Pangram vs GPTZero: which AI detector holds up better?
Pangram vs GPTZero comes down to evidence against convenience. Pangram has the stronger record in independent studies, most of all on AI text that has been reworded to slip past detectors. GPTZero is the better-known name and lets you try a text without an account. It also costs less if you pay for a year up front.
Our verdict: if catching AI text is the priority, Pangram has the better evidence. Every independent study we found that tested both tools had Pangram catching more AI text, and the gap grew on humanized text. GPTZero makes sense if you want a no-account trial, writing replay or the cheaper annual plan.
Pangram vs GPTZero at a glance
Prices and limits are as listed on each company's site at the time of writing (October 2026), in US dollars. IsThisAiWritten, our own free detector, is in the first column for comparison.
| IsThisAiWritten | Pangram | GPTZero | |
|---|---|---|---|
| Use without an account | Yes, any length | No, its help center says you must be logged in | Up to 10,000 characters per scan |
| Free allowance | Unlimited | 20 credits a day, about 2,000 words | 10,000 words a month with a free account |
| Entry paid plan | Free | $20 a month, or $15 billed yearly, 300,000 words | $23.99 a month, or $12.99 billed yearly, 300,000 words |
| Length per scan | No limit | 50 to 18,725 words in the dashboard | 50,000 characters (Premium), 150,000 (Professional) |
| Keeps your text | Never uploaded; checked in your browser | While your account is active; you can delete scans | Dashboard inputs stored; API inputs not stored |
How the two compare on features:
| Pangram | GPTZero | |
|---|---|---|
| Languages | More than 20 | Five fully supported |
| Mixed or AI-edited text | Labels segments as human, AI-assisted or AI | Mixed detection and sentence highlights on paid plans |
| Writing replay | Not listed | Yes, for Google Docs |
What independent studies say about Pangram vs GPTZero
Both companies publish tests that put themselves first. GPTZero's own Pangram comparison reports 99.6% accuracy for GPTZero against 97.5% for Pangram. Vendor tests are not neutral, so we looked for outside research that ran both tools on the same texts.
University of Chicago Becker Friedman Institute, August 2025
Economists Brian Jabarian and Alex Imas tested Pangram, GPTZero, Originality.ai and an open RoBERTa model on 1,992 human passages in six genres, from product reviews to novel excerpts, matched with AI passages from GPT-4.1, Claude Opus 4, Claude Sonnet 4 and Gemini 2.0 Flash. Pangram's false positive rate was essentially zero at its standard thresholds. GPTZero's stayed around 0.7% at every threshold.
The real split came with humanizers. After the AI passages went through StealthGPT, GPTZero missed around half or more of them in most genres. Pangram still caught nearly all of the longer ones. The authors also priced each tool per correctly flagged AI passage: $0.0228 for Pangram and $0.0575 for GPTZero.
University of Maryland, 2025
Jenna Russell, Marzena Karpinska and Mohit Iyyer compared expert human readers with detectors on 300 non-fiction articles. Overall, Pangram flagged 98.0% of the AI articles with a 2% false positive rate; GPTZero flagged 85.3% with 0.7%. So GPTZero made fewer false accusations here, but missed far more. On articles written by o1-pro and then humanized, Pangram caught 90.0% and GPTZero 46.7%. Each condition had only 30 articles.
Vrije Universiteit Brussel, June 2026
This peer-reviewed study in the International Journal for Educational Integrity ran 160 academic papers in four groups of 40 through Pangram, GPTZero, Copyleaks and Turnitin. For hybrid papers (human text with AI passages added), Pangram's score landed in the correct range for 37 of 40; GPTZero managed none. For humanized papers it was 37 against 1. Every tool cleared the fully human papers, though the authors noted a small tendency for GPTZero to over-flag.
RAID benchmark
RAID is a public benchmark from Liam Dugan and colleagues, presented at ACL 2024. Teams run their own detector on the test set and the maintainers score the predictions. GPTZero's entry catches 95.7% of AI texts when tuned to flag only 1% of human texts, and 92.9% once adversarial attacks are included. Pangram has no entry, so RAID cannot settle this matchup. Its AI texts also come from older models such as GPT-4 and Llama chat.
What Pangram and GPTZero claim about themselves
Pangram says it targets a false positive rate of 1 in 10,000. An October 2026 post on its blog reports 0.0041% across more than two million human-written documents, and says the model is tuned to prefer missing AI text over accusing a human. Its homepage reports 99.7% accuracy on samples from 26 AI models.
GPTZero's homepage claims 99% accuracy. It also says it retrains its model to keep the false positive rate on writing by English learners at 1%. Both sets of numbers come from the vendors' own testing.
Pricing and free limits
GPTZero is the quicker way to try a single text: paste up to 10,000 characters on the homepage, no account needed. A free account then gives 10,000 words a month. Pangram asks you to sign up first, but its free plan is more generous over a month: 20 credits a day at 100 words per credit, so about 2,000 words a day.
On paid plans, Pangram Individual is $20 a month ($15 billed yearly) and GPTZero Premium is $23.99 a month ($12.99 billed yearly), both for 300,000 words. Higher up, Pangram Professional is $65 a month ($45 yearly) for about 1.5 million words and includes $200 of API credit. GPTZero Professional is $45.99 a month ($24.99 yearly) for 500,000 words, with LMS integration and batch uploads of up to 250 files.
Features you will use day to day
Pangram reports how much of a document looks AI-written and marks passages that look AI-edited rather than fully generated. Paid plans add a plagiarism check. It accepts PDF, DOCX and RTF uploads, keeps a history of past scans, and connects to Canvas, Moodle, Google Classroom and Brightspace.
GPTZero highlights sentences, separates mixed human and AI text, and offers Writing Replay, which plays back how a Google Doc was typed, pastes included. Teachers can run it from its Chrome extension, Google Classroom or Canvas. GPTZero is now owned by Superhuman, the company formerly called Grammarly, after a deal announced in June 2026.
Privacy and where your text goes
Both tools process text on their servers. Pangram's privacy FAQ says content is stored for as long as your account is active, can be deleted from your history, is encrypted, and is never used to train its model. GPTZero's FAQ says it stores inputs made from the dashboard and uses them in aggregate to improve the service, while text sent through its API is not stored. Both say they are SOC 2 Type 2 compliant.
Who should pick which
- Pick Pangram if you check a lot of text, worry about humanizer tools, or work in a language other than English.
- Pick GPTZero if you teach and want writing replay, or if you only need to check the odd essay without making an account.
- Pick neither alone for a decision that could hurt someone. Both vendors say a score should not be the only evidence.
A free option before you pay: IsThisAiWritten
Both tools meter their free use, by the day or by the month. IsThisAiWritten doesn't: there is no account and no word limit, and the check runs in your browser, so the text is not uploaded. Every word is scanned, never sampled, so a whole manuscript or thesis can be checked in one piece without spending credits. Like any detector, it can be wrong in both directions, so treat a result as a reason to look closer, not as proof. The methodology page explains how it is tested.
Where Pangram and GPTZero are better than us: they handle more languages (we are English only), read uploaded files and plug into school systems such as Canvas and Google Classroom. Pangram adds a plagiarism check on paid plans and a history of past scans; GPTZero adds Writing Replay and a Chrome extension. If you need any of those, use one of them.
Side by side: IsThisAiWritten vs Pangram and IsThisAiWritten vs GPTZero.
Sources
- Jabarian and Imas, Artificial Writing and Automated Detection (BFI Working Paper 2025-116), checked October 2026
- Russell, Karpinska and Iyyer, University of Maryland study of human and automatic AI-text detection (2025), checked October 2026
- Van Vlasselaer, Van Droogenbroeck and Spruyt, Who wrote this? Evaluating the reliability of AI detection tools in higher education, checked October 2026
- RAID benchmark leaderboard, checked October 2026
- RAID repository and leaderboard submission rules, checked October 2026
- Pangram homepage, checked October 2026
- Pangram: What do I get with a free account?, checked October 2026
- Pangram: How to scan a text for AI, checked October 2026
- Pangram: Minimum and maximum input sizes, checked October 2026
- Pangram: Subscription tiers, checked October 2026
- Pangram pricing, checked October 2026
- Pangram data privacy FAQ, checked October 2026
- Pangram: Is Pangram accurate?, checked October 2026
- GPTZero homepage, checked October 2026
- GPTZero pricing, checked October 2026
- GPTZero FAQ, checked October 2026
- GPTZero privacy and security, checked October 2026
- GPTZero: GPTZero vs Pangram, checked October 2026
- GPTZero: accuracy statistics, checked October 2026
- BetaKit: Superhuman, formerly Grammarly, acquires AI detector GPTZero, checked October 2026
Is this AI written? Check it now.
Free, no sign-up, no word limit. The check runs in your browser, so your text never leaves your device.
Check text freeIsThisAiWritten is not affiliated with, endorsed by or sponsored by any company named on this page. Product names belong to their owners. Facts about other products come from their public pages on the date shown and may have changed since.