Bulk AI Detector
Drop in a folder of documents and get one row per file, sorted most AI-like first. Every file is read and scored on this device.
- Free Forever
- No Sign Up Required
- Unlimited Usage
- Your Text Never Leaves Your Browser
How bulk scanning works
- 1
Drop the whole folder
Text, Markdown, PDF, Word, CSV and Excel files, up to 200 at a time. Each one is read inside this browser tab — no file is uploaded, and there is no server that could receive one.
- 2
Every document gets a quick scan
A quick scan reads the document in wide overlapping passages and produces an overall score, a verdict and a confidence rating. It skips the passage breakdown, and is not what a list of thirty documents is for.
- 3
The list sorts itself, most AI-like first
So the two or three worth reading closely are at the top. Documents that could not be read, or that were too short to judge, stay in the list with the reason rather than quietly disappearing.
- 4
Open any row for the full picture
Opening a document runs the full scan on that one file and shows the ordinary editor with the passage heat map, the passage breakdown and the writing-style signals.
How long it takes
It depends on the machine, because the machine is doing the work. The detector is a 126 MB model that downloads once and then runs on your own CPU. A thousand-word essay is a handful of model passes on a quick scan and roughly seven times that on a full one.
On a desktop with memory to spare, several documents are scanned at once — each in its own worker, each holding its own copy of the model. On phones, tablets, low-memory machines and metered connections it stays at one, because the alternative is a tab that runs out of memory halfway through. You can stop a run at any point; it takes effect within a second or two, and everything already finished is kept.
What leaves your device
Nothing. The model comes to the browser rather than the documents going to a server, which is what makes this free and what makes it safe to run over other people’s writing. Batch results are held in the tab and are not written to disk: the single-document tool saves your own drafts because they are yours, but a batch is usually somebody else’s work, and storing that uninvited is a different matter. Closing the tab discards the run.
Exports are built in the browser too. The CSV and Excel files carry the accuracy and language warnings as part of the file, so a score cannot be forwarded without them.
Before you use this on students
A ranked list is a seductive object. It invites you to read the top row as a finding, and it is not one: these are probabilities from a model that is wrong in both directions, most often on short work, on technical or formulaic writing, and on writing by people whose first language is not English. Scanning thirty essays at once multiplies the chance that at least one score is badly wrong.
The model was also trained on English alone. Documents that do not look like English are flagged in the list and in the export, and their scores should be ignored entirely rather than discounted.
Use a high score to decide what to read closely, never as the finding itself. Look at drafts and version history, ask the writer about their process, and weigh the context.