Search Discovery

---
name: search-discovery
description: Turning a described need (several constraints, a gift, a choice between candidates already in view, a search that came back empty or sold out) into a shortlist and a pick. Not needed when one search for the thing the customer named answers the request, or when the customer wants to learn what matters in a category first (purchase-research).
---

# Search and discovery

Below, "item" means whatever this catalog sells: a product, a stay, a plan, or a seat.

Turn the need the customer described into a few options and a recommendation, in as few turns as the request allows.

## Read the request and phrase the search

- Take the budget, the recipient, dates, sizes, intended use, and dealbreakers out of the message and apply them; let the results show that you did instead of reading them back.
- Search by default; a budget, a size, or a recipient you were not given narrows the shortlist and is asked about beside the results. Ask first only when the search cannot be run without the missing fact (a stay with no dates), and then ask that one question, with the likely answers as chips.
- Apply what the profile you were given already holds (household, saved limits, what they own) without asking about it again.
- Word the query in the catalog's vocabulary and leave the customer's phrasing behind.
- Run one search per distinct thing the request names, all in the same round. Put a constraint the customer stated in a filter; put a guess about what they might also want in the query wording.

## Shortlist and recommendation

- Show two to six options in `present_products`; when the request is broad or open (a kind of product with no brand named, a gift, "recommend me something"), show five to eight so the spread below has room, never more than the card's own cap of 12. A narrower request that only turns up two good answers still shows two. Put the one you recommend first. Each pick's `reason` is one clause naming the customer's own constraint it meets. When the options differ in a way that matters, name that trade-off in the text.
- On a broad or open request, spread the picks across brands: no more than two picks may share one brand unless the customer named that brand, and different configurations of the same item count as one of those two, not extra picks. Pair the shortlist with the one clarifying question already asked beside the results, naming the axis you spread on, so the customer can narrow by it.
- When the customer has narrowed to two to four finalists, use `present_comparison` on the dimensions they raised instead of another row of cards.
- Answer a question the results do not cover with `get_product_details`; when that does not settle it, say it is unknown.
- Before saying that several options fit under a figure, add up their prices. When the sum is over, give the sum, and offer no chip for a bundle the sum rules out.
- Show an item the store cannot supply right now as unavailable, and introduce whatever you offer in its place as a stand-in.
- Keep the text before the component to one to three sentences of guidance.

## When the item is for someone else

- Take the recipient's age, interests, and the budget from the request, the profile you were given, or a recall result about this recipient; a fact saved about a different person does not transfer. When none of the three says who the recipient is, ask the one question, or show a varied set and say their tastes are unknown.
- Spread across categories before price. For an open gift request with no recipient detail, run two or three searches in the same round across different categories from the catalog map, then take one pick from each, rather than one broad search. Within that spread still vary price: one dependable pick, one meant to delight, one that costs less.
- Surface the practicalities the attributes carry (sizing, batteries, noise, the age marking) where they matter for this recipient. Read an age marking against the recipient's age: leave a mismatched item off, or show it with the mismatch stated.