Currently, nothing. It’s free to try.

You paste in documents using the text box on the home page. Make sure your documents are formatted cleanly. Copy/pasting text from HTML or PDF documents can produce garbled results if you haven’t made sure the line breaks are correctly separating your paragraphs.

Write us: closereadai@proton.me.

You don’t have to take our word for that. Feel free to test human or AI texts with origins you are certain about before using the detector on uncertain texts you want to evaluate.

It means you submitted a document with five separate paragraphs; the Detector thinks two of those paragraphs were predominantly AI-generated. Remember, that’s a paragraph-level diagnosis only: individual phrases or sentences in an “AI” paragraph might be human in origin, and vice versa.

Of course. See our Tips & Tricks page for advice on how to make the most of repeated submissions.

This happens all the time. The final verdict depends on which voice is more dominant, more often, within a given paragraph. For instance, if an AI rewrites a completed human text, it usually gets flagged by our Detector as “AI.” If a human being uses an AI-generated essay outline, but writes every sentence themselves, it generally reads as “HUMAN” to the Detector.

The position of closeread.ai and its creator is that in most such cases, the AI emerges as the dominant voice, influencing style, structure, and ultimately the meaning of the piece far more than its human users generally acknowledge. Human beings are quick to downplay the influence of generative AI on their writing; if they use it at all, it frequently comes to dictate what and how they write, regardless of how an author, or machine, frames the “assistance” provided by the bot.

The criteria and scoring of each paragraph, for now, has to remain somewhat mysterious to the public at large. We don’t like doing that, but it’s necessary in order to prevent our proprietary code from being stolen or reverse engineered.

What we can say is this: first of all, we intend to create a fully-featured, subscription-based service that provides in-depth, explanatory analyses of every inputted text. That will likely be publicly available by the end of 2026.

We can also say, with confidence, that this Detector does not use the methods that OpenAI, Turnitin.com, and others have tried. Their methodology, as everyone knows, produced such spectacularly bad results that the entire project of distinguishing human from machine writing has now begun to seem, to many writers, educators, and editors, like an impossibility.

Meanwhile, we here at CloseRead.AI know that a reliable AI Detector is possible. We built one.

Not yet, but that’s coming soon. Send us an email if you are interested!

No. It correctly identifies everything from AI-generated poetry, to AI imitations of specific writers, to human writing by James Joyce, Thomas Jefferson, Frederick Douglass, and many others.

We do not currently maintain any in-house records about the people using our site or the writing they submit. However, the software does rely upon AI reasoning frameworks owned by OpenAI and Anthropic. Treat submissions to closeread.ai with the same prudence you would apply before uploading promising ideas, sensitive materials, or confidential documents to those companies’ respective chatbots.