Why factuality guardrails matter more than 'creativity'

The most common failure mode of LLM rewriters isn't that they sound robotic. It's that they fabricate — confidently. They invent stats, round numbers the wrong way, attribute quotes to the wrong speaker.

If your rewriter saves you three hours of drafting but costs you three hours of fact-checking, you haven't saved any time. You've moved the work.

Voxmea's factuality pass extracts every named entity and numeric claim from the source transcript, checks them against the output, and flags drift before you see the draft. That's the difference between a writing *assistant* and a writing *risk*.