All guides
Guide

Uploading your manuscript to AI: what happens to your text

4 min read

When you paste a chapter of your manuscript into an AI chatbot or an online translator, the text goes to the company's servers, and what happens to it next depends on the tool and on your settings. In some tools, on the free version or with the default settings, the text can be used to train the model, and sometimes people at the company read some of the conversations. In almost every tool you can cut that down. Before uploading your manuscript to AI, it's worth spending five minutes on the settings.

Why is this even a question

An unpublished manuscript is something you want to control. You don't want passages from it floating around early, and you're not sure you want it used to train someone else's machine.

Two different things can happen to text you upload. First, the company uses it to improve its model, meaning it becomes part of the material the machine learns from. Second, employees or reviewers at the company read certain conversations, for example to check quality or safety. The answer is different for every tool, so you need to check.

What the privacy pages say

We read the pages of four common tools. This is what they say, as of early October 2026:

ChatGPT. According to OpenAI's data controls page, there's a setting called Improve the model for everyone. When it's turned off, new conversations won't be used to train the models. A temporary chat isn't used to improve the models, but it may be kept for up to 30 days for safety reasons. On managed business accounts, the default is that content isn't used for training.

Claude. According to Anthropic's privacy center, on personal accounts conversations are used to improve the models if you chose to allow it. In its announcement about the updated terms, the company says the change doesn't apply to its business services.

Gemini. In the Gemini privacy hub, Google says some conversations are read by human reviewers, and asks you not to enter confidential information you wouldn't want a reviewer to see. Even with history turned off, conversations are kept in your account for 72 hours.

DeepL. According to DeepL's privacy policy, in the free version the text and translation are processed for a limited period to train and improve its systems. In the Pro version, texts aren't stored permanently, are deleted at the end of the job and aren't used to improve the service.

The pattern that keeps coming up

Look at the four tools together and a clear pattern shows up:

Personal or free version Business or paid professional version
Training on your text Depends on a setting, and sometimes yes Usually not, by default
What you need to do Check and change the settings Make sure you're actually on that version

The big difference isn't between the companies. It's between free and business, and between people who checked their settings and people who didn't.

Uploading your manuscript to AI: five steps

  1. Read the tool's privacy page. Not all the terms, just the part about using content for training.
  2. Turn off training if there's an option for it. It's usually in the settings, under privacy or data.
  3. Prefer a temporary chat when you're just testing something.
  4. Don't upload the whole book when a passage will do. Checking how one dialogue sounds doesn't take a hundred pages.
  5. When you're paying for a translation service, ask. What happens to the file after the job, how long it's kept, and whether it's used for training. A serious service will answer that in writing.

What this means when you choose a translation service

If you're thinking about translating your book with an AI tool, whether a general chatbot or a dedicated service, the privacy question is part of the comparison, not a footnote. We wrote about comparing offers in how to compare two translation quotes, and about the difference between a general chatbot and translating a whole book in translating a book with ChatGPT?.

One last note: policies like these change. The Gemini privacy page, for example, was updated at the end of September 2026. Before you upload something that matters to you, look at the current page. More questions are collected in our frequently asked questions.

Frequently asked questions

Does ChatGPT use the text I paste in to learn?

On a personal account, it depends on a setting called Improve the model for everyone. According to OpenAI, when it's turned off, new conversations won't be used to train the models. On managed business accounts, the default is that content isn't used for training.

What about Claude and Gemini?

With Claude, according to Anthropic, conversations on a personal account are used to improve the models only if you chose to allow it. With Gemini, Google says some conversations are read by human reviewers, and asks you not to enter confidential information you wouldn't want a reviewer to see.

Does a free translator like DeepL keep the text?

According to DeepL's privacy policy, in the free version the text and translation are kept for a limited period for training and improvement. In the Pro version, texts are deleted at the end of the job and aren't used to improve the service.

What's the simplest thing to do before uploading a chapter?

Check the tool's privacy settings, turn off the option that uses conversations for training if there is one, and prefer a temporary chat or a paid version that commits not to use your text. And don't upload the whole book when a short passage will do.