What it produces
A Category Brief answers: what is this market actually about?
Structured sections:
- Category definition — bounded by which products, keywords, segments
- Audience segments — the demand pockets, sized
- Pain points — top recurring negative themes across all sources
- Growth levers — what's driving category expansion (or contraction)
- Defining properties — which features the market consistently evaluates
- Top use cases — what consumers actually do with category products
- Citation share landscape — who's winning the AI Overview / editorial game
What L1 reads
L1 reads the structured intelligence layer, not raw snippets. Specifically:
distinctive_keywords_segment— which keywords define each segmentkeyword_search_volume— sizing per demand pocket- Aggregated
rag_snippets(by canonical property, snippet_type, sentiment) — what the market thinks ai_overview_sources— citation landscapeproduct_taxonomy— what products are in scope
The L1 prompt then asks Claude Sonnet to synthesise, not extract. The hard work of extraction happened upstream.
Why structure beats free-form
The L1 brief is structured (fixed sections, fixed field types). This matters because:
01 — Comparable across categories. A Canon mirrorless brief and a Bose headphones brief have the same shape. Multi-category clients can compare directly.
02 — Machine-readable. L2 and L3 read L1 programmatically. A free-form prose brief would force re-parsing on every downstream call.
03 — Quality-checkable. Each section has acceptance criteria (e.g. "audience segments must sum to >80% of category volume"). The pipeline rejects briefs that fail these checks and re-runs.
Cadence
L1 briefs are monthly for client deployments. Categories don't restructure week-to-week — but they do shift over months. Monthly cadence is enough to capture genuine category drift without being noisy.
Re-running on weekly data would mostly produce a slightly different prose version of the same brief. Diminishing returns set in fast.
What good looks like
A high-quality L1 brief, for a category we know well (Canon mirrorless UK), should:
- Identify the 9 segments Leiden found (true positives)
- Surface 3-5 recurring pain points that show up across sources
- Name the defining property set (autofocus, image quality, video, portability, value, IBIS, viewfinder)
- Spot the content gap between what brands say and what market validates
- Position the AI Overview citation race with a leaderboard
The brief is 8-12 pages dense — enough for a category planning meeting, not so much that nobody reads it.
What L1 doesn't do
L1 doesn't make recommendations. It frames the market. Recommendations are the job of L3 Situation Analysis, which reads L1 (and L2) and proposes prioritised actions per product.
This separation matters: L1 is stable (the category is what it is). L3 is brand-specific (your situation in the category is what it is). Keeping the layers distinct keeps both useful.