Forty-five minutes per article. That's the human time investment one content team reported after publishing 47 AI-assisted articles across three blogs in a single week - down from the 4-6 hours they previously spent writing each piece from scratch.
The team ran the experiment over a quarter, building a five-stage pipeline: automated trend discovery, drafting with a fine-tuned model (meaning a language model specifically trained on their past content and style), human editing, automated scheduling, and a performance feedback loop that fed ranking data back into future drafts.
The Numbers That Actually Matter
Of the 47 articles published, 12 reached Google's first page within three weeks. The total AI cost for the entire batch came to $380, which breaks down to roughly $8 per article on the generation side. Add the 45 minutes of human editing time per piece, and you're still looking at a fraction of what a traditionally written article costs in either freelancer fees or staff hours.
Those are solid numbers, but they need context. "First page" covers positions 1 through 10, and there's a massive traffic difference between ranking third and ranking ninth. The team didn't share click-through rates, conversion data, or how those rankings held up beyond the initial three-week window. Rankings in the first month after publication are notoriously unstable - Google's "honeymoon period" can boost new content temporarily before settling it into a more permanent position.
The Pipeline Is the Product
The more interesting takeaway isn't the output volume - it's the workflow architecture. The human editing gate at 45 minutes per article is doing heavy lifting. That's not a quick proofread; that's a substantive edit that likely catches the generic phrasing, factual gaps, and structural problems that AI drafts consistently produce.
Fine-tuning the model on their existing content is also a meaningful detail. Off-the-shelf ChatGPT or Claude will produce competent but generic blog posts. A model trained on your past articles, your tone, your typical structure - that closes some of the quality gap, though it won't eliminate it entirely.
The feedback loop matters too. Most teams using AI for content treat it as a one-shot process: generate, edit, publish, move on. Feeding performance data back into the system means the trend discovery phase can learn which topics and angles actually gain traction, rather than guessing based on keyword volume alone.
What This Doesn't Tell You
The experiment has a few blind spots worth flagging. Three weeks is too short to know if this content will hold its rankings. Google's helpful content system evaluates sites on a rolling basis, and a sudden influx of 47 articles in one week is exactly the kind of publishing pattern that can trigger closer scrutiny over the following months.
There's also no mention of whether these articles competed in low-difficulty keyword spaces or went after more competitive terms. Publishing 47 articles targeting long-tail queries with minimal competition is a very different achievement than ranking for terms with established incumbents.
And $380 in AI costs is cheap until you factor in the human time. At 45 minutes each across 47 articles, that's roughly 35 hours of editing work in a week. For a solo creator, that's nearly a full work week of nothing but editing AI drafts. The economics only work if you have the editorial capacity to handle that throughput.
Still, the core finding holds: a well-structured AI content pipeline with genuine human oversight can produce rankable content at dramatically lower cost per piece. The teams that will get the most out of this approach are the ones treating AI as a first-draft machine inside a disciplined editorial process - not as a replacement for writers.