Summary
A pre-FDA medical device brand had a one-page website, no ranking keywords and a client who had declined to raise a $5,000-a-month paid budget. We built the compliance rules before the content: a 16-term banned-word list, 14 quality gates, a 25-point scorecard. Nine articles went through at 24 or 25 out of 25 and six cleared medical review with no rewrites.
Starting from nothing, in the hardest category there is
When I audited the company’s two domains in July 2026, the picture was as close to zero as a live website gets. The main site had one page, no ranking keywords whatsoever, and 306 referring domains that were almost entirely purchased spam. The robots file returned successfully with zero bytes.
The sitemap returned a 404. The second domain returned a 404 at its root and had earned one click and two impressions in Search Console across the whole window.
There was no patient-education content because there had never been any content. And the client had made a decision about money: they would not put more into paid, because awareness-only ads for a procedure that wasn’t yet FDA-cleared didn’t produce a return they could see.
So organic search had to build the patient audience before the product existed to sell them. And it had to do so without a single claim a regulator could object to.
In a regulated vertical, the gate is the product
Most content programs start with the content and add review later. In pre-FDA medical, that order is a regulatory problem. A word like “cure,” “painless” or “minimally invasive” isn’t a style choice; it’s a claim about a device that hasn’t been cleared to make it.
So we built the governance first, as a five-stage approval framework with the client’s clinical lead as the named final owner. Two of the five stages need the client.
Two revision rounds maximum, no third. About three business days of client time, seven to eight business days from brief to live.
Inside that framework:
- A 16-term banned-word list, tighter than regulators require: revolutionary, painless, cure, guaranteed, minimally invasive, non-surgical alternative and ten more. The list lived inside the drafting prompt, so Claude never saw those words as options.
- Fourteen quality gates per article, including: every cited source verified live at draft time, so no fabricated references; a named medical authority on every clinical claim, from PubMed, StatPearls, ACOG or peer-reviewed sources only; the direct answer in the first 115 words; five to nine FAQ entries; the structured data a medical article needs; and an honest byline that reads “Awaiting review” until a named clinician clears it. Never a fabricated name.
- A 25-point scorecard. Anything scoring below 24 goes back into the pipeline. It doesn’t get a human’s time until it earns it.
How the engine ran
Each article moved through a compliance-mode build of my Article Factory: five automated stages, then one hard human gate. The factory read the client’s context and persona files. It wrote a brief.
It drafted 2,300 to 3,300 words against the brief, the banned-word list and a twelve-part structure built for AI search. An optimization pass tightened the answer-first opening and FAQ.
The scorecard graded. Then, and only then, a clinician reviewed it in a hosted review app built for that batch, so the reviewer worked in one clean interface rather than a pile of documents.
Nine articles were written from zero, covering four patient personas across five stages of their journey. Alongside them: a content calendar, a publishing bundle ready to drop into the client’s site, patient guides and a quiz funnel for lead capture, and a KPI dashboard wired to analytics and consultation requests so the day it launched, it would be measured.
What came out
- Nine of nine articles scored 24 or 25 out of 25, with zero banned words and zero unsupported claims, disclaimer block in place.
- Six of nine cleared medical review and were approved and ready to publish. The third batch was in review when the engagement paused.
- The client’s fractional CMO, on a recorded retrospective: the content was “exceptional” and “quite impressive to showcase from a content generation standpoint.”
Where it stands
No articles are published yet. The engine was blocked on the client’s own website build and then on their FDA timeline, and the engagement paused at the end of August 2026 pending a regulatory filing. The KPI dashboard says so itself: published live, zero; numbers stay at zero until the site ships.
The engine is built, the approved articles are queued, and the dashboard is wired. When the site ships, it turns on.
At a glance
- The niche: Pre-FDA medical device, colorectal procedure, pre-revenue.
- The problem: Zero content, zero ranking keywords, a spam backlink profile, a broken robots file and sitemap; paid budget capped; every claim regulated.
- What was built: A five-stage content approval framework, a 16-term banned-word list, 14 quality gates, a 25-point scorecard, a compliance-mode build of my Article Factory with mandatory clinical sign-off, hosted review apps, a KPI dashboard.
- Implementation: Governance first, then nine articles across four personas and five journey stages; two revision rounds maximum; brief to live in seven to eight business days.
- Results: Nine of nine at 24 or 25 out of 25 with zero banned words; six of nine cleared medical review with no rewrites; the client’s CMO on record calling the output exceptional. Publishing follows the client’s site build.
- Built with: a compliance-mode build of my Article Factory.
