Use case · Content

One topic in, a keyword-targeted article batch out. A strategist agent plans the cluster, and every article arrives written, structured for search, and paired with an editorial hero image — the deliverable content agencies bill per article, produced by the batch at provider cost.

Ranking rarely comes from one article. Search engines reward clusters — several pieces that own a topic together, each answering a different intent. Producing a cluster is what makes content expensive; this graph makes it one run.

How it works

The strategist plans once; everything after the list generator runs once per article.

graph LR business["Business & site (String)"] seeds["Audience & seeds (String)"] count["Article count (Integer)"] strategist["Strategist (Agent)"] briefs["Briefs (ListGenerator)"] writer["Writer (Agent)"] hero["Hero image (TextToImage)"] business --> strategist seeds --> strategist count --> strategist strategist --> briefs briefs --> writer briefs --> hero
  1. Feed in the campaign. Who’s publishing and which pages the articles support, who searches and the seed topics, and how many articles to produce. (e.g. “ultralight hiking shop · beginner hikers · trail runners vs boots · 4 articles”)
  2. Plan the cluster. A strategist agent picks the topic cluster, assigns one primary keyword per article with its search intent, and divides coverage so the articles don’t compete with each other.
  3. Write the briefs. A list generator turns the plan into one brief per article: title, keyword, intent, H2 sections, and what the reader must be able to do afterward.
  4. Write the articles. For each brief, a writer agent produces the full Markdown piece — meta description, answer-first opening, the brief’s section structure, an FAQ, and one unforced CTA.
  5. Render the heroes. In parallel, each brief becomes a 16:9 editorial hero image with room left for a headline overlay.

One run, one cluster

Because every article comes from the same plan, the batch behaves like a cluster instead of a pile: keywords don’t cannibalize, internal logic is consistent, and the voice section of the plan keeps ten articles sounding like one publication. Point the seed inputs at the next cluster and run it again.

Make it yours

  • Retune the writer. The article rules — length, structure, point of view, FAQ — are a prompt, not a setting. Make it terse and technical or warm and story-led.
  • Raise the stakes per article. Swap the writer’s model up for money pages, down for supporting pieces.
  • Change the imagery. The hero prompt controls the visual language — photography, flat illustration, product-in-context.
  • Chain the next step. Add a Save Text node per article, or a translation branch to ship each cluster in every market’s language.

Models in this workflow

The template ships with no models selected — pick one per role when you open it. Called with your own keys; the bill comes from the provider.

Role Node Works well with
Cluster strategist Agent Any strong language model
Brief writer List Generator Any language model
Article writer Agent A long-output model you trust with prose
Hero images Text To Image FLUX, Imagen, Nano Banana

See Models & Providers to set up keys.

Next steps