Multi-Agent Content Generation — Quick vs Deep Research
What each agent does (Researcher, Strategist, Writer, SEO Optimizer, Headline Expert, review), which mode runs which agents, what reads your Brand Knowledge and Content Rules, and how brand mentions are kept in check.
The agents
| Agent | Job | Runs in |
|---|---|---|
| Researcher | Searches the web for competitor articles and public facts, finds the content gap | Quick and full |
| Strategist | Designs the outline: H2 rhythm, key takeaways, citation hooks; sees competitor knowledge for positioning | Full |
| Writer | Writes the body from the outline, following your tone and length rules | Quick and full |
| SEO Optimizer | Adds schema.org markup, internal links, keyword density | Full |
| Headline Expert | Produces three candidate H1s and the meta description | Quick and full |
| Review | Final pass: de-slop, consistency, brand-mention count preserved | Full |
Strategist and SEO Optimizer can be switched off in Content Rules; the others always run. Each agent can be given extra instructions there, and the three presets (professional, friendly, academic) replace all agent instructions at once.
Quick vs deep research
- Quick (Researcher → Writer → Headline): 30 to 90 seconds, no live web research beyond what the Researcher already knows, best for standard articles and how-tos.
- Deep research / full (all agents): 3 to 5 minutes, live web research, competitor facts checked against official sites, best for comparisons, lists and flagship guides. Requires 3 to 15 prompts (a counter-article started from Citation Sources can use one).
What the agents read
- Content Rules are read by every agent on every run. Business info is mandatory.
- Brand Knowledge is retrieved by topic for the multi-agent pipeline (strategy articles, deep research, Autopilot). The quick Answer Assets generator does not read it. Competitor-flagged knowledge is shown to the Strategist only, never quoted by the Writer.
- Competitors configured on the Competitors page are used for positioning; ownership and pricing claims about them are fact-checked and dropped when no official source confirms them.
Brand mentions and de-slop
The Writer works to a brand budget: two to four mentions in a platform version, at most one for developer communities, no limit for your own blog. The final de-slop pass removes filler while keeping the brand-mention count exactly as written, so what passes review is what gets published.
Frequently asked questions
Can I run two generations at once? One deep-research run per project at a time. Quick drafts queue.
Why did generation stop with "business info missing"? Fill in Business info in Content Rules; the pipeline will not start without it.
Does a failed run cost credits? Queued strategy articles and Autopilot jobs refund the held credits on failure.
Was this article helpful?