CAROL Bike: A Daily Content Pipeline That Drafted 42 Evidence-Grounded Articles and Let the Team Publish 39
CAROL Bike had 197 editorial articles on a template-driven WordPress site, and they were already good: six to twenty PubMed citations each, fresh, well-ranked. The audit in June 2026 showed the problem was not authority but demand capture. The main-site articles pulled 2.36 million impressions in 90 days and turned them into 8,213 clicks, with the biggest pages sitting at positions 5 to 8 and CTR of 0.1 to 0.6 percent. The queries with real volume ("exercise bike vs treadmill", "how long to see results from a stationary bike", "is a stationary bike good cardio") were generic, not REHIT-branded, and the site had one writer's worth of capacity for a keyword list that needed one article a day. Writing at that pace by hand meant either thin content or another headcount. The constraint from the start: nothing goes live without a human reading it.
A scheduled agent pipeline, running daily from 8 June 2026. Each run pulls the striking-distance keyword set from Search Console (queries where CAROL already ranked 11 to 25), checks the candidate against every existing article through the site's public posts endpoint so it complements rather than cannibalises an existing page, then drafts a 1,100 to 1,300-word article in CAROL's tone of voice with each claim grounded in a cited study (a PubMed connector; web research against publisher metadata as the fallback). The draft is pushed to WordPress as a draft post with category, featured image, display title, Yoast meta description and focus keyphrase already set, through a small authenticated shim installed as a code snippet on the multisite. A Notion queue records every run with the keyword, volume, prior rank and a dedupe note, and doubles as the fallback store: when the Shopify migration removed the WordPress REST API on 25 August, the pipeline detected the 404 and filed the finished drafts to Notion instead of failing silently. The CAROL team reviews and publishes; the pipeline never does. Peak cadence was one article per day for 16 consecutive days in July.
42 evidence-grounded articles drafted, 39 published by a human, zero by a machine