How We Automated Our Content Pipeline With AI

The problem was never writing—it was everything around it. Keyword research, briefs, internal linking, meta tags, and CMS formatting ate 70% of the hours per post. So we didn't try to automate the writing itself. We automated the connective tissue between the human decisions, which is where the real time went.

Stage one is research. We feed a target query into a brief generator that pulls the top 10 ranking pages, extracts their headings, and returns a structured outline with the questions real users ask. This runs in about 90 seconds and replaces what used to be an hour of manual SERP scraping in spreadsheets.

Stage two is drafting. We pass that brief to Claude with a fixed system prompt encoding our voice rules—answer-first paragraphs, no hype, concrete numbers. The model returns a full draft plus FAQ blocks and meta description. Critically, the prompt is version-controlled in Git, so when we improve the voice, every future article inherits it.

Stage three is editing, and this stays human. An editor fact-checks claims, adds first-hand experience the model can't invent, and cuts anything generic. This is non-negotiable: AI drafts are a starting point, not a finished product. We reject roughly one in five drafts outright and rewrite them, because a wrong number in an answer-first paragraph is worse than no article.

Stage four is publishing. Once approved, a script pushes the article straight into Webflow CMS via the API—title, body, meta tags, category, and internal links mapped automatically. No copy-paste, no formatting drift. What used to be 30 minutes of CMS wrangling is now a single command that takes seconds.

The results after six months: article output roughly tripled, from four posts a month to twelve, with the same two-person team. Time-per-article dropped from about two days to under three hours. Organic traffic to those pages is up because we now publish enough volume to actually build topical depth around a subject instead of one-off posts.

Two honest caveats. First, AI made our bottleneck the editor, not the writer—so we invest in editorial judgment, not more automation. Second, quality only held because a human owns the final call on every post. If you remove that gate to chase volume, search engines and readers notice fast. Automate the plumbing, keep the judgment human.