How

A clear signal while you browse. No new tab. No guessing from a screenshot.

How it works

Try the sample. Hover the faint green dot at the upper right of the photo to see the percent, then click it to rate. Hover the underlined text the same way.Tap the faint green dot at the upper right of the photo to see the percent, then tap it again to rate. Tap the underlined text the same way. More green leans human; more red leans AI. The same dots and underlines appear while you browse.

Sample photo of hand-painted ceramic vases used for the DotCheck score demo

Hover the upper-right dot to see the percent, then click to rate.Tap the dot to see the percent. Tap again to rate.

© Charcomphoto

How we do it

The score is made on our servers, not on your machine. Purkyne Gate reads the file first. If the file is quiet, our four models score the picture, the writing, the frames, or the sound. Shared memory returns a score we already have, so the dot is often there as soon as you see the picture.

Stay in the feed with the Chrome Extension, bring a file to Check, or use it in your own app with the Pro API.

Purkyne Gate

AI generators leave a print in the file: studio labels, credentials, origin tags. A studio or platform mark can do the same. Purkyne Gate reads those marks before any model runs. Clear origin locks a lasting 100% AI score.

Cameras and phones leave a print as well. On an original you upload that still carries it, that can lock 0%. It is not a claim that the scene is true. Chat apps and feeds usually remove that print; then the models decide.

Named for Jan Evangelista Purkyne, who first described the ridges of a fingerprint.

Models

Four models. The same on Free, Premium, and Pro. Paying unlocks reports, history, documents, optional Confirm, and the video sound setting.

Vermeer@14.2

Vermeer scores a still image, in your feed and in Check. Named for a painter known for how he treated light and texture.

Valla@14.3

Valla scores writing in English, Chinese, Spanish, French, Portuguese, German, Italian, and Dutch. It scores the writing, not whether the claim is true. Named for the scholar who unmasked a famous forgery by reading the Latin closely.

Muybridge@4.4

Muybridge scores a few frames from across a clip, so a video can carry a score without loading the whole file. Named for the motion study that first broke movement into frames.

Helmholtz@3.1

Helmholtz listens: speech, music, and general sound. On a video with a soundtrack it works beside Muybridge and the two fuse into one percent. A synthetic voice on real footage still counts. So does real speech on generated video. Silent clips use frames only. For sound on its own, drop an audio file into Check or the extension popup.

While you browse, scoring the sound of a video you click or play is a Premium and Pro setting. You can turn it off in Settings.

We publish measured results on Docs. Weights are open on Hugging Face under Apache-2.0.

Reports you can check

Anyone can screenshot a score. A DotCheck Report can be checked by the person who receives it. Premium and Pro can issue lasting Reports and Dossiers on Check. Free can try a SAMPLE on the Reports section of Use.

Pro can use Confirm, an optional second opinion after a DotCheck score, up to 300 times a month. Visual media uses Sightengine. Text uses Winston AI. Free and Premium never send everyday checks to those services.

The DotCheck Index

The DotCheck Index is the average of first scores from the last 30 days: how AI the content we score is, through our eyes. It describes what people check through DotCheck, not the whole internet, and not any one file.

Common questions

Does the Chrome extension read my whole page?

No. It scores the image, video frames, or supported-language text in view. On Premium and Pro, a video you click or play can send about four seconds of audio. We never use your microphone.

Do you save my photos and posts?

No album of your browsing. We keep an anonymous content hash and the score so the same item can return quickly. Uploads and Share renders are temporary.

Do you train your models on my data?

No. Shared memory is a hash and a score, not a training set. New versions learn from datasets we choose and license, not from your uploads or browsing.

Do you sell what I check?

No. We sell DotCheck, not a profile of you. Analysis runs in the Netherlands; the product is controlled from the Czech Republic. Made in the EU. Your rights: Privacy Policy.

Is Free a weaker detector?

No. The models are the same. On Free, browse video is scored from frames. Files you upload get picture and sound on every plan.

Do you always get it right?

No. Some generators are easier to spot than others. Purkyne Gate can be fooled. We publish how we measure on Docs.

Plans

Free to start. Premium for reports you can hand over, history, documents, and the video sound setting. Pro for Confirm and API keys. To cancel: Check → Account → Cancel subscription (Stripe Billing). See full plan details.

Straight talk

A score is a signal, not a verdict. It is not always perfect. Some generated photos, clips, and writing we know well. For kinds we have not trained on as much, the number can miss. Use your own judgment.