What it does
The writer drafts from a brief the agent produced: the target query, its intent, the sections a complete answer needs, the entities to name, the internal links to carry and the metadata to ship with. The output is a strong first draft for your team to edit — not a page to publish unread.
What makes a draft good or worthless
The model is not the differentiator. Two drafts from the same model can be wildly different in quality depending on what preceded them. A draft generated from a keyword produces filler. A draft generated from an intent analysis, a competitive read of what already ranks, and an outline of what a complete answer contains produces something worth editing.
| Input | Why it matters |
|---|---|
| Target query and intent | Decides the page type and the shape of the opening |
| Coverage analysis | Identifies the subtopics a complete answer has to include |
| Entity list | Keeps naming consistent and unambiguous across the site |
| Internal link targets | Links are planned, not bolted on afterwards |
| Existing coverage | Prevents writing a page that competes with one you already have |
The shape of every draft
- 01Direct answerTwo or three sentences that resolve the query completely — the passage most likely to be extracted by an answer engine, and the one that keeps human readers on the page.
- 02ContextWhy the question matters and what changed to make it worth asking.
- 03MechanismHow the thing actually works, in named steps or components.
- 04ComparisonA real table wherever the reader is choosing between options.
- 05PracticeWhat to do, in order, specific enough to act on.
- 06LimitsWhat this does not solve. Balanced writing is more citable than promotional writing.
- 07Follow-upsAn FAQ covering the questions the page itself raises.
Checked before it reaches you
- Semantic coverage against the subtopics the cluster requires.
- Entity consistency with the rest of your site.
- Internal links resolved to real pages with descriptive anchors.
- Heading hierarchy in order, with headings that state a claim.
- Metadata and structured data generated alongside the draft.
Where the human stays
The agent does not know your positioning, your roadmap, or which claims your legal team will accept. It removes the repetitive layer — research, structure, metadata, link selection — so the editorial pass is about judgement rather than formatting. Nothing publishes without approval; see how it works.
Written for both surfaces
The same structure that ranks also survives extraction by an answer engine: a self-contained opening, unambiguous entities, real tables and lists, and an FAQ. That is why drafting and GEO optimization are one workflow rather than two.