Every writing-tool homepage has "we don't train on your work" somewhere near the privacy promises. Few of them survive the fine print. The phrase is doing a lot of quiet work, and most of the time it's narrower than it sounds. Here is what it actually means, what to watch for, and exactly how WriteLoom handles your manuscript.

What "training" actually means

"Training" has a specific technical meaning: using your text to update a model's weights, so that what you wrote becomes part of what the model knows and can reproduce. That is the thing writers are rightly afraid of — your unpublished manuscript leaking into a model that then echoes your premise, your phrasing, or your world to someone else.

The trouble is that "we don't train on your work" only promises that one specific thing. It says nothing about a long list of other uses that can be just as consequential: how long your text is retained, whether humans review it, whether it's logged for "quality" or "safety," and whether it's used for the famously elastic phrase "improving our services." A company can honestly say it doesn't train on your work while still keeping your manuscript for years, letting staff read it, and feeding it into analytics. The headline is true and the picture is incomplete.

The weasel words to watch for

A few phrases should make you read more carefully, not less:

"We may use your data to improve our services." This is the catch-all. It is broad enough to cover almost anything, including uses that look a lot like training even if they aren't called that.

Opt-out instead of opt-in. If the privacy-protective setting is off by default and buried in a menu, the default tells you what the company actually wants.

Silence about sub-processors. Most tools send your text to third-party AI vendors. If a policy talks confidently about its own practices but says nothing about what those vendors are contractually allowed to do with your text, the promise has a hole in it exactly where the risk is.

"De-identified" or "aggregated." Sometimes meaningful, sometimes a euphemism for "we kept it, just without your name attached." Prose isn't a spreadsheet; a manuscript is identifiable by its content.

How WriteLoom actually handles your manuscript

Here is the concrete version, which is the only kind worth trusting.

Routing. AI features send only the context a given task needs to our AI vendors, and every vendor we use is bound by a written agreement that prohibits training on the content we send and prohibits retaining it beyond the request. The text goes to the vendor to do the thing you asked, and that's the end of its life there. No human-review pipeline, no "service improvement" reuse.

Storage. Your manuscript and every other artifact live in your project, stored so we can show it back to you and run the features you trigger — and nothing else. We don't mine it, build datasets from it, or use it for marketing. The full privacy policy names which vendor sees what, and why, rather than waving at "third parties."

Portability. Everything you put in is exportable to a standard format at any time. You are never holding your own book hostage inside our walls; if you want to leave, you take the manuscript, the cover, the comp set, and the rest with you.

Deletion. When you delete content or your account, it comes out of live systems immediately and out of backups within 30 days. Deletion means deletion, not "hidden but retained."

The one exception, named plainly

There is exactly one place your text leaves this arrangement, and we'd rather say it out loud than bury it. Audiobook narration runs on your own ElevenLabs API key, because ElevenLabs charges by the character of speech generated. When you generate audio, the text you narrate goes to ElevenLabs under your account and their terms, not ours. We flag this in the product at the point you'd use it, so the choice is yours and informed.

The test

The honest test for any tool's privacy promise isn't the headline on the homepage. It's whether the policy will tell you, specifically: which vendors see your text, what they're allowed to do with it, how long anything is kept, and what deletion actually removes. A company that's comfortable answering those four questions plainly is telling you something real. One that retreats to "we don't train on your work" and stops there is telling you something too — just not the part you needed to hear.