Check before it matters.
See which passages statistical detectors flag — and whether a technical proof of origin also exists. For text, image, audio, video and websites.
Your text
0 words3,000 words per free check
Supported formats
Text
Paste up to 3,000 words — the detector flags standout passages, optionally cross-checked with Pangram.
Website
Enter an address — page text is checked for standout passages, images for provenance signals.
Image
Content Credentials and metadata are read straight from the file.
Audio
Container details, duration and embedded provenance data.
Video
Metadata and embedded credentials from the video container.
How a check works
- 1
Add your content
Paste text, enter a URL, or upload a file.
- 2
The detector and provenance signals are checked
The Wmarks detector scores the text passage by passage, optionally cross-checked with Pangram. In parallel, supported provenance signals such as content credentials are checked.
- 3
See exactly what was found
Flagged passages and any provenance evidence are marked separately, never merged into one number.
More than one number
A single percentage doesn't tell you what to do. Flagged passages do.
Passages, not a percentage
Instead of one number, see exactly which sentences and sections a detector rates as standing out.
Compare detectors
The local Wmarks detector and, optionally, Pangram score independently — agreement between them is more telling than either result alone.
Stay legible
Every flagged spot can be opened and judged in context, instead of trusting a single traffic light.
Built for review, not for punishment
Wmarks is a verification and review tool. It gives students something to check before they submit, and gives teachers evidence to look at instead of a single percentage to trust.
For students
- Check an essay before submitting it.
- See which passages could trigger AI detectors.
- Compare what multiple detectors say.
- See which detectors flagged a passage.
For teachers
- Review the passages a detector marked, rather than trusting one percentage.
- Compare what different detectors conclude.
- Keep statistical detection and technical proof separate when you make a judgment.
Detection is not provenance
Two different questions, two different answers — we never merge them.
AI detection
A statistical detector estimates from patterns like word choice and predictability how likely a text is to be machine-written. That's a probability, not proof.
Provenance verification
Provenance reads a marker deliberately embedded in the content — a watermark, a signed manifest, or a metadata declaration. Only that can actually be verified.
Wmarks shows both, where available — kept separate, never merged into one number.
Pricing
Start free. Buy credits when you need more.
Free
€0.00
To try it out
- 1 anonymous text check per day
- 3,000 words
- All available signal checks
Flex
from €4.99
For individual checks
- Buy credits as needed
- No subscription
- Credits don't expire while your account exists
- Professional Pangram checks use credits based on text length
Pro
€9.99 / month
For regular checking
- 100 credits every month
- Unused credits roll over
- Remaining credits stay usable after cancellation.
How credits work
Professional Pangram checks are priced by text length. You always see the exact credit price before running a check.
- Text check
- 1 credit per 3000 words
- Image
- 1 credit
- Website page
- 2 credits per page
- Audio, per 5 minutes
- 1 credit per 5 minutes
- Video, per minute
- 1 credit per 60 seconds
The professional Pangram cross-check is a separate, optional action priced by word count and shown before you run it — it is not included in the costs above.
Questions
Can you tell me whether a text was written by AI?
No, and neither can anyone else with certainty. We check for provenance signals that a supported system actually embedded. If no such signal exists, the honest answer is that nothing is verifiable — not that a human wrote it.
What is the difference between detection and provenance?
Detection guesses from style and statistics. Provenance reads a marker that was deliberately embedded — a watermark, a signed manifest, or a metadata declaration. Only the second kind can be verified.
Why do some checks say 'Not verifiable'?
Because no public interface exists to verify that vendor's watermark. We would rather say so than show a number that means nothing.
Do you store my content?
Anonymous and opted-out checks are discarded after processing. For signed-in accounts, completed checks are saved for up to 31 days by default so they can be reopened without another charge. Wmarks does not use submitted content to train its own AI models.
Can you remove a watermark or make content undetectable?
No. This product identifies and explains provenance signals. It does not strip them, and it never will.
Does a detected signal mean someone did something wrong?
No. A finding can be a verified technical fact, a declaration, a forensic indication, or a verification limitation. None of those alone establishes misuse or authorship.