Semantic Brief skill
The playbook Moonrank's Writer Agent uses to design the spine of a page — including the one heading a competitor would never write.
Download the methodology, see the surprise gradient rule, or run it live inside the Writer Agent.
- Category
- Writer
- Agent
- Writer Agent
- Format
- semantic-brief.md
- Price
- Free to download
The skill file
What is the Semantic Brief skill?
The Semantic Brief skill is how Moonrank's Writer Agent designs a page before anyone drafts it. It decodes the query's primary and latent intents, lists the semantic vectors a page on that query is expected to contain, maps the micro-intents a reader carries before, during and after the decision, and turns all three into a heading structure where every H2 exists for a reason.
Its distinguishing rule is the surprise gradient: at least one heading must be counter-intuitive or say the thing competitors avoid. A structure where every heading is predictable produces a page that adds nothing to what a model already holds, which is a page it has no reason to retain or cite.
It's written as a SKILL.md, a plain-language instruction format that people and AI agents can both follow. Read it to brief your own writers, hand it to an agency, or let the Writer Agent build the plan for a target query.
See it in action
“Give me the plan for a page on replacing an electric car battery.”
Decisional intent, with three latent ones: fear of being ripped off, uncertainty about normal degradation, and the warranty question. Eight H2s, each mapped to a vector — and the opening one is "The €500 mistake most owners make", which none of the ranking pages says because it costs a garage money to admit. The FAQ is four bottom-of-funnel questions, not four generic ones, and the page needs a cost calculator because the intent is "how much".
From input to fix list
- 1
You give the keyword and intent
The query, the page's purpose, the sector, and any proprietary evidence you hold.
- 2
Moonrank decodes the intents
The primary intent, plus what the reader worries about but never types.
- 3
It builds the Hn structure
One heading per semantic vector, with a surprise gradient built in.
- 4
It places the proof
E-E-A-T signals and required formats, each with its position in the page.
Everything it checks
Primary and latent intent
What the reader typed, and what they are actually anxious about.
Semantic vector list
The entities, co-occurrences and expert vocabulary the query expects.
Micro-intent mapping
The questions before, during and after the decision — the source of the H3s.
The surprise gradient
At least one heading competitors avoid, so the page adds rather than repeats.
E-E-A-T placement
Which proof goes where, rather than a general instruction to be credible.
Multimodal requirements
The table, tool or video the intent needs, with the reason it's needed.
When to reach for it
- Brief a writer with a plan instead of a keyword.
- Rebuild a page that ranks below its writing quality.
- Stop producing pages that average the current top three.
- Find the heading your competitors won't write.
- Decide what proof a page needs before drafting it.