Is my writing app training AI on my novel?

What Google and Microsoft have said

Both companies answer this on their own help sites, and both say no, with conditions worth reading closely.

Google's Docs Editors Help page on how Gemini protects your data sets out commitments it says apply to all Workspace users, a term that on that page includes people using Docs with a personal Google account. It says Google does not use your Workspace data to train or improve the generative AI models behind Gemini, Search, and other systems outside Workspace without permission. Two qualifications follow. Anonymized or aggregated interactions, like accepting a spelling suggestion, may improve features such as spell check. And if a personal account chooses to share Workspace data with Gemini Apps or certain Search features, their terms apply instead, and under them the data may be used for training. For work and school accounts, the Workspace privacy hub for administrators says content is not used to train generative AI models outside your domain without permission.

Microsoft's Copilot privacy documentation on Microsoft Learn says personal data collected through connected experiences in Microsoft 365 is not used to train large language models, including the ones behind Copilot. Connected experiences were the setting at the center of an alarm in late 2024. For home plans like Microsoft 365 Personal and Family, its support page on Copilot privacy says your prompts, responses, and file contents are not used to train foundation models when you use Copilot in the apps. For work and school accounts, the same Copilot documentation says it of prompts, responses, and data reached through Microsoft Graph.

Those are statements of current policy, and policies are revised. Read the terms for the product and the kind of account you have, since personal, work, and school accounts are covered by different documents. Look for the words train, improve, and models. A company's own help pages carry more weight than a screenshot passed around social media, in either direction.

Checking your settings

• Word for Windows: File › Account › Account Privacy › Manage Settings. The connected experiences switches are here. Turning off the optional ones also turns off features that rely on them.
• Word for Mac: the same switches sit under Word › Preferences (or Settings) › Privacy.
• Google Docs: for a personal account, Google Workspace smart features on the General tab of Gmail's settings governs whether your Workspace data powers smart features across Google's Workspace apps. It is a feature switch rather than a training switch.
• Any AI feature you switch on: an assistant inside a word processor may have its own terms, separate from the editor's. Check those as well.

If your version shows these settings somewhere else, the product's help center will have the current path.

In Tabularia

Tabularia's privacy policy says it does not use manuscript content to train AI models. There is no assistant built into the app. Tabularia does not pass your writing to an outside AI company on its own behalf.

What you can do is connect an assistant you already use, like Claude or ChatGPT. It takes a Cloud subscription. The assistant can then read your manuscript and propose changes, most of which wait for you to accept or reject. Once you do, that assistant's settings decide whether it retains your pages or learns from them, not Tabularia's. Tabularia can neither see that setting nor change it. The guide's advice is to switch off training in the assistant before connecting, since consumer plans often start with it on; the switch is often labeled along the lines of Help improve our models. You can end the connection at any time from your account, and the FAQ sets out what passes through Tabularia while it is live.