Pangram vs GPTZero: Our Test Data on False Positives and Paraphrase (2026)
Disclosure: Some links on this page are affiliate links from tool recommendations; we may earn a commission at no extra cost to you. Rankings come from our own testing.
Pangram vs GPTZero is a comparison almost nobody can write from experience, because one tool is new and the other is hard to test. We can write it. In August 2026 we registered a free Pangram account and ran our 12-essay ground-truth set until the daily credits ran out, and we tried GPTZero in two separate rounds. The results are the most lopsided we have published: Pangram caught the sample everyone else misses and then flagged every human essay it saw. GPTZero produced no reading at all.
Pangram vs GPTZero at a Glance
The Pangram vs GPTZero scorecard, August 2026, with every cell labeled by source.
| Pangram | GPTZero | |
|---|---|---|
| AI models covered | Major models, per vendor | 5 model families (official site) |
| Catches paraphrased text? | Yes, verified by us (100% confidence) | Claimed on paid tiers; untested this round |
| Free allowance | ~20 credits/day, 1 credit per 100 words | 10,000 words/mo (official site) |
| Paid from | ~$20/mo | ~$15/mo |
| Human essays in our test | 3/3 flagged 100% AI (false positives) | Not tested this round |
One table, two failure modes. Pangram answers every question and gets the human ones wrong. GPTZero, for us, never answered. Any Pangram vs GPTZero page that gives you a clean winner is hiding one of those two facts.
What Our August 2026 Test Showed
We registered and email-verified a free Pangram account in round 2. The daily allowance (about 20 credits at 1 credit per 100 words) covered the four highest-value samples before running dry:
- paraphrase-01 (AI essay, hand-rewritten by us): flagged 100% AI. Correct, and notable, because paraphrasing is the cheat pattern free tiers miss most.
- human-03 and human-04 (handwritten essays): both flagged 100% AI. False positives.
- esl-01 (English learner’s essay): flagged 100% AI. False positive, and the most troubling of the three.
GPTZero’s side of the Pangram vs GPTZero ledger is empty by necessity: in round 1 its homepage Scan button never fired a request across four attempts, and in round 2 free registration required hCaptcha, which we do not solve by policy. We mark GPTZero “not tested this round” rather than repeat its marketing numbers.
The ESL False Positive Problem
The esl-01 result deserves its own paragraph, because it is an equity issue before it is a product issue. AI detectors reward unpredictable, varied prose; second-language writing tends to be more formulaic, which reads as machine-made to the model. A 2023 Stanford-led study in Patterns found that seven widely used detectors, taken together, flagged more than half of TOEFL essays by non-native writers in at least one tool while rarely flagging native speakers. Our single ESL sample drew a 100% AI verdict from Pangram, the same pattern in miniature. If a detector flags a student who writes in a second language, treat the score as the weakest kind of evidence, pull the draft history, and give the student the appeal path in our false positive appeal templates before anyone says the word “cheating.”
Free Limits and Pricing
Pangram vs GPTZero on cost: Pangram’s free account worked after email verification and gave us roughly 2,000 words of scanning a day; institution pricing starts near $20 a month per vendor pricing. GPTZero’s official free plan covers 10,000 words a month across five model families, with long pastes requiring an account; paid plans start near $15 a month. For a quick no-account scan, neither is the tool we reached first in practice; that was ZeroGPT, covered in our GPTZero vs ZeroGPT notes. The full per-sample matrix lives in the test archive.
Pangram vs GPTZero: How a Teacher Should Use Either Score
The usable lesson of Pangram vs GPTZero is that false-positive behavior, not marketing accuracy, is the first number a teacher should ask about. A detector that catches paraphrases but also calls honest essays AI-generated creates a worse classroom outcome than a detector that misses some AI. So: run any flag on a second tool, read revision history before reading the score, and talk to the student before escalating. When a case needs a documented second opinion, Originality.ai exports sentence-level reports and documents paraphrase coverage per its official site (from about $14.95 a month; not hands-on tested this round), which is why it is our overall #1 pick.
Get a Second Opinion from Originality.ai
Frequently Asked Questions
Pangram vs GPTZero: which caught paraphrased AI text?
Pangram, at 100% confidence on our hand-rewritten sample, the only clean paraphrase catch of the round. GPTZero could not be tested (scan never fired; hCaptcha-gated signup).
Did Pangram false-flag human writing in your test?
Yes: all three human samples came back 100% AI, including two handwritten essays and an English learner’s essay. Never present a Pangram score to a student without a second tool and draft history.
Why do AI detectors flag ESL writing more often?
Second-language prose is often more formulaic, which resembles machine output to the model. The 2023 Stanford-led Patterns study documented the pattern across seven detectors; our ESL sample matched it.
Pangram vs GPTZero: which should a teacher trust for an integrity case?
Neither alone. Build cases on draft history, a student conversation, and a second detector; use a paid tool with exportable reports when a dispute is likely.
What are the free limits on Pangram vs GPTZero?
Pangram: about 20 credits a day (roughly 2,000 words). GPTZero: 10,000 words a month per the official site, account required for long pastes. Paid entry runs near $20 and $15 a month respectively.
Affiliate Disclosure: Some links on this page are affiliate links. If you buy through them we may earn a commission at no extra cost to you. Our rankings come from our own testing, not from commissions.