Human Score: how our 0 to 100 detector scale is built
Human Score is the number the Humanize 360 AI detector returns for any text: 0 to 100, where higher means the writing reads more like a person wrote it. Above 70 reads human. Between 45 and 70 is mixed. Below 45 is very likely machine-written. Unlike a bare percentage, the score comes with a breakdown of the eleven signals that produced it, so you can see which habits are pulling it down and fix those rather than rewriting everything.
Two short passages show the range. "Furthermore, effective communication is crucial in today's fast-paced business environment. Additionally, it fosters collaboration and enhances productivity." That would land well below 45: stock phrases, a tell-word, two transition openers, twin sentences of similar length. "Nobody in the Kampala office read the memo. We found out because the printer was still warm on Friday." That would sit comfortably above 70. Neither passage is better English. One of them has a person in it.
The eleven signals behind the score
- Sentence-length variance (burstiness) and word predictability (perplexity).
- Stock-phrase density, tell-word ratio and transition-word overuse.
- Opener repetition, paragraph uniformity and structural sameness.
- Punctuation patterns, including em-dash and rule-of-three density.
- Contraction rate and vocabulary spread.
How the signals become one number
Each signal is measured against ranges observed in ordinary human prose and converted to a contribution. Some carry more weight than others: sentence rhythm and word predictability matter most, punctuation habits least. The contributions are combined and scaled to 0 to 100. There is no trained classifier and no third-party model; the analyser is our own code, runs in milliseconds and discards anonymous text as soon as the result is sent. The same text always gets the same score.
What each band means for you
- Above 70: leave the text alone. Rewriting rarely helps and can make it worse.
- 45 to 70: open the breakdown. Usually two or three signals are low. Fix those by hand or run Standard mode at Balanced strength.
- Below 45: the full set of tells is present. Use Heavy strength, or Academic mode for coursework, then check again.
- Any band, when assessing someone else: a score is not evidence of authorship. See /ai-detector-for-teachers.
What the score is not
It is not a prediction of what Turnitin, GPTZero, Originality or any other tool will report. Those products use their own models and thresholds, change them without notice and disagree with each other. A high Human Score means your text does not carry the common statistical tells. That is the part you control, and it is the only thing we claim to measure.
Common questions
Is a Human Score of 70 a pass?
It is the point above which text reads human on our scale. It is not a pass on any other tool, and no tool can promise that. Use it as an editing signal, not a certificate.
Why did my score change after a small edit?
Short texts are sensitive because a few sentences dominate the averages. Removing one stock phrase or splitting one long sentence can shift several signals at once.
Can I see the score for free?
Yes. 300 words per check, three checks a day, no sign-up. A free account includes 3,000 detector words a month; larger allowances come with plans.
Does the humanizer aim for a specific score?
It targets the signals, not the number. Check the score before and after; if it did not move, the text probably did not need humanizing.