Prompts that work well

The chat is most useful when you write messages that clearly signal what you want. This page collects patterns we’ve seen produce good results.

For triggering a clean search

Pattern: “Find/show me [product] [constraints]”

Claude reliably extracts as a search.

Examples:

  • “Show me bamboo socks under $20 with low competition in the UK.”
  • “Find FBA products in the home-fitness category around $30–$60 with at least 30% margin.”
  • “Show me yoga accessories that are trending up over the last 12 months.”

Pattern: “Products in [category] for [criteria]”

Same effect.

Examples:

  • “Products in beauty under $25 for sellers wanting low margin requirements but high volume.”
  • “Tools in home-improvement that aren’t already saturated.”

Pattern: “Find [adjective] [products] [marketplace]”

Adjectives like “trending”, “high-margin”, “low-competition” are interpreted as filter or weight signals.

Examples:

  • “Trending pet-care products in the US.”
  • “High-margin gift items in DE.”

For asking conversational questions (no search)

Pattern: “How does [feature] work?”

Examples:

  • “How does the demand sub-score work when SFR isn’t available?”
  • “Why did this product score 72?”
  • “What’s the difference between trend and seasonality sub-scores?”

Pattern: “Explain [concept]”

Examples:

  • “Explain why a low data-quality score affects how I should read the SellerScore.”
  • “Explain when I should bump the margin weight up.”

Pattern: “What’s a good [strategy] for [scenario]?”

Examples:

  • “What’s a good weights configuration for a new seller without brand registration?”
  • “What’s a good way to filter for products that ship cheaply via FBA?”

For refining a previous search

Pattern: comparative refinement

Examples:

  • “Tighten the previous search to 25%+ margin.”
  • “Same search but in DE instead of UK.”
  • “Now exclude products with fewer than 100 reviews.”

Claude understands “previous search” / “the search above” as a reference and either re-runs (1 credit) or filters in place (free, when possible).

Pattern: deep-dive on one result

Examples:

  • “Tell me more about product 3.”
  • “Why is product 1 scoring higher than product 2 even though product 2 has better margin?”
  • “Are any of these likely to be Amazon Basics competitors?”

These run as conversational analysis (free).

Patterns that don’t work as well

Multi-part messages

“Show me bamboo socks AND tell me about the demand sub-score AND explain how exports work.” Claude has to pick one tool; multi-question messages confuse intent extraction.

Fix: ask one thing at a time.

Vague messages

“Find good products.” Without specificity, Claude either picks no_search_intent and asks a follow-up, or extracts overly-broad params.

Fix: include a category, a marketplace, or at least a price band.

Asking for actions outside Hilal

“Email this to my supplier.” Claude can’t do that. The chat is research-only.

Fix: export to PDF/CSV (Exports) and email yourself.

Tips for ongoing conversations

  • Keep context short. Long chats consume more tokens per turn (the context grows). Start a fresh session when the topic shifts.
  • Star or rename important sessions. The auto-title is good but renaming to “Q4 launch research” makes it findable later.
  • Use chat for synthesis, guided for precision. When you need to nail down exact parameters for a repeated workflow, switch to guided.

Related articles