Rank with articles your agents write

An article job is not one model call. It is a staged pipeline, and each stage exists because skipping it produces the generic AI article everyone can smell.

• SERP research. The pipeline reads what currently ranks for the topic: the angles taken, the questions answered, the gaps nobody covers. Your article is positioned against the real competition, not written into a void.
• Outline. A structure is drafted first, sections, FAQs, and target headings, so the piece argues one thing instead of wandering.
• Parallel writing. Sections are written concurrently against the shared outline, which keeps a 2,000-word piece fast without letting it drift.
• QA. A separate pass checks claims against the research, cuts filler, and rejects drafts that fail. Failed drafts cost you a retry, not a published embarrassment.
• Metadata, JSON-LD, hero image. Every draft ships with title tag, meta description, Article structured data, and a generated hero image. Nothing to bolt on later.

Search engines reward experience, expertise, authority, and trust. A pipeline cannot fake your experience, but it can stop diluting it. The niche settings are where you put the things only you know: your point of view, your terminology, the claims you will and will not make, the products you actually use.

The practical move is to treat the niche's tone and topic pillars as an editorial policy, not a style preference. “We always state prices, we never say ‘game-changer’, we test before recommending” is policy the QA stage can enforce on every draft. Generic content is mostly the absence of policy.

Articles inherit the niche's audience and tone, but a few settings deserve their own pass before you turn the pipeline on:

• Topic pillars. Three to five themes the niche owns. Ideation proposes topics inside the pillars, so tight pillars produce a coherent site section instead of a scattershot blog.
• Search intent mix. Decide the ratio of informational (“how does x work”) to commercial (“best x for y”) pieces. Early on, weight informational; it ranks sooner and builds the authority the commercial pieces need.
• The daily cap. Articles cost less than videos, so a mixed niche needs headroom for both. Check today_spend or the dashboard strip if article jobs are getting refused late in the day.

A hundred orphan articles rank worse than thirty linked ones. The strategy that works is boring and old: hub and spoke. One pillar page owns the broad query; spoke articles answer the specific questions and link up to the pillar; the pillar links down to every spoke.

Map your pillars to hubs before you generate volume. When you enqueue a topic, you are really choosing which cluster it joins, and a topic that fits no cluster is usually a topic to skip. As spokes accumulate around a hub, the hub starts moving for queries you could not have targeted directly.

Search engines index rhythm. Three articles a week for six months outperforms ninety articles in a burst followed by silence, both in crawling behavior and in your own ability to read results. Set the niche's article cadence to something you can review sustainably, then leave it alone.

Expect the timeline to be honest but slow: indexing in days, movement on long-tail queries in weeks, competitive head terms in months. The pipeline's job is to make the waiting cheap; your per-article cost stays flat while the compounding does its work.

Weekly, not daily: impressions by cluster, average position on the queries each hub targets, and which spokes earn links or completions. Feed what you learn back as topic decisions, double down on clusters that move, prune pillars that stay flat for a quarter.

Intervene on inputs, not outputs. Editing one article by hand fixes one article; tightening a pillar, a policy line, or the intent mix fixes every article the niche writes from then on. That is the entire trick of running an editorial machine: push your judgment up a level and let the pipeline propagate it.