The short answer
ZeroGPT is a free AI detector whose own site claims 98.4% detection accuracy and under 1% false positives. Independent tests give lower and more mixed results, such as 64.35% accuracy and 16.67% false positives in one 2025 study. Plagino has not benchmarked ZeroGPT, so treat its score as a signal, not proof.
What is the ZeroGPT AI detector and how does it work?
ZeroGPT is a web-based AI content detector that estimates whether text was written by a large language model such as ChatGPT, Claude or Gemini. Its home page calls the method DeepAnalyse and says it covers more than 35 model families. The service is run by Olive Works LLC, a company organized in Wyoming, according to its privacy policy. You paste text or upload a file and get an overall AI percentage, a text label and highlighted sentences.
ZeroGPT's FAQ says the detector looks at token patterns, burstiness, entropy and ensemble classifier features learned from human and AI text. In plain terms, it measures how predictable the wording is and how much sentence style varies. The FAQ asks for at least 150 to 200 words, and says 500 to 1,000 words or more make the result more stable.
ZeroGPT is not GPTZero
The names look alike, but these are different companies and different products. This page covers ZeroGPT only. Plagino's benchmark tested GPTZero but not ZeroGPT, and the GPTZero results are on the Plagino vs GPTZero page.
ZeroGPT also sells other tools, including a plagiarism checker, a paraphraser, a grammar checker and a tool that rewrites AI text to read as human. Plagino is a checking tool and does not rewrite text, and this page does not cover that product.
Is ZeroGPT accurate? Its claims against independent tests
ZeroGPT's home page states 98.4% detection accuracy, a false positive rate under 1% and coverage of more than 35 model families. Its FAQ describes the goal as "pushing toward >98% on internal evaluations". The pages Plagino read do not publish the test set, the date or the method behind these figures, so they are claims, not results that others can check.
Independent tests give a much wider range. The table lists every source with real numbers that Plagino found, who ran each test and what to keep in mind. Dates and figures are as of October 2026.
| Source | Who ran it | What was tested | Result for ZeroGPT | Catch |
|---|---|---|---|---|
| ZeroGPT home page and FAQ (October 2026) | ZeroGPT | Not stated | 98.4% detection accuracy; under 1% false positives | Vendor claim; no test set, date or method published on those pages |
| RAID benchmark (Dugan and colleagues, 2024) | University researchers | More than 6 million machine-generated texts across 8 domains; ZeroGPT queried through its API | 1.42% of human text flagged at a 0.5 threshold; 65.5% of AI text caught when tuned toward a 5% false-positive target, which ZeroGPT could not reach in many domains | Older ZeroGPT version; API results, not the web app; the authors set the thresholds |
| Erol and colleagues, Acta Neurochirurgica (2025) | Academic study | 250 human and 750 ChatGPT-written neurosurgery abstracts and introductions, scanned 1 to 15 June 2024 | 94.4% sensitivity and 93.2% specificity at a cutoff of 75.3, so about 6.8% of human texts were flagged | One field; ChatGPT 3.5, 4 and 4o only; the cutoff comes from the authors, and the label you see on the site may use a different one |
| Pratama, PeerJ Computer Science (2025) | Academic study | 72 human abstracts published by 2021 and 144 abstracts written by ChatGPT o1 and Gemini 2.0 Pro Experimental | 64.35% accuracy; 16.67% of human abstracts flagged; 45.14% of AI abstracts missed | Abstracts only; one study; newer AI models than the 2024 tests |
| Scribbr 12-tool test (2026 edition) | Scribbr, which sells its own AI detector | 30 short texts of 1,000 to 1,500 characters, 5 of them human-written | 64% accuracy; 1 of 5 human texts flagged; caught 4 of 5 GPT-3.5 texts and 3 of 5 GPT-4 texts | Tiny sample; GPT-3.5 and GPT-4 era; the tester sells a rival tool |
| GPTZero review of ZeroGPT (July 2026) | GPTZero, a rival detector | 6 raw AI texts and 7 human texts | AI texts scored from 29.9% to 100% AI; human texts scored from 0% to 18.7% AI | 13 texts; the tester sells a rival tool; every human text still got a human verdict |
| Liang and colleagues, Patterns (2023) | Stanford researchers | Seven detectors, ZeroGPT among them, on 91 TOEFL essays by non-native English writers and a set of US student essays | Averaged across all seven detectors: 61.22% of TOEFL essays flagged as AI | ZeroGPT's own rate is not broken out in the sources Plagino read; March 2023 versions |
| Plagino benchmark (March 2026) | Plagino | Seven other detectors | Not benchmarked | Plagino has not tested ZeroGPT |
The sources are the RAID benchmark paper from researchers at several universities, Erol and colleagues in Acta Neurochirurgica, Pratama's study in PeerJ Computer Science, Scribbr's detector comparison, GPTZero's ZeroGPT review and the Stanford study by Liang and colleagues.
None of the independent tests reproduced both a 98.4% catch rate and under 1% false positives. The academic tests put ZeroGPT anywhere from about 64% overall accuracy to 94.4% sensitivity, and three things explain most of the spread. First, the texts differ: abstracts from one field, short samples and mixed documents are different tasks. Second, the AI models differ: the 2024 study used ChatGPT 3.5, 4 and 4o, and the 2025 study used ChatGPT o1 and Gemini 2.0 Pro Experimental. Third, the cutoff matters, because a detector looks more or less accurate depending on the score above which it says AI.
Blog tests also circulate figures such as a 20.5% false-positive rate from a 160-text comparison. Plagino could not verify the method or the texts behind those posts, so it does not use them. Small tests add noise: in a five-text human sample like Scribbr's, one flagged text moves the rate by 20 points.
How accurate is ZeroGPT on human writing? False positives explained
On human writing, ZeroGPT's error rate in independent tests ran from about 7% to about 17%, well above its claim of under 1%. Erol's team found about 6.8% of human neurosurgery texts flagged at their cutoff, and Pratama found 16.67% of 72 human abstracts flagged. RAID is a counterpoint: at a 0.5 threshold it flagged 1.42% of human text, so the result depends heavily on the text and the threshold.
A flag means ZeroGPT found patterns it links with AI writing, and human writing can share them. ZeroGPT's own FAQ says highly polished, formulaic or low-entropy writing can resemble AI. Its privacy policy adds that its outputs should not be treated as the sole factual determination of authorship, misconduct or academic integrity.
Writers who learned English as a second language face extra risk. In the Stanford study, seven detectors including ZeroGPT flagged an average of 61.22% of TOEFL essays as AI, while they classified US student essays accurately. Pratama's study found ZeroGPT's accuracy was similar for native and non-native authors, so the bias is not proven for ZeroGPT on its own, and Plagino has not tested it.
If a flag lands on work you wrote yourself, do not rewrite it to chase a lower score. Gather evidence of your process and use your institution's procedure. Our guide to AI detector false positives explains how to read an error rate, and the guide for students falsely accused of using AI walks through the next steps.
Is ZeroGPT reliable compared with Turnitin and Plagino?
ZeroGPT's results are not the same as Turnitin's, and no free checker can predict an institution's Turnitin report. They are separate products with separate models, and Turnitin is sold to institutions rather than to individual students. Our guide on whether Turnitin detects AI covers what Turnitin does and does not show.
The only head-to-head data Plagino can stand behind is its own benchmark. The corpus had 1,000 documents: 500 AI-written (GPT-4o, Claude and Gemini, some lightly human-edited) and 500 human-written. Every tool saw the same files through its own web app on default settings, in March 2026. ZeroGPT was not one of the seven tools, so the table shows the rows we have and marks ZeroGPT as not tested.
| Detector | AI documents caught | 95% interval | Human documents wrongly flagged | 95% interval |
|---|---|---|---|---|
| Turnitin | 98.2% | 96.6–99.1% | 1.4% | 0.7–2.9% |
| Plagino | 97.4% | 95.6–98.5% | 0.8% | 0.3–2.0% |
| Originality.ai | 95.6% | 93.4–97.1% | 4.1% | 2.7–6.2% |
| Copyleaks | 91.3% | 88.5–93.5% | 2.2% | 1.2–3.9% |
| GPTZero | 88.9% | 85.8–91.4% | 5.7% | 4.0–8.1% |
| Academi.cx | 86.4% | 83.1–89.1% | 1.9% | 1.0–3.5% |
| TurnDetect | 79.5% | 75.7–82.8% | 6.3% | 4.5–8.8% |
| ZeroGPT | Not benchmarked | Not benchmarked | Not benchmarked | Not benchmarked |
Because ZeroGPT's published tests use different texts and cutoffs, their numbers cannot be lined up against these rows, and reading across would mislead. What the table does show is the spread among tools that were tested: false-positive rates from 0.8% to 6.3%. If you want a tested option, try the Plagino AI detector, and for how other tools compare with Turnitin read our guide to the closest alternatives to Turnitin.
What to do if ZeroGPT flags your own writing
- 1
Keep your drafts and history. Save your notes, outlines and earlier drafts. If you wrote in Google Docs or Word, version history shows how the text grew over time.
- 2
Scan the whole document. ZeroGPT's FAQ asks for at least 150 to 200 words and says longer text gives more stable results. Scan the full piece, not just the paragraph that worries you.
- 3
Read the highlighted sentences. Check whether the highlights fall on definitions, method steps, lists or standard phrases. Those are the places where human writing most often looks formulaic.
- 4
Note the percentage, the label and the date. Write down the AI percentage, the text label and the date. In Scribbr's test, very different percentages sometimes carried the same label, so record both.
- 5
Use the official process. Share your drafts with your instructor and ask how the school reviews AI flags. A detector score alone should not decide an academic case.
Is ZeroGPT a good AI detector for the price?
ZeroGPT's free plan is large. Its pricing page lists 15,000 characters per scan and 1,250,000 words a month at no cost, with no card needed. The free tool shows ads, and the paid plans remove them, raise the limits and add PDF reports. The prices below were read in October 2026.
Whether it is a good AI detector for the price depends on what you are buying. The price buys volume and extra tools, not an independently tested accuracy figure.
| ZeroGPT | Plagino | |
|---|---|---|
| Free use | Free plan: 15,000 characters per scan and 1,250,000 words a month, no card, with ads | 3 scans per month, no card |
| Paid plans | Pro $8.99 a month billed yearly ($14.99 monthly), Plus $13.99 ($21.99), Max $19.99 ($30.99); 100,000 to 150,000 characters per scan | Pro: $19/month for 20 scans per day |
| Teams and schools | EDU plan up to 250,000 characters per scan, Expert plan up to 350,000, plus a pay-as-you-go API | Team: custom pricing for departments and writing centres |
| How you submit | Paste text or upload .txt, .docx or text-based .pdf; batch upload of 50 to 100 files on paid plans | .docx upload in the web app |
| Report | Overall AI percentage, a text label and highlighted sentences; PDF report on paid plans | PDF with sentence-level AI highlights and matched sources |
| Plagiarism check | Separate plagiarism checker: 300 words once on the free plan, 35,000 or 60,000 words a month on Plus and Max | Open-access academic literature via CORE only |
| Published accuracy test | Claims 98.4% on its home page; no test set published on the pages Plagino read | March 2026 benchmark, with intervals on the benchmark page |
Plagino's pricing page has the current plan details. If you need a plagiarism check as well, Plagino's plagiarism checker matches against open-access research through CORE, and does not search the public web.
What does ZeroGPT do with the text you scan?
ZeroGPT's FAQ says text checked on the site is not saved, shared, published or used to train its model, and is discarded after the result. Its privacy policy is more detailed. It says private content is not used to train models and is not sold, and that it is kept only as long as reasonably necessary to run the service, with longer retention possible for security, abuse prevention or legal reasons.
The same policy says the site shows advertising from providers including Google and Freestar, though it says private text is not shared with advertisers. The FAQ says ZeroGPT's infrastructure is hosted in Germany, while the policy lists processing in the United States, Germany and other places. These are the company's own statements. If you are scanning an unpublished thesis or manuscript, read both documents first. Plagino's own terms are on its privacy page.
Where ZeroGPT is the better choice
ZeroGPT does several things that Plagino does not, and for some users those matter more than a published error rate.
A very large free allowance
The free plan covers 15,000 characters a scan and 1,250,000 words a month, and the tool needs no sign-up to run a scan. Plagino's free plan gives 3 scans a month. For a quick look at a long document, ZeroGPT's allowance is a real advantage.
Batch scanning, an API and extra tools
Paid plans allow batch checks of 50 to 100 files, ZeroGPT sells an API, and the same account includes a paraphraser, summarizer, grammar checker and translator. Plagino offers an AI check and a literature match only, and its Team plan is custom.
Language coverage
ZeroGPT says it supports all languages. Plagino has not tested that claim, and its own benchmark covered English only.
Where Plagino is the better choice
Plagino is our tool, so read this with that in mind. Plagino cannot say it beats ZeroGPT on accuracy, because it has not tested ZeroGPT. What it can say is that its numbers come with the method: in the March 2026 benchmark Plagino caught 97.4% of AI-written documents and wrongly flagged 0.8% of human-written ones, with 95% intervals on the benchmark page. Turnitin, the reference, caught 98.2% and flagged 1.4%.
Plagino is built for academic drafts. You upload a .docx file and get a PDF report with sentence-level AI-writing highlights and matched academic sources, which you can keep and discuss with a tutor. The free plan gives 3 scans a month with no card, and Pro costs $19 a month for 20 scans a day. Plagino checks writing and does not rewrite it.
Whichever tool you use, a score is a signal, not proof. For other head-to-head results, see Plagino vs GPTZero, Plagino vs Originality.ai, Plagino vs Copyleaks, Plagino vs Winston AI and Plagino vs Scribbr. Our guide to the best AI detectors compares the wider field.
Test conditions and limitations
These limits apply to Plagino's results and to every outside test cited above.
What the benchmark does not cover
Plagino's corpus was English only, and its AI documents came from GPT-4o, Claude and Gemini, so results on other languages or newer models may differ. We scored at document level, so one flagged sentence in a human essay counted as a false positive, and we used each tool's default settings in March 2026. Each rate rests on 500 documents, which separates large gaps but not small ones. ZeroGPT was not in the test, and Plagino has not benchmarked it.
What the outside tests do not cover
The outside studies used ZeroGPT at different dates from 2023 to 2026, and ZeroGPT does not publish version numbers on the pages Plagino read. Several tests used short texts or a single field, and two were run by competing vendors. Treat each figure as a snapshot of one test, and scan your own work again if the result matters.
For a wider look at how detectors err and why, read our guide to AI detector false positives, or try the Plagino AI detector on your own draft.
Frequently asked questions
Is ZeroGPT accurate?
Not as accurate as it claims in the independent tests Plagino found. ZeroGPT says 98.4% detection with under 1% false positives, while a 2025 PeerJ study measured 64.35% accuracy and 16.67% false positives on abstracts. A 2024 study found 94.4% sensitivity and 93.2% specificity at a chosen cutoff, so results depend on the text and the test.
How accurate is ZeroGPT?
Independent figures range from 64% overall accuracy to 94.4% sensitivity, depending on the test. The lowest were 64.35% in a 2025 abstract study and 64% in Scribbr's 30-text test, and the highest was 94.4% in a 2024 neurosurgery study. Plagino has not benchmarked ZeroGPT, so it has no figure of its own.
Is ZeroGPT reliable?
It is reliable enough for a quick look, not for a decision about a person. ZeroGPT's own privacy policy says its outputs should not be the sole basis for deciding authorship or misconduct, and some independent tests flagged about 7% to 17% of human writing. Use the score as a signal and add other evidence.
Is ZeroGPT a good AI detector?
It is a good free tool for a first check and a weak basis for a verdict. The free plan is large and easy to use, but Plagino found no independent test that reproduces its 98.4% claim. For high-stakes work, use a detector that publishes its method and error rates, and still treat the score as a signal.
Why does ZeroGPT say my writing is AI when I wrote it myself?
ZeroGPT found patterns it links with AI writing, and human writing can share them. Its own FAQ says polished, formulaic or low-entropy writing can resemble AI. Short passages, method steps and work by non-native English writers are more at risk, so keep your drafts and ask for a review.
Are ZeroGPT's results the same as Turnitin's?
No. They are different products with different models, so a ZeroGPT score does not predict your school's Turnitin report. Turnitin is sold to institutions, and in Plagino's March 2026 benchmark it caught 98.2% of AI-written documents and wrongly flagged 1.4% of human ones. Plagino has not benchmarked ZeroGPT.
Is ZeroGPT the same as GPTZero?
No. They are different companies and products with similar names. ZeroGPT's privacy policy names Olive Works LLC, a Wyoming company, as its operator. Plagino's benchmark tested GPTZero but not ZeroGPT.
Is ZeroGPT free?
Yes, ZeroGPT has a free plan with 15,000 characters per scan and 1,250,000 words a month, as of October 2026. Paid plans start at $8.99 a month billed yearly, or $14.99 billed monthly, for Pro, which removes ads, allows 100,000 characters per scan and adds PDF reports. Check the pricing page for your own price before you subscribe.
Does ZeroGPT save my text?
ZeroGPT's FAQ says text checked on the site is not saved, shared, published or used to train its model. Its privacy policy says private content is kept only as long as reasonably necessary, and may be kept longer for security, abuse or legal reasons. These are the company's statements, so read them before scanning unpublished work.
Can ZeroGPT detect ChatGPT?
Often, but not always. A 2024 study found it caught 94.4% of ChatGPT-written neurosurgery texts at its chosen cutoff, and Scribbr's test found 4 of 5 GPT-3.5 texts and 3 of 5 GPT-4 texts. A 2025 study found it missed 45.14% of abstracts written by ChatGPT o1 and Gemini 2.0 Pro Experimental.
Does ZeroGPT work in languages other than English?
ZeroGPT says it supports all languages, but that is a vendor claim. Plagino has not tested ZeroGPT in any language, and the independent tests Plagino read used English text. Plagino's own benchmark was English only.
How much text can I check in ZeroGPT?
The free plan allows 15,000 characters per scan, and paid plans allow 100,000 to 350,000, as of October 2026. ZeroGPT's FAQ recommends at least 150 to 200 words. It also says 500 to 1,000 words or more make results more stable.
Who owns ZeroGPT?
ZeroGPT is operated by Olive Works LLC, a company organized in Wyoming, United States, according to its privacy policy. It is a separate company from GPTZero. Plagino found no other ownership details on the pages it read.
