Does QuillBot bypass Pangram? No — Pangram’s own benchmark caught QuillBot-processed text 100.0% of the time, the joint-highest detection rate of twenty tools. But QuillBot never claimed to try, and that is the part everyone quoting the number leaves out.
Key Takeaways
- Pangram’s own benchmark scores QuillBot at 100.0% detection accuracy — joint-highest of the 20 tools tested, meaning Pangram caught it every time (Bradley Emi, 27 Aug 2025).
- But QuillBot’s own FAQ states plainly: “Our Humanizer is not designed for bypassing AI detectors.” It names Turnitin and GPTZero only to disclaim them.
- Pangram’s own academic paper files QuillBot under “Paraphrasers”, not “Humanizers” — a distinction its blog does not make.
- QuillBot is the only tool in that benchmark that also sells an AI detector. Competitors run “Bypass QuillBot” landing pages against it.
- The benchmark discloses no sample size and no methodology. Pangram has never published one for it.
- Pangram 4 (July 2026) now reports a pooled 98.83% across “13 popular commercial tools” it does not name — so nobody can tell whether QuillBot is still in the set.
- On Reddit through 2025–2026, QuillBot appears overwhelmingly as a detector students check against, not a tool used to beat one.
Want to bypass Turnitin in 2026? Grab the free prompt pack.
Get the exact text-humanization prompts I use to drop an AI score by hand — copy, paste, submit. Free, straight to your inbox.
Send me the free prompts →Does QuillBot bypass Pangram?
No, and unusually for this question there is a specific published number rather than a guess. Pangram Labs scored twenty humanizing and paraphrasing tools in August 2025 and put QuillBot at 100.0% detection accuracy — meaning every QuillBot-processed sample it tested was correctly identified as AI-involved.
That is the joint-highest figure in the table. It is also close to meaningless as a verdict on QuillBot, for a reason the number alone cannot show you: QuillBot is not a bypass tool and says so in writing.
This page reviews the published record and reports no first-party test. Every claim is dated and attributed, and where a source has a commercial stake, we say so.
What does the 100% actually measure?
It measures Pangram, not QuillBot. The column header in the benchmark is “Accuracy” — how often Pangram caught the tool’s output. Higher is worse for anyone hoping to slip past it.
Read the table that way and the ranking inverts from how it usually gets quoted. Undetectable AI scored lowest at 90.3%, meaning it evaded Pangram more often than any other tool tested. TwainGPT followed at 92.7%, Just Done at 93.5%, StealthGPT at 95.6%. QuillBot sits at the opposite end, tied at 100.0% with Ahrefs, Grammarly, Ghost AI, Smodin, Surfer SEO and several others.
Look at who those neighbours are and the pattern is not about capability at all.
| Tool | Pangram accuracy | How the tool is sold |
|---|---|---|
| Undetectable AI | 90.3% | Detector bypass — evaded Pangram most often of the 20 |
| TwainGPT | 92.7% | Detector bypass |
| Just Done | 93.5% | Detector bypass |
| StealthGPT | 95.6% | Detector bypass |
| Scribbr | 99.0% | Academic writing service |
| QuillBot | 100.0% | Paraphrasing tool — and sells its own AI detector |
| Grammarly | 100.0% | Writing assistant |
Why is a writing tool in a bypass benchmark?
Because Pangram’s blog uses “paraphraser” and “humanizer” interchangeably. Its own research does not. The DAMAGE paper — written by Pangram’s own team — splits its 19 tools into two groups, filing QuillBot under “Paraphrasers” with DIPPER and Grammarly, separate from the list it labels “Humanizers.”
The distinction matters commercially. Tools in the second group sell detector evasion as the product. QuillBot sells rewriting, and markets an AI detector of its own alongside it — the only tool in the benchmark that does. Competing humanizers run “Bypass QuillBot AI Detector” landing pages, which tells you which side of the market QuillBot is understood to be on.
So the honest reading of 100.0% is not “QuillBot failed.” It is that a tool which never entered the race was scored on it anyway. Our full QuillBot review covers what it does do.
How much should you trust the benchmark?
Directionally, quite a lot. Precisely, less than the decimal point implies.
Pangram Labs is a commercial AI detector benchmarking the tools it competes with, which is a conflict worth naming even when the work looks careful. The post states no sample size, no source texts, no per-tool method and no false-positive rate. A figure of “100.0%” carries an implied precision that an undisclosed sample cannot support.
There is also a freshness problem, and it runs the wrong way. The benchmark measures a model two generations old. When Pangram 4 launched in July 2026 it reported detecting AI involvement in humanized output 98.83% of the time across “13 popular commercial tools” — pooled, with the tools unnamed. So Pangram’s per-tool transparency went down between August 2025 and July 2026, and whether QuillBot remains in the tested set is simply not knowable from outside.
We track that pattern across detectors in how much AI detectors disagree and in the Pangram review.
Has anyone else tested QuillBot against Pangram?
No. As of August 2026 we could find no independent, third-party test of QuillBot against Pangram from anyone — academic, journalistic or commercial. Pangram’s own benchmark is the entire public record.
What exists instead are rival detectors making adjacent claims. Originality.ai says on its own blog that it identifies QuillBot-paraphrased writing with 96% accuracy, citing the third-party RAID benchmark rather than its own testing. GPTZero’s blog quotes a secondhand figure putting QuillBot’s detector at 80% accuracy. Both publishers sell competing detectors, and neither ran QuillBot against Pangram.
The nearest thing to a neutral read is an analyst comparison from May 2026 that tested nine human and 27 AI texts, and found QuillBot’s detector missed every Gemini-generated sample. That measures QuillBot detecting, not QuillBot evading — which is the direction most published QuillBot research runs in.
What do users actually say?
The organic record is striking mostly for what is absent. Across 2025 and 2026, QuillBot shows up on Reddit overwhelmingly as a detector people run their own essays through — students posting that QuillBot flagged work they wrote themselves at 28% or 31% AI, or comparing its score against ZeroGPT and Grammarly.
It barely appears as a tool for beating detection. In one thread a humanizer vendor describes his product as “designed to bypass QuillBot and GPTZero” — QuillBot as the obstacle, not the method. Even inside a subreddit dedicated to bypassing AI detection, QuillBot gets named alongside Grammarly as a writing tool.
The single place we found QuillBot positioned as a bypass tool was a promotional listicle in a vendor-run subreddit, ranking it third behind two tools the poster was selling. That is the exception that confirms the pattern.
What should you do instead?
If your worry is a false AI flag on writing you actually wrote — which is what most of those Reddit posts are really about — a rewriting tool is the wrong instrument. It changes the text without changing the evidence, and it is exactly the use QuillBot’s own FAQ warns against.
Keep your drafting history. Version history in Google Docs or Word costs nothing and is the strongest defence available, and it matters most for the writers detectors treat worst — see ESL writers and AI detection and false positives for neurodivergent students. Check before you submit rather than after, using the best pre-submission check and the Turnitin self-check walkthrough.
And if you are researching the Pangram question specifically, the tools that do market bypass have their own dedicated answers: Undetectable AI, StealthGPT, GPTHuman and WriteHuman, plus the overview at bypassing the Pangram AI detector. For the Turnitin side of the same question, start with does Undetectable AI bypass Turnitin and what to remove before using an AI humanizer.
Get the free prompt pack instead
The manual rewriting prompts that lower AI signals without a subscription — and without betting a submission on a number nobody has reproduced. Free, no card.
Frequently asked questions
Does QuillBot bypass Pangram?
No. In Pangram’s own benchmark, QuillBot-processed text was detected 100.0% of the time — the joint-highest detection rate of the 20 tools scored. But the more useful answer is that QuillBot never claimed to try: its humanizer FAQ explicitly states the tool is not designed for bypassing AI detectors.
What does the 100.0% figure actually mean?
It is Pangram’s accuracy at detecting AI text that had been run through QuillBot — not QuillBot’s success rate. Higher is worse for anyone hoping to evade detection. Undetectable AI scored lowest in the same table at 90.3%, meaning it evaded Pangram most often of the twenty.
Does QuillBot claim to beat AI detectors?
The opposite. Its AI Humanizer page carries a FAQ headed “Will this tool help my writing bypass AI detectors like Turnitin or GPTZero?” and answers: “Our Humanizer is not designed for bypassing AI detectors.” The same page tells users to follow their institution’s AI guidelines and to cite AI assistance.
Why is QuillBot in a humanizer benchmark at all?
Because Pangram’s blog treats paraphrasers and humanizers interchangeably. Its own peer-reviewed-style paper, DAMAGE, does not — it classifies QuillBot under “Paraphrasers” alongside DIPPER and Grammarly, separately from the tools listed as “Humanizers.”
How reliable is the benchmark?
The figure comes from Pangram Labs, a commercial AI-detector vendor scoring the tools its product competes against. The post discloses no sample size, no source texts, no per-tool methodology and no false-positive rate. Treat the direction as informative and the precision as unverified.
Is the August 2025 number still current?
It is still Pangram’s newest published QuillBot figure, but it describes a model two generations old. Pangram 4 shipped in July 2026 reporting 98.83% detection across “13 popular commercial tools” — and does not name them, so whether QuillBot is among them is not verifiable.
What does QuillBot cost?
The free tier humanizes up to 125 words with 6 uses a day. Premium is $19.95/month, $39.95 quarterly, or $99.95 annually, which works out at $8.33/month. Monthly pricing is under live A/B testing, so you may be shown a different figure.
If I want to avoid a false AI flag, does QuillBot help?
It may reduce surface-level AI patterns, which is what it says it does. It will not make text undetectable to Pangram, and using it that way is the misuse its own FAQ warns against. Keeping drafting history is a stronger defence against a false accusation than any rewriting tool.
