humanflow

QuillBot's AI detector

QuillBot claims a 99% detection rate citing RAID, requires 80 words, and states that when a result is unclear its model leans towards calling text human. That last disclosure is rare in this category and it changes how both of its verdicts should be read.

Last reviewed 16 August 2026 · The HumanFlow team

What the company claims

Read from their AI detector page on 16 August 2026. A 99% detection rate, attributed to independent evaluations from the RAID benchmark. Coverage of content from ChatGPT, Gemini, Claude and Copilot. Multilingual detection. A score from 0 to 100 expressing the likelihood that text was AI-generated, with an explainer card giving the reasons behind a flag.

A detection rate is not an accuracy figure and it is certainly not a false positive rate. It answers how much machine text the tool catches. It says nothing about how often it flags a person, which is the only number that matters to somebody defending their own work — and QuillBot does not publish it. Nor does anyone else in this category, ours included.

The disclosure worth reading twice

Their documentation states that when the result is unclear, the model tends to classify texts as human-written, which reduces false positives. That is a threshold decision stated openly, and almost nothing else in this market states one.

It has a direct consequence for how the output should be read. If ambiguous cases are resolved towards human, then a human verdict from this tool is weak evidence — it includes everything the model was unsure about. An AI verdict, by the same logic, is comparatively stronger. Most readers assume the two carry equal weight, and on this tool they do not.

It is also the honest trade. Every detection threshold buys fewer false accusations at the cost of more missed machine text, and there is no setting that avoids both. QuillBot has chosen the direction that protects the person being scored, and said so.

What they say their tool gets wrong

Their page names two categories: heavily paraphrased writing, and highly formulaic content such as legal jargon and academic writing. It also notes that longer texts produce more reliable scores than short ones.

Sit with the second category for a moment. A detector vendor is stating that conventional academic prose is more likely to be misread by its own tool — which is precisely the finding of every independent study of this problem, arriving here from the company selling the detector rather than from a critic.

What a QuillBot score does not prove

That anyone used AI. QuillBot's own documentation says the tool is not perfect and that you should never rely on an AI detector alone when reviewing results. We would put it slightly more strongly — a score is a statistical judgement about how text reads, not a record of how it was produced — but the direction is the same and it is their sentence, not ours.

If a QuillBot result has been quoted at you, the useful ground is the evidence that actually helps, which is drafting history rather than a counter-score, and what a score can support procedurally.

Where we stand, since we sell against them

QuillBot competes with our humanizer, not with our detector, and we have written about their paraphrasing suite elsewhere. That is a conflict, and the mitigation is that every figure above is theirs and linked.

On the substance we have no complaint to register. A vendor publishing a minimum text length, naming the writing its tool struggles with, disclosing which way it resolves ambiguity, and telling users not to rely on it alone is behaving better than most of this category. The gap we would still press on is the false positive rate, which they do not publish — and neither do we.

Compared against

  • Scribbr vs QuillBot QuillBot against Scribbr on what independent testing found, who each is sold to, and what happens to the text you submit.
  • Copyleaks vs QuillBot QuillBot against Copyleaks on what independent testing found, who each is sold to, and what happens to the text you submit.

Figures read from QuillBot's AI detector page on 16 August 2026: quillbot.com/ai-content-detector. Vendor claims are quoted as theirs and are not independently verified here.

Common questions

How accurate is QuillBot's AI detector?
QuillBot states a 99% detection rate, citing independent evaluations from RAID. That figure is a detection rate — how much AI text it catches — and is not the same as a false positive rate, which is the number that matters if you are the one being flagged. They publish the first and not the second.
How much text does QuillBot need?
Their page states a minimum of 80 words, and separately advises that texts of 300 words or more produce more reliable scores than short inputs. Both figures are theirs. A verdict on a paragraph is worth much less than one on an essay, on this tool as on every other.
Does QuillBot lean towards calling text human?
By their own account, yes. Their documentation states that when the result is unclear, the model tends to classify text as human-written, which reduces false positives. That is an unusually direct disclosure of a threshold choice, and it means a human verdict from this tool carries less information than an AI one.
What kind of writing does QuillBot say it gets wrong?
Their own page names heavily paraphrased writing and highly formulaic content such as legal jargon and academic writing. That is a vendor stating that careful, conventional academic prose is more likely to be misread by its tool — the same population every independent study of this category has flagged.
Can a QuillBot result prove someone used AI?
No, and QuillBot says so. Their documentation states that the detector is not perfect and that you should never rely on an AI detector alone when reviewing results. That is the correct position and it is theirs, not ours.
Is QuillBot's detector free?
The AI detector is offered free on their site, and QuillBot also sells a paraphrasing suite. We compete with the paraphraser rather than the detector, which is a conflict worth knowing about when reading anything we write about them.