QuillBot AI Detector: What It Checks, What It Reports and How Accurate It Is

The QuillBot AI detector is a free checker from the company best known for its paraphrasing tool. You paste at least 80 words, and it returns a score from 0 to 100% for how likely the text is to be AI-generated, with sentence-level highlights. QuillBot cites a 99% detection rate on a public benchmark. Two published studies that name the tool found a mixed record: one saw no false positives on 49 human answers, another flagged 10 of 30 human essays as AI. This page sets out what the tool reports, what those numbers mean, and how to read a result fairly.

What the QuillBot AI detector is and what it shows

QuillBot's AI detector page, read on November 2, 2026, describes a tool that looks for "repetitive or predictable phrases" and other patterns and "gives a confidence score between 0 and 100% that states the likelihood of the text being AI or human written." Its FAQ says the tool also distinguishes how a text was made: "entirely AI-written, entirely human-written, or human-written with some AI refinement." Flagged sections come with explainer cards that say why they were marked.

The page sets practical limits. A scan needs at least 80 words, and QuillBot says "longer texts (300+ words) produce more reliable scores than short inputs." The detector reads text only; images go to a separate image detector. It says the tool supports "20+ languages," naming English, Spanish, French and Portuguese among them.

The same page is plain about what a score is not. "Scores are signals, not verdicts," it says, and it tells readers never to make "an academic or employment decision on a detector score alone." It also says the tool "can not verify whether a piece of writing was created by a human or original," and that "no AI Detector can provide 100% accuracy."

Is the QuillBot AI detector accurate?

A 99% catch rate measured where 1 in 20 human texts is flagged is not a 1% error rate.

QuillBot's page says the detector finds AI text "with a 99% detection rate, according to independent evaluations from RAID." RAID is a benchmark built by researchers at the University of Pennsylvania and others and presented at the ACL 2024 conference. Its public leaderboard scores detectors on a hidden test set of machine-written and human-written texts across eight domains, from news and abstracts to recipes and poetry.

Two details matter when you read that figure. First, the detector's maker runs its tool on the test texts and submits the predictions; RAID scores them. QuillBot's RAID results file lists a QuillBot contact as the submitter and a release date of January 30, 2025. So the scoring is third party, but the entry is not a study someone else chose to run.

Second, the leaderboard measures how much machine text a detector catches after its threshold is set, domain by domain, so that exactly 5% of the human texts are flagged. On that basis QuillBot caught 54,248 of 54,400 machine-written texts without adversarial edits (99.7%), and 650,303 of 652,800 when texts with attacks such as paraphrasing and misspellings were included (99.6%). At a stricter 1% false positive rate it caught 99.2% and 98.1%. Those are strong results, but in a setting where 1 in 20 human texts is flagged, on thresholds tuned to the benchmark. The public tool uses its own threshold, and QuillBot says that "when the result is unclear, our model tends to classify texts as human written."

What published studies found

Few peer-reviewed studies name QuillBot's detector; the two below are preprints posted on arXiv. Yeadon and colleagues (2024) ran 60 short physics essays through five detectors: 30 written by students before ChatGPT existed and 30 generated by GPT-4. With results reduced to AI or human, QuillBot flagged all 30 AI essays and 10 of the 30 student essays, for an accuracy of 83% and a precision of 75%. In the same test ZeroGPT made no false positives.

Alshammari and Rao (2025), in a study conducted in March 2025, used 49 human-written question answers from before the chatbot era, 49 answers generated by the DeepSeek model and 196 altered versions. QuillBot had an average accuracy score of 100% on the human answers, so no false positives, and 98.4% on the unaltered DeepSeek answers. After the answers were run through QuillBot's own paraphraser it scored 95.49%. On DeepSeek text rewritten in a "humanize" mode, its score fell to 58%.

The spread is the lesson. Ten false flags in 30 essays and none in 49 answers are both real results; they came from different texts, different years and different versions of the tool. Other detectors show the same spread, as the evidence on GPTZero's accuracy shows.

How to read a QuillBot result fairly

1. Check the length first

   Under 300 words, QuillBot itself says the score is less reliable.

2. Separate the three labels

   "AI-refined" means the tool thinks a human draft was polished with AI help. That may be allowed under your course rules; it is not the same as AI-written.

3. Read the highlighted sentences

   The score rests on the passages it marks. Read them yourself: are they generic, or simply formal?

4. Consider the kind of writing

   QuillBot warns that "formulaic human writing (academic definitions, legal copy, structured templates)" can score higher. Lab reports and definitions fall in this group.

5. Ask for the process

   Drafts, notes and version history answer what a percentage cannot. Use them before any decision.

A percentage is a likelihood, not a share of the text a tool wrote. What an AI detection score actually means explains the difference, and the guide to what AI text detectors can tell you covers how these tools work in general.

A worked example: one paragraph, two readings

Here is an example paragraph written for this page. A student drafted it, then used a grammar tool to fix two commas.

• Example sentence 1: "Photosynthesis is the process by which plants convert light energy into chemical energy stored in glucose."
• Example sentence 2: "It takes place in the chloroplasts, which contain the pigment chlorophyll."
• Example sentence 3: "When I grew basil on my windowsill last spring, the plant leaned toward the glass within a week."

The first two sentences are textbook definitions, the formulaic writing QuillBot says can score higher. The third is specific and personal. A flag on the first two would say more about the genre than the writer. The paragraph is also under 80 words, so QuillBot would not scan it alone.

Checking with a second detector

Writers checking their own drafts sometimes compare two tools to see whether a flag holds up. You can run the same text through a second AI detector such as Walter Writes AI, an AI humanizer with a built-in AI detector that supports over 80 languages; every plan includes voice matching and the detector. The same limit applies as to QuillBot: no detector, Walter's included, can guarantee 100% accuracy, and two detectors can disagree about the same paragraph.

Teachers comparing tools for a class can start with the guide to choosing an AI detector as a teacher. If you need to show your own process, how to show you wrote your own work lists the records that help most.

Pitfalls

• Quoting 99% as an error rate. The RAID figure is a catch rate at a fixed 5% false positive rate, not the chance a flag is wrong.
• Scanning a fragment. Under 80 words the tool will not run, and under 300 its own page says scores are less reliable.
• Treating AI-refined as cheating. Grammar help or light editing may be permitted. Check the rule before the score.

Expert tips

• Scan the whole text at once. QuillBot recommends analyzing the entire piece rather than pasting sections separately.

The QuillBot AI detector is free, quick and more candid about its limits than many rivals. Its benchmark result is strong, but it was measured at a fixed false positive rate on tuned thresholds, and small independent studies found errors in both directions. Use its highlights to decide what to reread, and let drafts and notes carry the weight a score cannot. More guides to checking and editing AI writing are on the AI writing hub.