Guide

GPTZero AI Detector: We Tested It on 11 Texts We Knew the Origin Of (2026)

Published on November 13, 2026·9 min·By Eliot Byteau
GPTZero AI Detector: We Tested It on 11 Texts We Knew the Origin Of (2026)

The tool tested in this guide

Try GPTZero on your own writing and check our findings for yourself.

Try GPTZero →

Disclosure: DigitSelect is a member of GPTZero's affiliate program. If you subscribe through a link on this site, we may earn a commission at no extra cost to you. GPTZero did not sponsor or review this article, and we ran every test on a free account without paying.

An AI detector is only useful if you know how often it gets things wrong. So we gave GPTZero's free AI detector eleven texts whose origin we knew for certain: six written by people before ChatGPT existed, four generated by an AI model, and one that mixes both. Here is what it got right, what the free plan actually includes, and what a GPTZero score can and cannot tell you.

How we tested

We created a free GPTZero account on November 13, 2026, skipped the Premium trial, and pasted each text into a new document in the GPTZero dashboard. The result panel showed the model as "GPTZero AI Detection, Model 5.0b". We used the same eleven texts as in our Paperpal AI detector test:

#TextOriginWords
H1Abstract, Ruscetti et al. (2018), PLOS ONEHuman, published in 2018204
H2Abstract, Rosário et al. (2019), PLOS ONEHuman, published in 2019225
H3Abstract, Ruhl et al. (2019), PLOS ONEHuman, published in 2019241
H4Abstract, Nam & Park (2020), PLOS ONEHuman, published in 2020210
H5Abstract, Gonzalez et al. (2020), PLOS ONEHuman, published in 2020209
H6John Dewey, Democracy and Education (1916), chapter 1Human, published in 1916257
A1–A4Four paragraphs on academic topics: an essay, a discussion, a first-person reflection and a results section with statisticsGenerated by an AI model (Claude), unedited158–185
M1The first three sentences of H2 followed by three AI-generated sentencesMixed: 75 human words, 67 AI words142

All five abstracts were published under a CC BY license before ChatGPT was released in November 2022, so no generative AI wrote them. Dewey's book is in the public domain. The first text used our one free Advanced scan; the ten others ran on the Basic scan, with the same model version. We ran each text once and repeated one to check consistency. This is a small, dated test, not a benchmark.

Results at a glance

#OriginGPTZero's resultGPTZero's verdictCorrect?
H1Human (2018)Human 100%"Highly confident this text is entirely human"Yes
H2Human (2019)Human 100%SameYes
H3Human (2019)Human 100%SameYes
H4Human (2020)Human 100%SameYes
H5Human (2020)Human 100%SameYes
H6Human (1916)Human 100%SameYes
A1AIAI 100%"Highly confident this text was AI generated"Yes
A2AIAI 100%SameYes
A3AI, first personAI 100%SameYes
A4AI, with statisticsAI 100%SameYes
M13 human + 3 AI sentencesMixed 100%"Highly confident this text is a mix of AI and human", 3 of 6 sentences likely AIYes

Human writing: no false alarms

The texts most likely to trip a detector are short, formal research abstracts: concise, structured, full of the phrasing students are taught to use. GPTZero rated all five of ours "entirely human", with 0% AI and 0% mixed, and did the same for the 1916 excerpt. Formal academic writing is where detectors most often raise false alarms, so this was the most useful result of the test.

GPTZero AI detector: a 2019 research abstract rated 100% human

We scanned the 2019 science policy abstract a second time a few minutes later: same result, 100% human.

AI writing: all four caught

All four AI-generated paragraphs scored 100% AI with a "highly confident" verdict. That includes the hardest one for a detector, a first-person reflection about group projects that reads like a student assignment, and a results paragraph full of percentages and a chi-square test.

GPTZero AI detector: an AI-generated first-person reflection rated 100% AI

Keep in mind what we did not test: our AI texts were pasted exactly as generated. We did not test AI text that a person had edited or rewritten, which is harder to detect.

Mixed writing: the right sentences

Our mixed paragraph started with three sentences from a 2019 abstract and ended with three sentences written by the AI. GPTZero called it "a mix of AI and human" and said 3 of 6 sentences were likely AI generated. The three sentences it highlighted in the editor were exactly the three AI sentences; none of the human sentences was highlighted.

GPTZero AI detector: a half-human, half-AI paragraph, with the three AI sentences highlighted

One detail to read carefully: the result showed "Mixed 100%". That number is GPTZero's confidence that the whole text is mixed, not the share of AI in it. The "3/6 sentences" line and the highlighting are what tell you how much of the text was flagged.

Is GPTZero free?

Yes, with limits. As of November 13, 2026, the free plan includes 10,000 words a month, the Basic AI scan, one free Advanced scan, and five free AI highlights in the Chrome extension. Each scan needs at least 250 characters, and our credits went down by roughly one per word scanned.

After our first scan, the dashboard showed "0 advanced scans left". Every following scan used the Basic scan, which still gave a verdict, the AI/mixed/human breakdown and, on the mixed text, sentence highlighting. Expect several upgrade pop-ups along the way; they close without affecting the result.

GPTZero pricing as of November 13, 2026: Free, Premium and Professional

PlanWords per monthMain additions
Free10,000Basic scan, 1 Advanced scan
Premium300,000Advanced scan, multilingual detection, downloadable reports, plagiarism check
Professional500,000Larger files and batches, page-by-page scanning, LMS integration

Paid plans are cheaper when billed annually, and GPTZero offered us a free Premium trial during sign-up. Prices and offers vary by country and change often; check the GPTZero pricing page for the current amounts.

How to read a GPTZero score

  1. Look at the verdict and the breakdown together. "Human 100%" means GPTZero is confident the whole text is human, not that 100% of the words were checked by a person.
  2. For mixed texts, read the sentence count and the highlights. They show which parts the model reacted to.
  3. A clean result on our texts is not a guarantee on yours. Edited AI text, very short texts and other languages were not part of this test.
  4. Keep evidence of how you wrote. Drafts, notes, outlines and your document's version history are what show authorship, not a detector score.

If your own work is flagged by your university, talk to your instructor and show your drafts. If you used an AI tool for part of the work, check your institution's rules and disclose it.

Limits of this test

  • Eleven short texts (142 to 257 words) are a small sample. A perfect score here does not mean GPTZero never makes mistakes.
  • Our human texts are published abstracts and a classic book, which are publicly available online. We have no indication that this affected the results, but we cannot rule it out.
  • Our AI texts were generated by a single model and left unedited.
  • We tested on one day, with model version 5.0b. Detectors are updated often, and results can change.

FAQ

Is GPTZero accurate?

In our test, yes: it classified all 11 texts correctly, six human texts written before ChatGPT, four AI-generated texts and one mixed paragraph. It also identified the exact AI sentences in the mixed paragraph. Eleven texts is a small sample, so treat this as one data point, not proof.

Does GPTZero flag human writing as AI?

Not in our test. Five formal research abstracts and a 1916 book excerpt were all rated 100% human. Formal academic writing is often where detectors produce false alarms, so this was the most useful result of the test.

Is GPTZero free?

Yes. The free plan includes 10,000 words a month, the Basic scan and one Advanced scan. After the Advanced scan is used, the Basic scan still gives a verdict and a breakdown. Premium raises the allowance to 300,000 words a month.

What is the difference between GPTZero's Basic and Advanced scans?

Both showed the same model version in our test. The Advanced scan added a detailed sentence-by-sentence panel; on the free plan you get one. Our ten Basic scans still returned correct verdicts.

Does GPTZero give the same result twice?

In our test, yes. We scanned one abstract twice and got the same 100% human result.

Can I use a GPTZero score to prove I didn't use AI?

No. A detector gives an estimate, not proof. Your drafts and version history are better evidence of how you wrote.

Want to see it on your own text?

GPTZero has a free plan, so you can run the same checks on your own work.

Try GPTZero →
Eliot Byteau

Author

Eliot Byteau

Editor — AI tools for academic writing

Eliot Byteau writes DigitSelect's guides on AI tools for academic writing. Each guide starts from official documentation, public pricing, and university integrity policies; no test is claimed unless it was actually run. Limits and risks (plagiarism, fabricated citations, AI detection) are always stated.

Read my story →
ShareXLinkedInFacebook