Every recommendation should make the user more confident, not just more likely to click.
Most users do not need a marketplace grid. They need a small ranked set with clear tradeoffs:
- Budget
- Best for most people
- Upgrade or best specialized fit
Ask follow-up questions for constraints that materially affect ranking:
- Budget
- Size or fit
- Compatibility
- Delivery timing
- Style preference
- Safety requirement
Do not turn shopping into an interview.
Facts:
- Price
- Availability
- Specs
- Merchant
- Shipping
- Return policy
Judgment:
- Best for a commuter
- Most giftable
- Better long-term value
- Not worth the upgrade
The UI should make this distinction visible.
A trustworthy recommendation says who should avoid it.
Examples:
- "Do not buy this if you need dishwasher-safe lids."
- "Not ideal for wide feet."
- "Too large for most under-seat airline use."
- "Great price, but seller reliability is unclear."
If a shopping answer shows price or availability, it needs:
- Source
- Timestamp
- Merchant
- Variant
- Confidence
Without that, the answer should speak carefully.
Paid placements can exist, but they must be labeled and separated. The user's trust should not depend on guessing which answer was bought.
The best AI shopping answers will be built on structured merchant data, not only scraped pages and review summaries.
When the user leaves for a merchant, the handoff should preserve:
- Product identity
- Variant
- Price
- Return policy
- Merchant disclosure
- Reason this product was selected
If the system cannot verify a product, seller, claim, or price, it should say so. Refusing to fake certainty is a competitive advantage.