GPTZero review: the word caps, trial terms and data rules behind the detector

What GPTZero is

GPTZero estimates whether a piece of writing was produced by an AI model. You paste text or upload files, and it returns an overall assessment plus highlighted passages. Paid tiers add an Advanced Scan, multilingual detection, downloadable reports, batch uploads and, on Professional, LMS integration. There is also a Chrome extension and a developer API.

We read the pricing page, GPTZero's plan data, the terms of use, the privacy policy, the FAQ and three help-center articles. We did not scan documents ourselves, so this review is about the rules, not about detection quality.

What GPTZero claims, and what it concedes

GPTZero's FAQ calls it 'the most accurate AI detector across use-cases' and says that, for results flagged with high confidence, 99.1% of human articles are classified as human and 98.4% of AI articles as AI. Those are the vendor's figures, not ours.

The same FAQ is frank about the edges: there are always cases where human text is labeled AI and vice versa, procedural or formulaic writing can be flagged, and the model is not trained to catch AI text that has been heavily edited. It recommends looking for a long-term pattern rather than a single result, and says outright that results 'should not be used to punish students'.

Using a detector in a classroom

Detection tools have become part of academic-integrity processes at many schools, and they are controversial for exactly the reasons GPTZero lists: false positives exist, and a probability is not proof. Students who are flagged may have written the work themselves. A fair process treats a high score as a reason to talk, look at drafts and version history, and compare with earlier work.

One detail educators should notice: the terms say the site is intended for users who are at least 18, so check how your institution handles accounts for younger students before asking them to sign up.

Trials, renewals and refunds

• Trial conversion. The terms describe a 14-day free trial for new users that bills your chosen plan when it ends. Put the end date in your calendar.
• Renewal. Recurring purchases charge your card until you cancel. Cancelling in settings stops the next renewal; access runs to the end of the term.
• Refunds. There is no published window. The help center invites requests from users unhappy with accuracy or performance and says each is reviewed.
• History. Once a paid plan lapses, documents older than 30 days are no longer accessible, so export reports you need first.

Which plan fits which reader

The free tier's 10,000 words a month and 10,000 characters per scan are enough for an occasional check on a few essays. Premium, at $23.99 a month or $12.99 a month billed yearly, lifts the quota to 300,000 words and adds the Advanced Scan, multilingual detection and downloadable reports, which matter if you need to document a concern. Professional, at $45.99 a month or $24.99 billed yearly, is built for volume: 500,000 words with overage up to 2 million, 250-file batches, page-by-page scanning and LMS integration. Teams buy Professional seats, with a two-seat minimum and card-only payment.

Since unused words are not described as rolling over, size the plan to a typical month, not your busiest one.

If you only need occasional checks, there is little reason to leave the free plan. Upgrade when you actually need reports to share with colleagues or students, because those downloadable reports are where the paid tiers start to earn their price.

What happens to the documents you scan

The answer depends on how you scan. Through the website, inputs are stored and used in aggregate to improve the service. Through the API, documents are not stored unless you enable persistence. Student essays are personal data in most settings, so schools should read the privacy policy and ask for a data processing agreement before rolling GPTZero out widely. The policy lists 'content for ai detection' among collected data but gives no specific retention period for it.