Guide
ZeroGPT: How Accurate It Is, What It Gets Wrong, and Who Runs It
A plain-English guide to ZeroGPT: what its accuracy claim is measured on, what independent studies found, how it differs from GPTZero and zerogpt.net, what its terms of use say, who operates it, and what powers its image detector.
Quick answer: ZeroGPT is a free, no-signup AI text detector at zerogpt.com that returns a percentage and highlights the sentences it believes were machine-written. ZeroGPT states 98.4% detection accuracy with a false positive rate under 1%, produced by what it calls DeepAnalyse Technology (ZeroGPT). Independent studies have reported lower figures: a 2025 study in PeerJ Computer Science put ZeroGPT’s overall accuracy on journal abstracts at 64.35%, with a 16.67% false positive rate and a 45.15% false negative rate (Pratama, 23 June 2025), while a November 2025 study found it returned 6.5% AI on human-written text and 92.5% on fully AI-generated text (Cheng et al.). ZeroGPT is not GPTZero; they are separate companies, and a third site at zerogpt.net uses the same brand name. ZeroGPT’s terms of use provide the service “as is” and disclaim warranties of accuracy and reliability (ZeroGPT terms). Its image and video detectors are powered by TruthScan.
This guide covers ZeroGPT specifically. For the underlying detection mechanism across all tools, see how AI detectors work. For ranked alternatives, see best AI detectors. For the institutional detector most students are assessed by, see Turnitin AI detection.
What ZeroGPT is
ZeroGPT is a web tool. You paste text into a box, press a button, and get a percentage back.
It is free and needs no account. ZeroGPT’s pricing page lists a Free plan at “$0 forever” with no credit card required (ZeroGPT pricing).
It returns a percentage with highlighted sentences, and offers a PDF export. ZeroGPT also sells Business/EDU plans and an Enterprise plan with a user-management dashboard; it publishes no roster or class integration.
It is a suite. The site lists an AI Humanizer, AI Image Detector, AI Video Detector, Plagiarism Checker, AI Paraphraser, AI Grammar Check, AI Summarizer, AI Translator, Word Counter, Dictionary and AI Email Helper alongside the text detector.
It is operated by Olive Works LLC. ZeroGPT’s terms of use name the operating entity as Olive Works LLC, at 5830 E 2nd St, Ste 7000 #9537, Casper, Wyoming (ZeroGPT terms).
ZeroGPT is not GPTZero, and neither is zerogpt.net
ZeroGPT and GPTZero are different products from different companies. GPTZero is at gptzero.me and is now part of Superhuman; ZeroGPT is at zerogpt.com and is operated by Olive Works LLC. They have different detection models, accuracy claims, pricing and independent test results. If someone tells you “the detector said 92%”, the first question is which detector.
There is also a third site. zerogpt.net uses the ZeroGPT brand name, carries its own copyright line (”© 2026 ZeroGPT. All rights reserved.”), names no operating entity, and advertises a 0.0% false positive rate. The site states no relationship to zerogpt.com. Check which domain produced a result before quoting it.
| Property | ZeroGPT (zerogpt.com) | GPTZero (gptzero.me) | zerogpt.net |
|---|---|---|---|
| Operator named on site | Olive Works LLC, Wyoming | GPTZero (Superhuman) | None named |
| Accuracy claim | 98.4% | 99% | Not published |
| False positive claim | Under 1% | At or below 1% | 0.0% |
| Sign-up required for free use | No | No | No |
| Detection technology name | DeepAnalyse | Not branded | Not published |
For ranked comparisons with independent scores, see best AI detectors.
How ZeroGPT’s detector works
ZeroGPT attributes its results to DeepAnalyse Technology, a trademarked name used on its homepage and marketing (ZeroGPT). ZeroGPT does not publish a technical description of how DeepAnalyse works.
The interface highlights individual sentences as AI-generated and returns a document-level percentage. For the mechanisms detectors generally use, see how AI detectors work.
ZeroGPT’s FAQ frames detection results as something to use “within a fair review process” (ZeroGPT FAQ). Its terms of use say a detection score “does not constitute definitive proof” of authorship and must not be “the sole basis for a decision” that materially affects someone (ZeroGPT terms).
What ZeroGPT will and will not scan
| Requirement | What ZeroGPT states | Where it is stated |
|---|---|---|
| Maximum input, free tier | 15,000 characters per check | Homepage and detector page |
| Minimum length for a reliable result | 150 to 200 or more words; 500 to 1,000 or more for stability | FAQ |
| Languages | ”All available languages are supported across all the tools” | FAQ |
| Site interface languages | Six: English, French, German, Spanish, Italian, Portuguese | Site interface |
| Text retention | Text is “not saved, shared, published online, or used to train” the model and is “discarded” after processing | FAQ |
The minimum length is in the FAQ. ZeroGPT’s FAQ says the tool needs 150 to 200 or more words for a reliable reading, and that 500 to 1,000 or more gives better stability (ZeroGPT FAQ). The detector will still return a percentage on shorter text, so a score on a short passage falls below the length ZeroGPT recommends.
Languages. ZeroGPT’s homepage says it supports “all the languages with the highest accuracy rate”. It does not publish per-language accuracy figures. The independent studies cited on this page tested English.
How to read a ZeroGPT score
A percentage is not a proportion of the document. It is an aggregate of sentence-level judgements, not a count of words that came from a chatbot.
The highlighted sentences are where the score came from. Look at those sentences. ZeroGPT’s FAQ says that “highly polished, formulaic, or low-entropy writing can resemble AI” (ZeroGPT FAQ).
Different detectors give different answers. A ZeroGPT score does not predict another tool’s score on the same text.
Which AI models ZeroGPT says it can detect
ZeroGPT’s site says it covers “35+” LLM families and names ChatGPT, GPT-3, GPT-4, GPT-5, GPT-6, Gemini, Grok, Perplexity AI, Claude, DeepSeek and Llama models (ZeroGPT). It does not publish a detection rate for each model.
The independent studies below tested ZeroGPT against output from ChatGPT o1 and Gemini 2.0 Pro (Pratama, 2025) and earlier ChatGPT versions. None of the studies cited here tested GPT-5, GPT-6, Claude or Grok output. The PeerJ study reported “poor performance” on the latest model versions for ZeroGPT and DetectGPT (Pratama, 2025).
What ZeroGPT says about accuracy
The homepage states “98.4% Detection accuracy” and “<1% False positive rate” (ZeroGPT).
The FAQ describes the figure as from internal evaluations: “best-in-class performance — pushing toward >98% on internal evaluations” (ZeroGPT FAQ).
ZeroGPT does not publish the test set, sample size or definition of accuracy behind these figures.
What independent testing found
| Study | Date | What it measured | ZeroGPT result |
|---|---|---|---|
| Weber-Wulff et al., International Journal for Educational Integrity | 25 December 2023 | 14 detectors, ZeroGPT included, on human, AI and machine-paraphrased text | ”The highest accuracy values are below 80%” across the tools tested |
| Kar et al., Indian Journal of Psychological Medicine | 11 May 2024 online | 10 free detectors on one AI-written article and three paraphrased versions | 95.03% AI on the original; 96.74% on the Grammarly paraphrase; 90.61% on the ChatGPT paraphrase; 32.31% on the QuillBot paraphrase |
| Pratama, PeerJ Computer Science | 23 June 2025 | 3 detectors (GPTZero, ZeroGPT, DetectGPT) on journal abstracts, human-written and generated or enhanced by ChatGPT o1 and Gemini 2.0 Pro | 64.35% overall accuracy; 16.67% false positive rate; 45.15% false negative rate |
| Cheng et al., Advances in Simulation | 22 November 2025 | 3 detectors across five AI-use conditions, 150 introduction sections | 6.5% AI detected on human text; 20.2% (lightly edited AI) to 92.5% (fully AI-generated); significant overall difference across the five conditions (p < 0.001) |
Weber-Wulff and colleagues tested 14 detectors in 2023 and reported that “the highest accuracy values are below 80%”, that the tools leaned towards classifying text as human-written, and that paraphrasing made results worse (Weber-Wulff et al., preprint at arXiv:2306.15666).
Pratama’s 2025 study put ZeroGPT’s overall accuracy on journal abstracts at 64.35%, with a 16.67% false positive rate and a 45.15% false negative rate (Pratama, PeerJ Computer Science 11:e2953). It found ZeroGPT’s results varied by field, with a median of 0% detected in technology and engineering abstracts against 30.48% for interdisciplinary work (p < 0.001), and no significant difference between native and non-native authors.
Cheng and colleagues, testing manuscript introduction sections, found ZeroGPT returned an average of 6.5% AI content on human-written text and 92.5% on fully AI-generated text, falling to 20.2% on lightly edited AI text, with a significant overall difference across the five conditions (p < 0.001) (Cheng et al., Advances in Simulation). The authors concluded that these tools “should not be relied upon alone at this time.”
Kar and colleagues found ZeroGPT scored a QuillBot-paraphrased version of AI text at 32.31% AI, against 95.03% for the original and 96.74% for a Grammarly-paraphrased version (Kar et al.). It was a single sample of under 500 words.
The RAID benchmark (Dugan et al., ACL 2024) measured ZeroGPT at 65.5% average accuracy at a fixed 5% false-positive rate.
One widely cited figure is a seven-detector average. The 2023 Stanford study by Weixin Liang and colleagues reported a mean false positive rate of 61.22% on TOEFL essays by non-native English speakers, against about 5.19% on US eighth-grade essays, across seven detectors that included ZeroGPT (Liang et al., Patterns). The 61.22% is the average across the seven, not ZeroGPT’s own figure.
The 2026 Vrije Universiteit Brussel study did not test ZeroGPT. It tested GPTZero, Copyleaks, Turnitin and Pangram (Van Vlasselaer et al.).
False positives
ZeroGPT’s FAQ says regular writing is most at risk: “highly polished, formulaic, or low-entropy writing can resemble AI” (ZeroGPT FAQ). It suggests adding “concrete details, sources, and varied sentence structures” and avoiding generic language.
The field and language results from Pratama’s study are above: ZeroGPT’s results varied by discipline, and did not differ significantly between native and non-native authors.
What ZeroGPT’s terms of use say
The terms provide the service “as is”. ZeroGPT’s terms of use provide the services and outputs “as is” and “as available” and disclaim warranties including accuracy and reliability. They say a detection score “does not constitute definitive proof” of authorship and must not be “the sole basis for a decision” that materially affects someone (ZeroGPT terms of use). Terms of this kind are common in consumer software.
Liability is capped. The terms limit the total liability of ZeroGPT and Olive Works LLC to the greater of US $100 or the amount the user paid for the service in the six months before the claim. They are governed by Wyoming law.
The image and video detectors are powered by TruthScan
ZeroGPT’s AI image detection is powered by TruthScan. On 22 January 2026 ZeroGPT announced a partnership with TruthScan, a deepfake-detection company, saying its image detection is “powered by TruthScan’s software” (press release issued by EZ Newswire on behalf of ZeroGPT). The release quoted ZeroGPT chief executive Rawad Baroud: “People use our platform to detect AI content because they trust the accuracy of our product.”
NewsGuard tested the image detector. NewsGuard’s audit, published on 8 May 2026, tested five image detectors on 45 images and reported that ZeroGPT flagged 3 of 15 authentic news photographs as AI-generated, while catching 93.33% of manipulated images (NewsGuard). Baroud told NewsGuard that resizing, compression, lighting, contrast and blur can affect results on real photographs. A ZeroGPT image score and a ZeroGPT text score come from different systems.
ZeroGPT’s video detector page also carries a “Powered by TruthScan” badge (ZeroGPT).
The detector and the humaniser
ZeroGPT lists an AI Humanizer and an AI Paraphraser in the same menu as its AI detector. For those tools generally, see best AI humanizers.
What it costs
| Plan | What is published |
|---|---|
| Free | $0 forever, no credit card required; 15,000 characters per check; text detection, highlighting and PDF report |
| Personal | Monthly or annual billing; “40% off annually. It’s like 147 days free” |
| Business and EDU | Monthly or annual billing |
| Enterprise | Custom quote; security, user-management dashboard and integrations |
| API | Pay as you go |
Source: ZeroGPT pricing. Prices for paid plans are shown on the pricing page and may appear in your local currency.
How ZeroGPT compares with tools it gets confused with
| Tool | Who sees the score | Independent result | Vendor accuracy claim |
|---|---|---|---|
| ZeroGPT | Whoever pastes the text | 64.35% overall accuracy on journal abstracts (Pratama, 2025) | 98.4% |
| Turnitin | Instructors and administrators only | Average 0.0% AI detected on fully AI-generated papers (Van Vlasselaer et al., 2026) | Under 1% false positives above a 20% score |
| Pangram | Whoever runs it | Average 85.7% AI detected on fully AI papers; 45.5% on hybrid papers whose actual AI share was 46.3% (Van Vlasselaer et al., 2026) | — |
| Copyleaks | Whoever runs it | Average 14.0% AI detected on fully AI papers; 0.0% on human papers (Van Vlasselaer et al., 2026) | Over 99% |
The figures come from different studies on different text, so they are not a like-for-like ranking. In the 2026 peer-reviewed comparison, only Pangram’s average on fully AI-generated papers came close to their true share; Turnitin averaged 0.0%, GPTZero 14.3% and Copyleaks 14.0% (Van Vlasselaer et al.). Full rankings are on best AI detectors.
What to do, by situation
If you are a student who has been flagged by ZeroGPT. The strongest material you have is process evidence: version history from Google Docs or Word, drafts with timestamps, notes, outlines and references. ZeroGPT’s own terms say a score must not be “the sole basis for a decision”. In Matter of Newby v Adelphi University, decided on 28 January 2026, a New York court annulled an integrity finding that followed a Turnitin AI score, after finding the university’s process unfair; the student had submitted reports from two other detectors rating the essay human-written.
If you are a teacher considering ZeroGPT. The 2025 PeerJ study measured a 16.67% false positive rate on journal abstracts, and ZeroGPT’s own FAQ frames results as one input “within a fair review process”. Best AI for teachers covers assessment approaches that do not depend on detection.
If you are checking your own writing. The free tier covers this. Paste more than 200 words, as ZeroGPT recommends, read the highlighted sentences rather than the percentage, and expect a different answer from a different tool.
If you are checking an image. ZeroGPT’s image detector is powered by TruthScan. See best AI image detectors for how image detectors compare.
If you are deciding whether to pay. Test the free tier on text whose origin you know first.
Who runs ZeroGPT
The operating entity named in ZeroGPT’s terms of use is Olive Works LLC, at 5830 E 2nd St, Ste 7000 #9537, Casper, Wyoming (ZeroGPT terms). The chief executive is Rawad Baroud, named in ZeroGPT’s January 2026 press release (EZ Newswire on behalf of ZeroGPT). The site’s published contact address is support@zerogpt.com.
Summary
- The product: a free, no-account text detector at zerogpt.com returning a percentage and highlighted sentences, capped at 15,000 characters per check, alongside image, video, humaniser and other tools.
- ZeroGPT’s claim: 98.4% accuracy and under 1% false positives, which its FAQ describes as from internal evaluations.
- Independent results: 64.35% overall accuracy on journal abstracts (PeerJ, 2025); 6.5% AI on human text and 92.5% on fully AI text (Advances in Simulation, 2025); 65.5% average in the RAID benchmark (2024).
- The terms: the service is provided “as is”, a score is not definitive proof, and liability is capped.
- Operator: Olive Works LLC of Casper, Wyoming; chief executive Rawad Baroud. Image and video detection are powered by TruthScan.
- Not GPTZero, and not zerogpt.net, a separate site using the same name.
Frequently asked questions
Is ZeroGPT accurate?
ZeroGPT states 98.4% detection accuracy and a false positive rate under 1%, and its FAQ describes the figure as coming from internal evaluations. Independent studies have reported lower figures: a study in PeerJ Computer Science on 23 June 2025 put ZeroGPT’s overall accuracy at 64.35% on journal abstracts, with a 16.67% false positive rate and a 45.15% false negative rate. A November 2025 study found ZeroGPT returned 6.5% AI content on human-written text and 92.5% on fully AI-generated text.
Is ZeroGPT the same as GPTZero?
No. They are separate products from separate companies with separate detection models and accuracy claims. ZeroGPT is at zerogpt.com and is operated by Olive Works LLC of Casper, Wyoming; GPTZero is at gptzero.me and is part of Superhuman. If a score is being quoted to you, establish which tool generated it.
Is ZeroGPT free?
Yes, and the free tier needs no account or credit card. ZeroGPT’s pricing page lists a Free plan at “$0 forever” that includes text detection with sentence highlighting and a PDF report, capped at 15,000 characters per check. Paid Personal, Business, EDU and Enterprise plans are offered with monthly and annual billing, with an advertised 40% annual discount.
Why does ZeroGPT say my writing is AI when I wrote it myself?
Detectors respond to how predictable and regular prose is, not to who wrote it. ZeroGPT’s FAQ states that “highly polished, formulaic, or low-entropy writing can resemble AI”. A 2023 Stanford study found seven detectors, ZeroGPT among them, flagged TOEFL essays by non-native English writers at an average false positive rate of 61.22% across the seven; the 2025 PeerJ study found no significant native/non-native difference in ZeroGPT’s results.
Can ZeroGPT detect ChatGPT?
ZeroGPT lists ChatGPT, GPT-3, GPT-4, GPT-5 and GPT-6 among the model families it says it covers, but publishes no detection rate for each model. A 2024 study scored an unedited ChatGPT article at 95.03% AI. The 2025 PeerJ study, using ChatGPT o1 and Gemini 2.0 Pro, recorded a 45.15% false negative rate on AI-generated abstracts. None of the studies cited here tested GPT-5 or GPT-6 output.
Does ZeroGPT detect paraphrased or humanised text?
Results vary with the paraphrasing tool. In a 2024 study in the Indian Journal of Psychological Medicine, ZeroGPT scored an AI-written article at 95.03% AI, a QuillBot-paraphrased version at 32.31%, a ChatGPT-paraphrased version at 90.61% and a Grammarly-paraphrased version at 96.74%, on a single sample of under 500 words. The 2023 Weber-Wulff study found paraphrasing made results worse across the 14 tools it tested.
How many words does ZeroGPT need to give a reliable result?
ZeroGPT’s FAQ says 150 to 200 or more words, and recommends 500 to 1,000 or more for stable results. The detector still returns a percentage on shorter text. The maximum input on the free tier is 15,000 characters per check.
Is ZeroGPT safe to use with my own unpublished writing?
ZeroGPT’s FAQ states that submitted text is “not saved, shared, published online, or used to train” its model and is “discarded” after the result is produced. Its terms provide the service “as is” and cap liability. For confidential material, such as client work under NDA, check your own obligations before pasting it into any web tool.
Do teachers use ZeroGPT?
Some do. ZeroGPT offers Business/EDU and Enterprise plans, and its FAQ frames results as something to use “within a fair review process”. Its terms say a score must not be “the sole basis for a decision” that materially affects someone.
How does ZeroGPT compare to Turnitin?
They are different kinds of tool. Turnitin is an institutional product whose AI writing indicator is visible only to instructors and administrators, with a published false-positive claim of under 1% on documents scoring above 20%. ZeroGPT is a free tool anyone can use on any text. In the June 2026 peer-reviewed comparison, Turnitin scored all 40 fully AI-generated papers below its 20% reporting threshold; the 2025 PeerJ study put ZeroGPT’s overall accuracy at 64.35%. The two were measured in different studies on different text.
Who owns ZeroGPT?
ZeroGPT’s terms of use name the operating entity as Olive Works LLC, at 5830 E 2nd St, Ste 7000 #9537, Casper, Wyoming. Its chief executive is Rawad Baroud, named in ZeroGPT’s January 2026 press release announcing its image-detection partnership with TruthScan. The site’s published contact address is support@zerogpt.com.
Is zerogpt.net the same as zerogpt.com?
No. zerogpt.net is a separate site using the same brand name, carrying its own ”© 2026 ZeroGPT” copyright line, naming no operating entity, and advertising a 0.0% false positive rate. It states no relationship to zerogpt.com. Every figure and study on this page concerns zerogpt.com.
Does ZeroGPT work in languages other than English?
ZeroGPT’s FAQ states that “all available languages are supported across all the tools”, and the site interface is available in English, French, German, Spanish, Italian and Portuguese. ZeroGPT does not publish per-language accuracy figures, and the independent studies cited on this page tested English text.