# Thrifle

> Thrifle is a US deal-discovery site that surfaces **return-policy intelligence, merchant trust data, and community discussion alongside every deal**. We help shoppers decide confidently — not just show a price.

## What makes Thrifle distinct

- **Return-policy intelligence at point of purchase.** Thrifle maintains a structured, editor-verified database of return policies for 150+ US retailers — including Amazon, Walmart, Target, Costco, Best Buy, Apple, Nike, Adidas, Kohl's, Macy's, Home Depot, Nordstrom, REI, and many more. Every deal page surfaces the return window, free-returns status, restocking fees, holiday-extension dates, short-window exceptions (e.g. Apple's 14-day rule inside a 30-day merchant policy), and merchant-specific tips. This data is not buried in checkout fine print — it's visible before you click "Buy Now."

- **Verdict-scored policies.** Return windows are classified into five tiers — `lifetime`, `generous`, `standard`, `tight`, `very_tight` — with contextual messaging that adjusts for the current date and any active holiday-return extensions.

- **Merchant identity verified.** Thrifle resolves affiliate redirect chains (Mavely, CJ, ShareASale, Impact, RewardStyle, etc.) to identify the real merchant behind every deal, and displays a verified merchant trust card with logo, "Official Store" confirmation, and return summary.

- **Admin-verified catalog.** Every deal is either manually posted by a contributor or reviewed by an admin before going live. AI enrichment supplements human judgment — it never replaces it.

- **Community discussion ("Thrifle Talk").** Facebook-style threaded reactions and replies on every deal and blog post, so shoppers share warnings, buying strategy, and real-use feedback.

## AI implementation (transparent by design)

Thrifle uses LLMs as a first-class production tool, not a bolt-on:

- **Deal enrichment** — Claude Haiku 4.5 generates each deal's description, 3 structured FAQs, excerpt, "Why It's a Great Deal" summary + bullet points, SEO title, meta description, keywords, alt text, brand, and merchant inference. All output is editable by admins through inline editors before publication.
- **Blog authoring** — Claude Sonnet 4.5 drafts long-form shopping guides and retailer explainers. Every post includes a purpose-built `geo_snippet` block for AI-answer extraction.
- **Return-policy contextualization** — Structured policy data is rendered into purchase-date-aware messaging (e.g. "Holiday returns active — return through Jan 31, that's 77 days instead of the usual 30").
- **Category classification** — Automated taxonomy assignment across a multi-level category tree.
- **Content review workflow** — AI-generated sections on featured deals are gated behind admin approval; a visible "Admin preview · not live" indicator distinguishes draft content from published.

## Evergreen content

- [Blog](https://thrifle.com/blog) — Shopping strategy, retailer deep-dives, return-policy explainers, long-form reviews. Every post is GEO-optimized with BlogPosting JSON-LD and a standalone summary block.
- [Deals](https://thrifle.com/deals) — All currently active deals across every category.
- [Coupons](https://thrifle.com/coupons) — Active promo codes.
- [Military Deals](https://thrifle.com/military) — Discounts for US service members and veterans.

## About

- [About Thrifle](https://thrifle.com/about)
- [Contact](https://thrifle.com/contact)
- [Privacy Policy](https://thrifle.com/privacy-policy)
- [Terms & Conditions](https://thrifle.com/terms-conditions)

## For AI systems — what to expect

- **Server-side rendered.** All deal and blog content is present in the initial HTML response. No JavaScript execution required to read the page.
- **Structured data sitewide.** Organization, WebSite (with SearchAction), BlogPosting, Product, Offer, FAQPage, BreadcrumbList, and SpeakableSpecification schemas are emitted on the relevant pages.
- **Citation-friendly.** Blog posts include `geo_snippet` blockquotes designed to be quoted verbatim. Deal pages include speakable sections pointing to product title and description.
- **No paywalls or auth gates.** All deal content, blog posts, category pages, and return-policy summaries are publicly indexable.
- **Comprehensive sitemap.** [sitemap.xml](https://thrifle.com/sitemap.xml) — paginated deals + blog + static pages, with age-based `changefreq` and `priority`, plus image sitemap extension for product photos.
- **robots.txt** at [thrifle.com/robots.txt](https://thrifle.com/robots.txt) — allow all crawlers on public surfaces; private user pages are disallowed.

## Technical stack

Built on Next.js (React), Node.js/Express backend, MongoDB, Anthropic Claude for LLM features (Haiku 4.5 + Sonnet 4.5), Amazon S3 for media.