Running this across four sites. Posting it because the "AI content farm" posts I see are either selling a tool or hiding the editing cost.
The pipeline
- Keyword set clustered by SERP overlap, not string match
- One outline pass per cluster, reviewed by a human in about 2 minutes
- Draft generated section by section against the outline, with the SERP's actual top 3 as context
- Automated fact extraction into a checklist a human then verifies
- Human edit pass: 18 minutes average per article
Costs, monthly
Tokens $240 · editors $4,800 (two people, 400 articles, 18 min each) · tooling $190. Total about $5,230, or $13 an article.
What it does well
Breadth. Covering 400 long-tail queries a month for $13 each is not possible any other way.
What it cannot do
- Anything requiring first-hand experience. It will confidently invent having used a product
- Anything where being wrong is expensive — medical, legal, financial
- Original data. It has none, and neither do you after it finishes
- A reason for anyone to link to the page
That last one is the business problem. 400 pages a month with no reason to be cited is a traffic strategy with a ceiling, and the ceiling is getting lower.
AI content pipelines. The model is the cheap part.