AI brief

An AI brief is a Claude-generated paragraph (or short series of bullet points) attached to each product, structured as four sub-sections:

  • Opportunity — why this product might be worth selling.
  • Risks — what could go wrong.
  • Positioning — how to differentiate from existing listings.
  • Sourcing notes — what to look for from suppliers.

It’s a synthesis of the metrics, the listing data, and Claude’s domain knowledge. Useful as a sanity check before committing.

In this guide:

  • What goes into a brief
  • What a good brief looks like
  • The 48-hour cache (why briefs are sometimes instant)
  • When to trust the brief

What goes in

The brief generator gets:

  • The product’s listing metadata (title, brand, BSR, reviews, rating).
  • All six sub-scores and the data-quality flags.
  • Profitability detail (sell price, COGS estimate, fees, net margin/profit).
  • The original search context (keyword, marketplace).
  • A system prompt instructing Claude to focus on Amazon-seller actionable insights.

It does not get the open internet — Claude isn’t doing real-time research; it’s reasoning over the data Hilal already has.

What a good brief looks like

Example for a low-saturation, high-margin product:

Opportunity — Bamboo socks at the $18–$24 sell price hit a sweet spot in the UK market: enough margin (32%) to support modest ad spend without eroding profit. Mid-tier brand presence means there’s room to differentiate on packaging and material claims.

Risks — The trend sub-score (62) is solid but not exceptional; this isn’t a hyper-growth category. A flood of new entrants would compress margins fast. Also, “bamboo” claims are increasingly scrutinized — buyers expect proof of fibre source.

Positioning — Lead with material certification (rayon-from-bamboo vs. lyocell-from-bamboo distinction), not generic “eco-friendly” claims. Photography should emphasize texture and feel; current top listings underuse close-ups.

Sourcing notes — Typical supplier MOQ at this price point is 500–1,000 pairs. Look for OEKO-TEX certification specifically; AliExpress/Alibaba listings often advertise “bamboo” loosely.

The brief is scannable — you can read it in 20 seconds and make a yes/maybe/no call.

The 48-hour cache

Briefs are expensive to generate (a full Claude call). To keep cost down:

  • Briefs are cached 48 hours in Redis and Postgres.
  • The cache is shared across users — when one Hilal user opens a product detail and triggers a brief, every subsequent user sees that same brief for 48 hours.
  • The cache key is (asin, marketplace) — same product, same brief.

What this means in practice:

  • Most popular products already have a brief; opening their detail loads instantly.
  • Niche products you research first get a brief generated on-demand (5–15 seconds).
  • 48 hours later, the brief refreshes if you re-open.

The cache is a feature, not a limitation. If you’ve made a decision based on a brief that’s a few hours old, the underlying metrics haven’t changed materially.

When to trust the brief

Treat the brief as a second opinion, not a primary source. It’s good for:

  • Catching things the metrics don’t show (“this category is brand-driven; new entrants struggle”).
  • Suggesting positioning angles you might not think of.
  • Surfacing sourcing-related caveats.

It’s less good for:

  • Quantitative claims (always defer to the actual sub-scores).
  • Marketplace-specific regulations or compliance details (Claude doesn’t have real-time legal data).
  • Specific supplier recommendations (the brief stays generic).

If the brief and the SellerScore disagree, the SellerScore is the quantitative ground truth; the brief is qualitative color.

Refreshing a brief

If the brief feels stale (cache age, product changed materially), top-right of the brief panel: Regenerate. This forces a new Claude call (charged separately as a small AI call, not a credit). Use sparingly.

Related articles