Is QuillBot's AI detector accurate? What QuillBot, Turnitin and studies say

How does QuillBot's AI detector work?

It compares your text with patterns from human and AI writing and scores it sentence by sentence. QuillBot says the detector “analyzes text features to determine whether they are more similar to human-written or AI-generated samples”. The signals it names include how predictable the text is, variation in sentence structure, and repetitiveness.

You get a score from 0 to 100% for how likely the text is AI-generated, plus highlighted sentences. QuillBot's results can also sort text into four categories:

• Generated by AI
• Generated and refined by AI
• Written by a human and refined by AI
• Written by a human

QuillBot says the “AI refined” labels cover text changed “through the use of a paraphrasing tool”.

Documented limits

• Length: QuillBot needs at least 80 words to scan and says texts of 300+ words give more reliable scores.
• Unclear results: when a result is unclear, the model “tends to classify texts as human written”, which QuillBot says reduces false positives.
• Free use: the detector page lists the free plan as up to 1,200 words per scan and up to 6 scans a day; Premium has no word limit and unlimited scans. A QuillBot blog post says free scans of fewer than 1,200 words need no sign-up.
• Languages: the detector page says it supports 20+ languages. A QuillBot blog post revised in July 2026 lists six: English, French, Spanish, German, Dutch and Portuguese. Check that yours is covered.
• Proof of authorship: QuillBot says the tool “can not verify whether a piece of writing was created by a human or original”.

This article covers only the detector. For QuillBot's humanizer, see our QuillBot humanizer comparison.

How accurate does QuillBot say its detector is?

QuillBot claims a 99% detection rate, and it also says no detector is perfect. Its detector page says it detects AI text “with a 99% detection rate, according to independent evaluations from RAID”. The same page says “no AI Detector can provide 100% accuracy” and “Results are probability estimates, not verdicts.”

RAID is a public benchmark described in a 2024 research paper. Detector makers submit predictions on a test set whose answers are not published, and RAID scores them at a set false positive rate, 5% by default in its evaluation code. Its leaderboard includes a QuillBot entry. But RAID's test text comes from news, books, abstracts, reviews, Reddit, recipes, Wikipedia and poetry, not student essays. And in the RAID paper, which evaluated 12 detectors, the authors found that current detectors “are easily fooled by adversarial attacks” and by unseen AI models. QuillBot was not one of those 12; its leaderboard entry was added later. A benchmark score describes the benchmark, not your essay.

QuillBot is open about false positives. It says “All AI checkers present the risk of false positives”, and that heavily paraphrased writing and highly formulaic content, such as legal jargon or academic writing, is more likely to be flagged. Its advice is direct: “Never make an academic or employment decision on a detector score alone.”

What do independent studies say about QuillBot?

We found two studies that test QuillBot directly. Both were posted on arXiv, used small samples and tested older versions of the tool, and they disagree on false positives. Treat them as snapshots.

• Physics essays (2024). Yeadon and colleagues ran 60 submissions through five detectors. Half were student work from before ChatGPT and half were written by GPT-4, and each held five short essays. QuillBot flagged all 30 AI submissions but also labeled 10 of the 30 human submissions as AI, for 83% accuracy and 75% precision. ZeroGPT had no false positives in the same test.
• DeepSeek text (2025). Alshammari and Rao tested six detectors on 49 human-written answers and matching DeepSeek-v3 answers. QuillBot scored 100% on the human-written text, with no false positives. After the DeepSeek answers were run through QuillBot's own paraphrasing tool, QuillBot's detector still scored them 95.49% on average (the study's “accuracy” is the mean AI score across 49 samples). A separate set of answers paraphrased in a “humanize” mode was the hardest case: scores fell to 58% for QuillBot, 71% for Copyleaks and 52% for GPTZero.

Wider research is cautious about detectors in general. A 2023 study of 14 tools, including Turnitin, concluded they were “neither accurate nor reliable” and that obfuscation techniques significantly worsen their performance. Another 2023 study found detectors “consistently misclassify non-native English writing samples as AI-generated”. QuillBot itself lists non-native English writing styles among content that may need extra context. For the mechanics behind these errors, see how AI detectors work.

Does QuillBot paraphrasing show up as AI?

It can, and QuillBot says its Paraphraser is not meant to hide AI use. Its help center says the Paraphraser is “not designed to alter text in a way that avoids detection” and that AI detection tools are not 100% accurate.

What you see depends on what you paraphrased. If you wrote the text and only reworded parts, QuillBot says its checker accounts for AI-refined text to reduce false positives on human writing edited with AI. If the original was AI-generated, paraphrasing may not remove the signal: in the small 2025 study above (49 samples), QuillBot's detector still gave DeepSeek text paraphrased with QuillBot's tool an average AI score of 95.49%.

Research also shows the outcome depends on the paraphraser and the detector. One 2023 study built its own paraphrasing model and cut the detection accuracy of a research detector, DetectGPT, from 70.3% to 4.6%, while the physics-essay study reported, without figures, that the paraphrasing tools it tried appeared to have “minimal impact on detection rates”.

If your course limits AI tools, a paraphraser can count as AI use even when the ideas are yours. QuillBot's advice: if you used AI, “disclose it rather than revising to change a score”. Our responsible AI use guide covers disclosure.

Does QuillBot show up as AI on Turnitin?

It can: Turnitin names QuillBot in its own documentation. Turnitin's guide to the AI Writing Report says its detector “detects when likely AI generated text may have been further modified by an AI bypasser, AI-paraphrasing tool or AI word spinner, such as Quillbot”. That text counts toward the same AI percentage and is highlighted in blue.

Some limits apply. Paraphrase and bypasser detection works only for English submissions. Turnitin won't list the paraphrasing tools it trained on, saying it is “unable to disclose the names of these tools”. It also warns that its model “may misidentify human-written, AI-generated, and AI-paraphrased text”, and its FAQ lists “text that has been paraphrased without developing new ideas” among content that can produce false positives.

For more, see our guides to Turnitin's AI detection and Grammarly and AI detectors.

Is QuillBot's AI detector as accurate as Turnitin?

There is no fair way to rank them from the published evidence. The studies we read do not run both on the same essays, and each vendor reports a different kind of figure. Turnitin says it aims to keep false positives “under 1% for documents with over 20% of AI writing”. QuillBot cites its RAID detection rate. One is a false positive target; the other is a benchmark detection rate.

If you don't know which tool checks your work, our guide to which AI detector schools use explains how to find out.

Why do QuillBot and Turnitin give different scores?

They are different models with different rules:

• Different training. Turnitin says its model is trained on AI-generated and authentic academic writing. QuillBot describes “LLM-based linguistic pattern analysis” trained on large datasets.
• Different thresholds. Turnitin shows an asterisk instead of a number for scores from 1% to 19%, because false positives are more common there. QuillBot reports the full 0 to 100% range and leans toward “human written” when unsure.
• Different text counted. Turnitin scores only prose sentences in long-form writing, from 300 words, and does not detect bullet points or tables. QuillBot scans from 80 words.
• Different updates. QuillBot says its models are “regularly retrained and refined”, and Turnitin says model changes are likely to end up “affecting the AI percentage”. Text checked months apart can score differently.

So a 0% from QuillBot does not mean Turnitin will show 0%, and a high QuillBot score does not mean Turnitin will flag you. If your course uses Turnitin, its report is the one your instructor sees.

How should you use a QuillBot AI score?

Treat it as a reason to look closer, not a verdict. Both vendors say so: QuillBot calls scores “signals, not verdicts”, and Turnitin says deciding on misconduct takes “further scrutiny and human judgment”.

1. Scan the whole text at once. QuillBot says 300+ words give more reliable scores.
2. Read the flagged sentences, not only the percentage. QuillBot says formulaic writing can score higher than expected.
3. Don't treat a low score as clearance. Detectors produce false negatives as well as false positives.
4. Keep your drafts and version history. If you are flagged, our guide to AI detector false positives explains how to respond.
5. Follow your course policy and disclose AI help where it is required.
6. Teachers: QuillBot says to “pair detector results with a conversation or further investigation”. Check school rules before uploading student work: Vanderbilt University raised “real privacy concerns” about outside detectors.

Common questions

Sources

Checked on November 10, 2026. Vendor pages describe each vendor's own product. Studies tested the tool versions available at the time.

• QuillBot: AI Detector page (Internet Archive copy, October 25, 2026)
• QuillBot: Quillbot's AI Detector: The Best AI Detector (revised July 13, 2026)
• QuillBot: How to Use an AI Detector & When You Should
• QuillBot: How Do AI Detectors Work?
• QuillBot Help Center: I paraphrased my content and tried the AI Detector, but it was still flagged as AI. Why?
• Turnitin: Using the AI Writing Report
• Turnitin: AI writing detection capabilities FAQs
• Dugan et al.: RAID: A Shared Benchmark for Robust Evaluation of Machine-Generated Text Detectors (2024)
• RAID benchmark: Dataset, evaluation and leaderboard documentation
• RAID benchmark: QuillBot leaderboard submission
• Yeadon et al.: Evaluating AI and Human Authorship Quality in Academic Writing through Physics Essays (preprint, 2024)
• Alshammari and Rao: Evaluating the Performance of AI Text Detectors, Few-Shot and Chain-of-Thought Prompting Using DeepSeek Generated Text (preprint, 2025)
• Weber-Wulff et al.: Testing of detection tools for AI-generated text, International Journal for Educational Integrity (2023)
• Krishna et al.: Paraphrasing evades detectors of AI-generated text, but retrieval is an effective defense (2023)
• Liang et al.: GPT detectors are biased against non-native English writers, Patterns (2023)
• Vanderbilt University: Guidance on AI Detection and Why We're Disabling Turnitin's AI Detector (2023)