You pay for human-written data because it makes better models. ScribeLens checks each piece of text for AI writing patterns before it reaches your dataset, and shows your reviewers the exact sentences behind every result.
Fine-tuning sets, preference data and paid writing tasks are only as good as the people who write them. When a contributor pastes in assistant output, it can pass a quick read and end up in training. ScribeLens flags those submissions as they arrive, so your team reviews them before they are used.
Your reviewers don't have to reread a whole submission. They go straight to the flagged sentences and see why each one was flagged. When a contributor has edited AI output, the flags gather around the passages that came from the assistant, which is exactly where a reviewer needs to look.
We retrain the detector as AI writing tools change. Detector v4.0 was retrained in October 2026, and in our own testing it is markedly better at detecting text written with current assistants. Every result names the detector version that scored it, so your records show exactly how each submission was checked.
Text sent to the API is analysed and the result is returned. The text itself is not stored; only the word count is recorded, for your usage. API keys are issued with your agreement and use a single header to authenticate.
API access comes with a monthly allowance agreed for your volume. We can set up evaluation access first, so your team can run ScribeLens on your own data before you commit.
Any written English text your contributors produce: essays, answers, explanations, rewrites and preference responses. Send whole submissions to the main endpoint, or single responses to the sentence endpoint.
ScribeLens detects the writing patterns that AI assistants share, so one check covers text from any of them. It shows where AI-like patterns appear, not which product wrote them.
Each flagged submission comes with the sentences and reasons behind it. That gives your reviewers something specific to check and your contributors something specific to respond to.
Yes. We arrange evaluation access so you can run it on your own material. Tell us what you want to test and roughly how much text a month you expect.
Call the usage endpoint at any time. It returns the words used and remaining this month and the request count, and it does not use up any of your allowance.
Tell us about your data and your monthly volume, and we will set up evaluation access for your team.