It depends entirely on which study you read, and the honest summary is that it is one of the weaker performers on false positives. In the largest peer-reviewed multi-tool test (Weber-Wulff et al., 2023), GPTZero recorded the highest false-positive probability of the fourteen detectors examined at 50%, against 0% for the best. A 2025 NBER working paper placed it in a "secondary tier", unsuitable for very short text and susceptible to humanizing tools. Neither study is a reason to dismiss it outright, but both are reasons not to treat its output as proof.

There is a free tier: 10,000 words a month, up to 10,000 characters in a single scan, plus a small number of advanced scans. Paid plans raise the volume rather than change the method — Premium covers 300,000 words a month and Professional 500,000, with larger per-scan limits.

How much does GPTZero cost?

As of August 2026, GPTZero's own plan catalogue lists Premium at $119.88 a year and Professional at $45.99 a month or $299.88 a year. A cheaper Essential tier exists in the catalogue but is hidden from the pricing page by default, so most visitors never see it. Prices are localised by region, so what you are shown depends on where you are — we saw rupee pricing rather than dollars from one location.

Can teachers see if I used GPTZero on my own work?

No. Checking your own writing in a detector leaves no trace on your submission and is not something an institution can observe. What it also cannot do is clear you — a low score from one detector says nothing about what a different detector will report on the same text.

GPTZero vs Turnitin — which should I trust?

They answer to different people, which matters more than which is "better". Turnitin is sold to institutions and arrives inside the systems a university already runs; GPTZero is a consumer product an individual instructor can sign up for in a minute. In the one peer-reviewed test covering both, Turnitin scored highest of fourteen tools and GPTZero had the worst false-positive probability in the set. Our full comparison goes through the differences.

Does GPTZero detect humanized text?

Sometimes, and less reliably than its marketing implies. The 2025 NBER paper specifically found GPTZero susceptible to humanizing tools, and Weber-Wulff et al. measured 26% accuracy across all fourteen detectors on machine-paraphrased text. We will not tell you our own humanizer defeats it, because we have not run that test publicly — see our benchmark protocol.

GPTZero, reviewed

The best-known consumer AI detector, built by a student in the weeks after ChatGPT launched, and now the tool most likely to be the one an individual instructor reaches for. It is also the tool that recorded the highest false-positive probability in the largest peer-reviewed test of these systems. Both of those are true at once.

Last reviewed 16 August 2026 · The HumanFlow team

What it is

GPTZero was released publicly in January 2023 by Edward Tian, then a Princeton undergraduate, a few months before Turnitin shipped its own indicator. That origin shapes the product: it is a consumer tool first. There is a free tier, a browser extension, an API, and a dashboard aimed at individual teachers rather than at institutions.

The practical consequence is who runs it. Turnitin arrives bundled inside the system a university already licenses, so a student rarely chooses to encounter it. GPTZero is something one instructor can sign up for between classes — which means the person reading your score may have no institutional guidance on how to read it, and your institution may not know the tool is in use at all.

What independent testing found

Two studies are worth knowing, and neither comes from a vendor.

Weber-Wulff et al. (2023), the largest peer-reviewed multi-tool evaluation — fourteen detectors, 54 test cases, 756 tests — concluded that detection tools “are neither accurate nor reliable”, with every tool scoring below 80% accuracy. Within that set, false-positive probability ranged from 0% at the best to 50% at the worst, which was GPTZero. That is the single most important number on this page: on that test set, half the human-written samples were flagged.

Jabarian & Imas (2025), an NBER working paper — not peer-reviewed, and the paper says so — tested 1,992 human and 1,992 AI passages across four frontier models. It placed GPTZero in a “secondary tier”, noting it is not suitable for very short text and is susceptible to humanizing tools.

Neither finding means GPTZero is useless. It means a GPTZero score is a signal to look closer, never a finding of fact — and if you have been flagged by it, that 50% figure is a reasonable thing to raise. Our page on what to do when you are falsely accused covers the rest.

What it costs

Read from GPTZero’s own live plan catalogue in August 2026. Treat these as a snapshot: prices in this category move often, and GPTZero’s are localised by region, so the figures you see may differ from the ones below.

Plan	Price	Words / month	Characters / scan
Free	$0	10,000	10,000
Premium	$119.88 / year	300,000	50,000
Professional	$45.99 / month or $299.88 / year	500,000	150,000

One detail worth knowing if you are comparing prices: a cheaper Essential tier exists in GPTZero’s catalogue but is hidden from the pricing page by a default setting, so most visitors never see it. Any review quoting an entry price around $15 is quoting a plan the typical visitor is not shown.

What we are not telling you

We sell a humanizer and a detector, so here is where our interest lies and what we will not claim. We have not published a measured pass rate against GPTZero, and until we have, we will not tell you our humanizer defeats it. Our benchmark protocol is published in advance precisely so the method can be criticised before it produces a number that flatters us.

We would also say this: if your goal is to get a specific number down on a specific tool, no honest vendor can promise that, because detectors retrain and none of them publish when. What is durable is writing that reads like you — which is a different objective and the only one we think is worth selling.

Compared against

Pangram vs GPTZero — GPTZero against Pangram on what independent testing found, who each is sold to, and what happens to the text you submit.
GPTZero vs Copyleaks — GPTZero against Copyleaks on what independent testing found, who each is sold to, and what happens to the text you submit.
Sapling vs GPTZero — GPTZero against Sapling on what independent testing found, who each is sold to, and what happens to the text you submit.
GPTZero vs Originality.ai — GPTZero against Originality.ai on what independent testing found, who each is sold to, and what happens to the text you submit.

Turnitin vs GPTZero — the full side-by-side.
Turnitin, Copyleaks, Originality.ai, ZeroGPT, Scribbr, Compilatio and SafeAssign, each reviewed the same way.
How accurate AI detectors are — every published figure with its conditions.
How detectors work — perplexity, burstiness, and the ceiling the method runs into.

Two related pages: is GPTZero accurate? works through the published testing, and ZeroGPT vs GPTZero untangles the name collision, which matters if a score has been quoted at you and nobody said which tool produced it.

Common questions

Is GPTZero accurate?: It depends entirely on which study you read, and the honest summary is that it is one of the weaker performers on false positives. In the largest peer-reviewed multi-tool test (Weber-Wulff et al., 2023), GPTZero recorded the highest false-positive probability of the fourteen detectors examined at 50%, against 0% for the best. A 2025 NBER working paper placed it in a "secondary tier", unsuitable for very short text and susceptible to humanizing tools. Neither study is a reason to dismiss it outright, but both are reasons not to treat its output as proof.
Is GPTZero free?: There is a free tier: 10,000 words a month, up to 10,000 characters in a single scan, plus a small number of advanced scans. Paid plans raise the volume rather than change the method — Premium covers 300,000 words a month and Professional 500,000, with larger per-scan limits.
How much does GPTZero cost?: As of August 2026, GPTZero's own plan catalogue lists Premium at $119.88 a year and Professional at $45.99 a month or $299.88 a year. A cheaper Essential tier exists in the catalogue but is hidden from the pricing page by default, so most visitors never see it. Prices are localised by region, so what you are shown depends on where you are — we saw rupee pricing rather than dollars from one location.
Can teachers see if I used GPTZero on my own work?: No. Checking your own writing in a detector leaves no trace on your submission and is not something an institution can observe. What it also cannot do is clear you — a low score from one detector says nothing about what a different detector will report on the same text.
GPTZero vs Turnitin — which should I trust?: They answer to different people, which matters more than which is "better". Turnitin is sold to institutions and arrives inside the systems a university already runs; GPTZero is a consumer product an individual instructor can sign up for in a minute. In the one peer-reviewed test covering both, Turnitin scored highest of fourteen tools and GPTZero had the worst false-positive probability in the set. Our full comparison goes through the differences.
Does GPTZero detect humanized text?: Sometimes, and less reliably than its marketing implies. The 2025 NBER paper specifically found GPTZero susceptible to humanizing tools, and Weber-Wulff et al. measured 26% accuracy across all fourteen detectors on machine-paraphrased text. We will not tell you our own humanizer defeats it, because we have not run that test publicly — see our benchmark protocol.

Sources

1.Testing of detection tools for AI-generated text — Weber-Wulff et al. — International Journal for Educational Integrity 19:26, 2023
2.Artificial Writing and Automated Detection (NBER working paper, not peer-reviewed) — Jabarian & Imas — NBER Working Paper 34223, 2025
3.Simple techniques to bypass GenAI text detectors: implications for inclusive education — Perkins, Roe, Vu, Postma, Hickerson, McGaughran & Khuat — Int. J. of Educational Technology in Higher Education 21:53, 2024
4.GPT detectors are biased against non-native English writers — Liang, Yuksekgonul, Mao, Wu & Zou — Patterns (Cell Press), 2023