Authors have strong feelings about this, and reasonably so: surveys of working writers show the large majority believe their consent should be required before their work trains an AI model, and most say they'd want compensation if it did. What's less consistent is how many writers actually check whether the tools they use every day already do this by default.
Three things worth checking in any AI writing tool's terms
- Who owns what you create. A legitimate tool transfers full rights to whatever you write using it, to you. This should be stated plainly, not buried in ambiguous language.
- Whether your text trains the model. This is the core question. Some tools use your input to improve their models by default; others contractually don't, sometimes described as zero data retention.
- Where your text is stored. Whether it's kept indefinitely, for a limited window, or not retained beyond the immediate request matters for how much exposure your unpublished manuscript actually has over time.
The tier your tool falls into matters
Consumer-grade AI tools, often free ones, commonly train on user input by default, requiring you to find and use an opt-out setting if one exists. Paid, professional-tier tools more often build a contractual prohibition against training on user content directly into their terms. The practical problem is that this distinction isn't obvious from the outside; a tool doesn't announce which tier it falls into, and the honest answer usually requires actually reading the terms of service rather than assuming based on price or reputation alone.
The simplest way to remove the question
A tool that doesn't use AI at all doesn't have this question to answer. Quillen's Analyze Document feature, its prose self-editing tool, flags filter words, passive voice, vague words, and six other categories of common weaknesses using rules, not a language model, and nothing you write is sent anywhere; the analysis runs entirely on your own computer. There's no terms-of-service clause to check here, because there's no AI in the loop to have trained on anything in the first place.
Common questions
Do all AI writing tools train on what you type?
No, but the default varies by tier. Consumer-grade, typically free, AI tools commonly train on user input by default unless you manually opt out. Paid, professional-tier tools more often contractually prohibit training on user content and use what's sometimes called zero data retention. The difference is rarely obvious without checking.
How do I check if a specific writing app trains AI on my text?
Search its terms of service or privacy policy for the words "training" or "model improvement." Reputable tools state their policy clearly, either that they don't use your content for training or that they do by default with an opt-out. If you can't find a clear answer after a few minutes of looking, that absence is itself worth treating as a answer.