- Training on user conversations is a provider policy, and policies vary.
- Consumer tiers are more likely to use conversations than paid business tiers.
- Opt-outs commonly exist but are not usually enabled by default.
- Training on your text is different from your text appearing in someone’s book.
- A tool that states clearly that it never trains on your work removes the question.
Some might, depending on which tool you use and which plan you are on. It is a policy setting rather than an inherent property of AI. Consumer tiers more often use conversations for model improvement; business and API tiers typically do not. Check your provider’s current terms and your account settings, because both change.
Chapter i·Why it matters
The fear here is usually that a model will regurgitate your plot to another writer, which is not really how it works. The practical concern is simpler and more reasonable: you did not agree to hand over an unpublished book, and you should know whether you have.
Chapter ii·What to include
- Which tools touch your manuscript, including browser extensions and plugins.
- The current policy for each, per plan, read at the source.
- Whether an opt-out exists and whether it is switched on.
- A rule for collaborators, so co-authors are not on different settings.
- A note of when you last checked, since terms are revised.
Chapter iii·Example
A small press asks its four authors which AI tools touched their manuscripts. Two have opt-outs enabled, one is on a business tier, and one has never opened the settings page. The press writes a one-paragraph policy, and the question stops recurring.
In WriteLoom
Everything this article talks about, in one project
WriteLoom never trains on your manuscript. Your book stays yours, and it leaves with you in any format you like, whenever you want.
Not ready for a workspace yet? Get the free six-part series on winning the writing game: finishing drafts, editing in the right order, querying with real math, and launching on a runway.