AI Detector

Is this text AI-generated?

A zero-shot detector suite with an honest fused verdict.

Detectors
Operating preset

Frequently asked questions

Can you tell me for certain whether something was AI-generated?
No. Detection is probabilistic, not proof. Every AI-text detector produces false positives (human text flagged as AI) and false negatives (AI text that looks human, especially after editing). No tool can certify authorship, and this one does not claim to.
What is a zero-shot detector?
A detector that does not need fine-tuning on one specific model's output. We score a passage with an open language model — likelihood, entropy, log-rank and perplexity plus the DetectGPT, Fast-DetectGPT, Binoculars and DetectLLM methods — then fuse them into a single verdict. Because it is zero-shot, it generalizes across models instead of memorizing one.
Why don't the detectors all agree? Which should I trust?
The suite is intentionally diverse: a statistical baseline plus several zero-shot methods. They disagree because they measure different signals and have different false-positive rates. The fusion strategy (weighted mean, vote or rank) combines them into one score, but a fused score is still an estimate — not ground truth. We show each detector's raw value so you can see where the agreement and disagreement come from.
What does the low-FPR preset do?
It biases the suite toward fewer false positives: it raises the bar before calling text AI, so it is more conservative about flagging human-written work. Balanced trades some of that caution for sensitivity. Choose low-FPR when a false accusation would be costly.
What is a false positive, and how common is it?
A false positive is human-written text flagged as AI. Zero-shot detectors can trip on careful, formal or highly-polished prose, non-native writing, or text on a topic the scoring model is unusually confident about. The low-FPR preset reduces but cannot eliminate this.
Is a detector verdict enough to accuse someone?
No. A detector output must never be the sole evidence of AI use. Academic-integrity officers and publishers should treat it as a screening signal and confirm with context, drafts, version history or a human review. Using a probability as proof of misconduct is unsafe and can carry legal consequences.
How should I interpret a score near 50%?
P(AI) runs from 0 (almost certainly human) to 1 (almost certainly AI). Values close to 0.5 mean the detectors are divided or uncertain — treat that as indeterminate rather than a coin flip. We show the per-detector breakdown so you can see why the verdict is uncertain.
Can a detector be fooled?
Yes. Heavily edited AI text, AI text in a non-native style, or text deliberately rewritten to defeat detectors can be misclassified. Detectors also drift as new models ship. This is an active research area with no perfect solution.
What about the EU AI Act and academic policy?
The EU AI Act (Article 50) and similar transparency rules may require you to disclose AI-generated content when you publish. This tool is a screening aid, not a compliance or certification service — you remain responsible for meeting disclosure obligations and for following your institution's academic-integrity rules.
Is my text stored or shared?
No. Your text is sent to the detection engine, scored, and discarded. We do not retain, train on or share your content. We keep only your email (for the daily quota) and an anonymous usage counter.
Is it free?
Yes — the core tool is free with a daily quota (3 detections per day for signed-in users). Paid plans raise the quota and the per-detection length limit. There is no content retention and no training on your text.

Honest limits

AI Detector reports probabilities, not proof. AI-text detection is error-prone: it produces false positives on careful human writing and false negatives on edited AI text, so no verdict here is evidence of authorship and it must never be used as the sole basis for an accusation, discipline decision or automated rejection. Treat the output as a screening signal and confirm with context, drafts or human review before acting on it.

You are responsible for how you use the output and for meeting your own obligations, including AI-disclosure rules such as the EU AI Act (Article 50) and your institution's academic-integrity policy. By using this tool you agree to use it responsibly; we disclaim liability for misuse, including unfounded authorship claims. See the full Disclaimer in the Terms of Service.

Unlock your result

Sign in to see your detection result. 3 free detections every day. No content is stored or tracked.

or

Accept the Terms of Service

To use the Service, you must review and accept our Terms of Service, including the Disclaimer. You are responsible for how you use the results and for complying with applicable laws, academic policies and platform rules.

We record when you accept so you never have to be asked again. See the Privacy Policy.

Read the Terms

Settings

Profile
Profile picture
JPEG, PNG, WebP or GIF up to 2 MB.
Name
Shown in the header and on shared results.
Password
At least 8 characters.
Email
Your sign-in address.
Signed in
This browser session stays active until you sign out.
Account
Delete account
Stops your subscription, deletes your API keys and signs you out everywhere. Your email stays on record.
Appearance

System follows your device. Light and dark apply everywhere immediately.

Plan
API keys

Keys authenticate the /v1 API and the MCP server. A key is shown once, at creation.

Pay as you go
Topped-up balance
Credits cover /v1 and MCP requests.
$0.00
Top up
Checkout is handled by Stripe.
Low-balance alert
Warn me when the balance drops below this amount.
Auto top-up
Off.
Referral and earnings Loading…

Payout threshold: 25 USD. Full partner hub.

Your link