I'm a radiologist — here's how an AI editor replaces the authoring part of PowerScribe (and the use-case SEO experiment behind it)
Update for those following the RONTGEN build.
Quick background: I'm a radiologist from Brazil, and I started RONTGEN because of a daily frustration nobody warns you about in this job — the report takes longer to write than the study takes to read. Reading the images was never the bottleneck; turning what I see into a clean, structured, signable paragraph is. Multiply that across a worklist and the documentation is the job.
The software that's supposed to help — PowerScribe, Dragon — is genuinely good. But it's hospital software: priced per radiologist per year (a number that only makes sense for a whole department), wired into the PACS, gated behind IT, and tied to one workstation. The moment you're on a locum shift, moonlighting, or doing teleradiology from home, you're back to copying templates off Google Drive or a pen drive.
So I wrote an honest guide to what "dictating reports with AI" actually means in 2026 — and where a tool like RONTGEN fits. The thesis: for the authoring part of reporting, a modern AI editor can replace that hospital software instead of sitting beside it. In practice:
→ Voice-to-text with a personal dictionary — add an eponym or drug name once and it stops getting it wrong, on every machine. → Agent pipelines that apply your template and house style in one pass — define it once, run it on every report. → A copilot for the impression — draft from your findings, or get a second opinion / alternative wording from a different model. → Every major model in one place — and if one's ever down, priced out, or restricted, you switch without rebuilding anything. → Optional local, private models (Ollama) for sensitive text — nothing leaves your machine. → A fraction of enterprise pricing — a small monthly plan plus credit packs that scale with how much you use, or your own API keys at cost.
The honest boundary (candor is the whole point with this audience): RONTGEN is the authoring layer, not the hospital's records system. It doesn't reach into PACS to pull images or push the signed report into the hospital DB — you deliver the finished report into whatever system the site uses. You stay the radiologist; the AI handles the slow part, the writing.
Now the part IH might care about more — the meta. This post is piece #1 of a content experiment. My funnel diagnosis is brutal: plenty of signups, but my organic traffic is ~100% brand search (people who already know the name). Basically zero discovery. So instead of more generic "AI writing tool" posts, I'm testing use-case / persona SEO: enter through one concrete persona (radiologists), speak their exact pain, and try to get found by people searching the problem — not my brand. Radiology is the wedge because it's the one niche I've actually validated, and it's my own domain so I can write it credibly.
Read the post: https://rontgen.app/blog/dictate-radiology-reports-with-ai.html
Two things I'd genuinely love input on:
If you've used use-case/persona SEO to drive real activation (not vanity traffic), what actually worked — and how long before it showed up?
For anyone who's sold into healthcare or other regulated niches: how did you handle the "this isn't a certified medical device" line without killing momentum?
