publishers use it
- It combines AI detection and plagiarism scanning in one report for outsourced content.
- It offers team accounts and scan history, which suits an editorial workflow.
- It is marketed on aggressive detection, which some buyers want and some writers dislike.
- It is paid per scan or by subscription; check their pricing page for current terms.
What is known and what is not
Known: it is a trained classifier approach, it updates models, it reports per-sentence results, and its vendor publishes its own accuracy tests. Not known to anyone outside the company: the training data, the current weights, the exact threshold and how often any of these change. That is normal for a commercial detector and it is why two scans of one document weeks apart can differ. We make no claim about how a text will score there.
How the Human Score differs
On /ai-detector the reading runs 0 to 100 with higher meaning more human, and the eleven signals are named and shown: sentence-length variance, word predictability, stock-phrase density, opener repetition, tell-word ratio, punctuation patterns, paragraph uniformity, contraction rate, vocabulary spread, transition-word overuse and structural sameness. There is no trained classifier, no third-party model, and the thresholds are published. The plagiarism checker is a separate paid tool, kept apart on purpose because the two questions are different.
Common questions
What Originality.ai score is safe?
We do not know their thresholds and will not invent one. Their report labels the verdict; read that label and their own documentation rather than any number quoted on a forum.
Does Humanize 360 predict an Originality.ai result?
No. We measure the general statistical tells that all detectors respond to, and we cannot know their current model. A high Human Score is not a promise about any vendor.
Why is Originality.ai known for flagging human writing?
Aggressive tuning catches more machine text and more human text at the same time. Formal and formulaic prose is most at risk. Our detector shows the specific signals so a writer can point to them.
Is the plagiarism part the same as the AI part?
No. Plagiarism scanning searches an index for matching text. AI detection estimates texture. Originality bundles them in one report; we keep them as separate tools.