DetectGPT logo
Content Authenticity Detection · Academic Research

DetectGPT

DetectGPT is a web and mobile tool that scores how likely a text was written by AI, checks it for plagiarism, and can rewrite flagged passages. Aimed at teachers, students and writers, it starts with a 3,000-word free trial.

Active Free trial Subscription No public API 18+ Verified by Guidaio
Overview

What is DetectGPT?

DetectGPT is a content-authenticity tool sold as two halves of the same subscription. The first half is a detector: you paste up to 5,000 characters, or upload a file, and it returns the probability that the text was machine-generated, with a separate share for passages it reads as mixed. The homepage illustrates this with a sample verdict of 80% Probability AI Generated and 15% mixed. The publisher says the model was trained on billions of words and covers any AI model, naming ChatGPT, GPT-4 and Gemini. Around the detector sit a web-wide plagiarism checker, fact checking, a readability score, PDF report export and an Original Content Certification: a shareable analysis link a freelancer can hand to a client as evidence of originality. Detection of AI-generated images is announced but carries a Coming Soon badge.

The second half is the AI Humanizer, on its own page, which rewrites AI-written text so that it stops being flagged. Its promise is unusually explicit: it claims to defeat all AI detectors, including DetectGPT's own. The same three plans buy both tools, so one subscription pays for the test and for the way around it.

Pricing is short and public: Essential at 15 USD a month for 50,000 words and one seat, Pro at 29 USD for 250,000 words and three members, Team at 59 USD for 500,000 words and unlimited members. Every plan carries the same feature list; only quota, seats and batch size change. A yearly toggle advertises six months free, and a 3,000-word trial needs no payment card.

Two caveats belong in any description. Accuracy is claimed at 95% in the page metadata and 99% in the feature block, with no methodology or false-positive rate published anywhere. And an API is advertised on two pages, yet every link points to a dead anchor and no documentation exists. The site itself is tiny: five URLs, no about page, no blog and no social accounts.

What it does

  • Score a text for the probability that it was generated by AI
  • Check a document for plagiarism against content published on the web
  • Rewrite AI-generated text with the AI Humanizer so it is no longer flagged
  • Scan files in batches of 10 to 50, depending on the plan
  • Export a PDF report of an analysis
  • Issue an Original Content Certification and share the analysis link with a client
  • Run checks on the move from the iOS or Android app
Audience

When to use DetectGPT / When not to

A quick filter to help you decide if DetectGPT is the right fit.

When to use DetectGPT

  • Teachers and lecturers checking whether a submitted assignment was written by a student or by a chatbot
  • Students who want to screen their own draft before handing it in
  • Freelance writers and agencies who need a shareable originality certificate for a client
  • Editors and newsrooms vetting text received from outside contributors
  • Recruiters reading application letters and written tests at volume

When not to use DetectGPT

  • Anyone needing a defensible verdict: the site advertises 95% and 99% accuracy on the same page, with no methodology or false-positive rate
  • Teams that require a Data Processing Agreement, an EU Article 27 representative or a security certification such as SOC 2
  • Developers wanting to automate detection: the API is announced but no documentation is published anywhere
  • Buyers who expect to be able to get their money back, since the policy rules out refunds in every case
  • Users looking for a permanent free tier, or for AI image detection, which is still badged Coming Soon
Get started

How to use DetectGPT

A typical end-to-end flow, from setup to results.

  1. Open the homepage, where the scanner sits in plain view and can be tried before signing up
  2. Create an account with an email address and password, or sign up with Google
  3. Claim the 3,000-word free trial; no payment card is requested
  4. Paste your text into the box, watching the 5,000-character counter, or click Upload File
  5. Press Scan Text, which also counts as accepting the Terms of Service
  6. Read the verdict as a percentage of AI-generated content, with a separate mixed share
  7. Switch to the AI Humanizer page and press Humanize to rewrite a flagged passage
  8. Export the analysis as a PDF report, or issue an Original Content Certification to share
  9. Queue several files at once if your plan includes batch detection
  10. Manage or cancel the subscription from the account settings at any time
Quick read

Pros & Cons

Pros

  • Detection and rewriting bundled in one subscription, with no add-on to buy
  • A 3,000-word trial with no payment card, so the tool can be judged before paying
  • Simple public pricing: three tiers, amounts shown in full, from 15 USD a month
  • Generous monthly quotas for the price, up to 500,000 words on the top tier
  • Unlimited seats on the 59 USD Team plan, which is rare at that price
  • Batch processing, PDF reports and a shareable certificate fit real marking and delivery workflows
  • The privacy policy names its subprocessors and caps retention at three months after account closure

Cons

  • Accuracy is advertised as both 95% and 99% on the same homepage, with no methodology or false-positive rate
  • Selling the detector and the humanizer designed to defeat it is a structural conflict of interest
  • No refunds under any circumstances, including cancellation part-way through a paid term
  • No permanent free plan, only a one-off 3,000-word trial
  • The API is promoted on two pages but has no documentation and no working link
  • The legal entity differs between documents: AI Stack LLC in the terms, DetectGPT LLC in the privacy policy
  • One shared email address for support, legal and privacy, no contact form, no named leadership, and legal pages unchanged since December 2023
Pricing

Pricing & Plans

There is no permanent free plan. The service is sold as a subscription with three monthly tiers, the cheapest being Essential at 15.00 USD per month. A free trial of 3,000 words is available without a payment card, and annual billing is advertised as including six months free, although the annual amounts are not published on the page. All payments are made in US dollars, sales tax may be added, and no refunds are granted in any circumstance.

Plan 1
Essential
  • 15 USD per month - 50
  • 000 words per month
  • 1 user seat
  • premium AI detection models
  • batch file detection (10 files)
  • PDF report generation
  • AI Content Humanizer
  • plagiarism checker
Plan 3
Team
  • 59 USD per month - 500
  • 000 words per month
  • unlimited team members
  • premium AI detection models
  • batch file detection (50 files)
  • PDF report generation
  • AI Content Humanizer
  • plagiarism checker
Plan 4
Yearly billing
  • advertised as six months free
  • the annual amounts are computed in the browser and do not appear in the published page
Plan 5
Free trial
  • 3
  • 000 words
  • no payment card required
  • a trial
  • not a permanent free tier
Special offers — Free trial of 3,000 words, with no payment card required · Annual billing advertised as six months free, though the annual amounts are not published · Affiliate programme open through Tolt, with commission terms not disclosed · No student, academic or hardship discount is published, despite teachers and students being the stated audience
Prices and plans listed above may evolve. Always check the official pricing page before subscribing.
Trust & Privacy

Data, GDPR & hosting

A consolidated view of how DetectGPT handles your data.

GDPR overview

Implementation is partial and never named as such: the words GDPR and General Data Protection Regulation appear nowhere on the site. The privacy policy nevertheless grants residents of the EEA, the UK and Canada the right to access and obtain a copy of their data, to rectification or erasure, to restriction of processing, to portability and, in some circumstances, to object. It links to the list of European supervisory authorities and to the Swiss regulator, allows consent to be withdrawn at any time, and promises a reply to access, correction or deletion requests within thirty days. What is missing is just as telling: no legal basis for processing is stated, no Article 27 EU representative is designated, no DPO is named, no Data Processing Agreement is offered and no transfer safeguards are described. The policy has not been updated since 10 December 2023.

Who owns the data?

The terms draw a line between two kinds of content. Contributions, meaning anything you post, stay yours: AI Stack LLC states it asserts no ownership and that you keep full intellectual property rights. In the same breath, posting a Contribution grants the company a worldwide, perpetual, irrevocable, transferable and sub-licensable royalty-free licence to use it for any purpose, including commercial and advertising, with your moral rights waived. Submissions, meaning feedback, ideas or suggestions, are treated differently and become the company's sole property, without compensation or attribution. Nothing in either document addresses the status of the text you paste into the scanner itself.

Reuse rights

Nothing in the terms restricts what you do with what the tool gives back. No clause claims rights over a detection score, a PDF report or a humanised text, so you may reuse them freely, including commercially, without asking permission. On the company's side the privacy policy sets out broad purposes: operating, maintaining and improving the service, analytics, personalisation, support, measuring marketing campaigns and meeting legal obligations. Personal data may reach the named service providers Open AI, Supabase, Cloudflare, GitHub and ScraperAPI, plus Stripe as merchant of record for payments, and could be transferred if the business is sold or merged. Do Not Track signals are not honoured. The site never says whether submitted text is used to train models.

Data retention & training

Retention summary
The privacy policy sets one clear ceiling: no purpose it describes justifies keeping personal data more than three months after an account is closed, unless a law on tax, accounting or similar grounds requires longer. Once there is no legitimate need, data is deleted or anonymised; where that is impossible because it sits in backup archives, it is stored securely and walled off from further processing until deletion becomes possible. Asking to close an account triggers deactivation or deletion from the active databases, though some records may be kept to prevent fraud, resolve disputes, assist investigations or enforce the terms. Requests to review, correct or delete data go through a form and are answered within thirty days. No separate retention period is stated for the text submitted for analysis.
Subprocessors disclosed
Yes
GDPR contact

Hosting summary

The terms of service state plainly that the site is hosted in the United States, and the privacy policy adds nothing more precise: no region, no data centre, no residency option and no European hosting alternative. Users outside the United States should assume their data crosses the Atlantic, and the documents describe no transfer safeguards, no standard contractual clauses and no Article 27 representative to address in Europe. The named service providers give only a partial picture of where processing happens: Supabase for the database, Cloudflare in front of the site, GitHub, ScraperAPI and Open AI, with Stripe as merchant of record for payments; none has a stated region. The domain resolves to a Cloudflare anycast address, which reveals the delivery network rather than the location of the servers themselves, so it cannot be used to infer a hosting country. In short, one jurisdiction is stated, the United States, and everything below that level is undocumented.

Hosting countries
🇺🇸 United States
Watch-outs

Things to keep in mind

Risks and trade-offs to weigh before adopting DetectGPT.

  • A probability score is not proof: with no published false-positive rate, an accusation of cheating built on this number can harm an innocent student or contributor
  • The vendor sells both the detector and the humanizer designed to beat it, so neither result should be read as a neutral verdict
  • Leaning on automated detection can erode the habit of reading closely and judging a piece of writing on its merits
  • Pasting confidential material, exam papers, client drafts or unpublished manuscripts sends it to a third party whose named subprocessors include Open AI
  • The humanizer invites the opposite misuse: passing machine-written work off as your own, which is a serious breach in academic and professional settings
  • Two company names, a shared registered-agent address and no identified leadership make recourse difficult if something goes wrong
  • Legal documents frozen since December 2023 may no longer describe the product as it works today, and no refund is possible if it disappoints
Setup

Setup & Integrations

Technical difficulty

Very easy, with nothing to install. The scanner sits on the homepage and can be tried before signing up; an account takes an email address and password or a single click with Google, and no payment card is needed for the trial. Using it means pasting text, minding the 5,000-character limit per pass, or uploading a file, then pressing one button. Mobile apps cover iOS and Android. The only setup that is not feasible is the programmatic one: the advertised API has no published documentation, so an integration cannot realistically be attempted.

Deployment

Web appMobile appIOS appAndroid app

Apps stores

Company

Behind DetectGPT

Company name
AI Stack LLC
Founded
20/12/2022
Country of origin
🇺🇸 United States
Headquarters
16192 Coastal Highway, Lewes, DE 19958, USA
UBO
INFORMATION_NOT_FOUND
UBO country
INFORMATION_NOT_FOUND
Domain registrar country
🇺🇸 United States
Legal contact
Support contact
Official links

Resources

All the official URLs gathered for verification and reference.

FAQ

Frequently asked questions

What does DetectGPT actually do?
It scores how likely a text was generated by AI, checks it for plagiarism against web content, and offers an AI Humanizer that rewrites flagged text. Reports can be exported as PDF and shared as an originality certificate.
How accurate is it?
The site claims 95% accuracy in its page description and 99% in its feature block, on the same homepage. No methodology, false-positive rate or independent evaluation is published, so the figure should be treated as a marketing claim rather than a measurement.
Is there a free plan?
No. There is a free trial of 3,000 words that does not require a payment card, but no permanent free tier. After the trial, use requires one of the three paid subscriptions.
What does it cost?
Three monthly plans: Essential at 15 USD, Pro at 29 USD and Team at 59 USD, giving 50,000, 250,000 and 500,000 words per month respectively. An annual option is advertised as including six months free, but the annual amounts are not published.
Can I get a refund?
No. The refund policy states that no refunds are offered on any transaction, including when a subscription is cancelled part-way through a term. You keep access until the end of the period you have already paid for.
Is there an API?
The site advertises one on both the homepage and the humanizer page, but every Learn More link points to a dead anchor, no documentation exists and no developer subdomain resolves. In practice, integration is not currently possible.
How long is my data kept?
The privacy policy states that no purpose requires keeping personal data more than three months after an account is terminated, unless the law requires longer. Data is then deleted or anonymised, and backup copies are isolated until deletion is possible.
Where is my data hosted, and who else sees it?
The terms state the site is hosted in the United States. The privacy policy names Open AI, Supabase, Cloudflare, GitHub and ScraperAPI as service providers, with Stripe as merchant of record for payments.
Is there a minimum age?
Yes, 18. The terms state the site is intended for users aged at least 18 and that under-18s may not register, and the privacy policy repeats that it does not knowingly collect data from minors.
Does it detect AI-generated images?
Not yet. Image detection appears on the homepage as a feature card but carries a Coming Soon badge, so only text detection, plagiarism checking and humanising are available today.
Conclusion

Should you pick DetectGPT?

DetectGPT does something people genuinely need, at a price that makes it easy to try: paste a text, get a probability that a machine wrote it, and hand a client a shareable certificate if it comes back clean. Three flat tiers from 15 USD a month, quotas that are generous for the money, batch processing and PDF reports make it usable inside a real marking or delivery workflow, and a 3,000-word trial costs nothing but an email address.

The reservations are structural rather than cosmetic. The same subscription sells the detector and the AI Humanizer built to defeat it, including DetectGPT's own detection, which puts the product on both sides of the problem it describes. Accuracy is given as 95% in the page metadata and 99% a few blocks below, and neither figure is backed by a published methodology or false-positive rate. That matters, because the buyers named on the homepage are teachers, and a percentage without an error rate is not evidence about a student.

Around the product, the paperwork is thin. Two different company names appear across the legal pages, the postal address is a shared Delaware registered-agent address, no founder is named anywhere, a single email address serves support, legal and privacy questions, and the terms and privacy policy have not been touched since December 2023. There is no DPA, no Article 27 representative, no security certification, and the advertised API has no documentation at all.

Treat it as a fast, inexpensive first-pass indicator, useful for spotting text that deserves a closer human look. Do not treat its score as proof, and do not build a disciplinary or hiring decision on it. Anything consequential should be verified by other means.