Most journalists didn’t really choose where they drafted their work. They chose what worked. What was fast. What editors already used. What made collaboration easy. Google Docs became the default not because anyone loved it, but because it solved a lot of practical problems at once. It synced everywhere. It handled comments well. It felt frictionless. And for years, that was enough.
But something has changed. Quietly at first, then all at once. The tools we draft in are no longer neutral containers. They’re active systems with incentives, pipelines, and downstream uses that extend far beyond spelling suggestions and version history. And for people whose work depends on trust, timing, and discretion, that shift is hard to ignore.
Drafting Is Part of the Reporting
Unpublished journalism isn’t raw material waiting to be processed. It’s the work. Drafts contain context that never makes it into print. Notes reflect conversations that only happened because someone trusted you. Early versions reveal angles considered and discarded for good reasons. Reporting lives in these in-between states long before a story is finished.
When that work passes through tools designed to ingest, analyze, and reuse content at scale, the relationship changes. Even if nothing malicious is happening, even if the product “works as advertised,” journalists are left with a nagging question: who else is touching this before I’m ready to publish?
This isn’t paranoia. It’s professional instinct. Journalists are trained to think about sources, exposure, and unintended consequences. Applying that same lens to drafting tools is overdue.
The AI Shift Made the Trade-offs Visible
For years, concerns about document platforms felt abstract. Companies promised security. Terms of service were long but ignorable. Most people assumed drafts were just drafts. Then AI entered the picture in a very real way.
Large platforms now explicitly train models on enormous volumes of text. The line between “your document” and “our systems” has blurred. Even when companies say content isn’t used for training in certain contexts, the policies are complex, change often, and usually require trust rather than verification.
The tools we draft in are no longer neutral containers.
For journalists, that’s a problem. Not because every draft is sensitive, but because some are. And you don’t always know which ones will matter most until later. The cost of getting it wrong isn’t theoretical. It can affect sources, legal exposure, and credibility.
Google Docs is an incredible product. So are many of the tools journalists rely on. But they’re built by companies whose priorities aren’t journalism. They’re built to optimize scale, data reuse, and future capabilities. That doesn’t make them evil. It just makes them misaligned for certain kinds of work.
Jurisdiction Is Part of Editorial Judgment
Where your drafts live determines which laws govern them, who can compel access, and what recourse you have if something goes wrong. That’s not a technical detail. It’s an editorial one.
A document can be stored “in Canada” and still be operated by a U.S.-based company, subject to U.S. corporate structures and legal frameworks. Infrastructure providers like AWS power huge portions of the internet, including newsrooms, but they don’t erase jurisdictional realities. Ownership matters. Control matters.
Journalists are used to thinking about legal environments when publishing stories. It’s strange that we’ve spent so little time thinking about them while drafting those same stories. As the gap between writing and publishing fills with automation and data extraction, that blind spot gets harder to justify.
A More Deliberate Way to Draft
Some journalists are starting to rethink their tools not because they want fewer features, but because they want clearer boundaries. They want to know that drafts stay drafts. That notes don’t quietly become training data. That deleting something actually deletes it. That line of thinking is what led me to build cDox.
What is cDox?
A document platform hosted in Canada, governed by Canadian law, and never used for AI training. Private by default. Publishable when you're ready.