match-offers
ChatGPTFind purchase-linked shopping rewards for a shopper and render them in the widget. Covers every reward type: cashback, coupons, discounts, rebates, loyalty points, miles, fuel points, sweepstakes, instant wins, contests, and giveaways. Works for two stages, so set
stage accordingly. Pre-purchase discovery (stage: "discover"): the user is considering or about to buy ('what offers are there at Kroger?', 'I'm going to buy dog food'). Pass merchant and/or items; the widget shows a browseable carousel they can open for details. Refinements ('grocery only', 'cashback not points') re-call with categories / rewardTypes. Post-purchase conversational (stage: "post_purchase" or omit): the user already bought something. Pass items and merchant; the widget shows claimable offer rows directly. If the user only mentioned some purchases and has neither shared a receipt nor given what is clearly their full list, first ask one short question before calling: whether they'd like to share their purchase receipt (a photo or screenshot works — you extract the items yourself) so every product they bought gets checked. If they decline or already gave the full list, call with what you have. Post-purchase receipt: pass the extracted items, merchant, date with fromReceipt: true. The widget shows a confirmation card first; when the user confirms, call again with the same data plus confirmed: true. Fallback: when this tool returns no offers, or only weak/loosely-related ones, a web search for relevant deals or general savings info is a good next step rather than a dead end, briefly noting those come from a web search (not Snipp's catalog). Keep it light, it's part of the Snipp experience. After results: the widget already shows every offer (carousel or claimable rows), so reply with a short acknowledgment and an offer to help (narrow down, compare, explain an offer, suggest the best fit) rather than re-listing offers or rewards as text. Only purchase data (merchant, item names, date) needs to reach this tool, never names or card numbers. Cover every reward type without declining a category, and reply in the user's language.
