How I Built GiftHive: An AI Gift Finder That Actually Picks Gifts People Want

Most "AI gift finders" are a search box with a chatbot glued on. I wanted to build something different — GiftHive, a quiz-driven gift recommender that ranks real Amazon products by who the recipient actually is, not just keywords. In this post I'll walk through the architecture, the conversion tricks I learned shipping it, and the bits I'm proudest of. The Problem Picking gifts is emotionally expensive. You scroll Amazon for an hour, second-guess every option, and end up buying a gift card. Existing tools don't help because they optimize for keyword match, not recipient fit. GiftHive flips the input: instead of "show me gifts under $50", you answer a 30-second quiz about the person (relationship, interests, occasion, budget) and get a ranked shortlist with explanations of why each gift fits. Stack Next.js (App Router) — SSR for fast first paint, RSC for product data Tailwind CSS — design system + dark mode via CSS variables Cloudflare Pages — edge-deployed, free tier covers the traffic Amazon Associates — affiliate revenue model The Funnel The whole site is a 3-step conversion funnel: Landing page — exit-intent modal + social proof toasts prime the visitor Quiz — 30-second, one-question-per-screen flow, no login Results — ranked products with countdown bar and "X people found gifts this week" social proof Every step has a single primary CTA. The exit-intent modal is route-aware — it only fires on / and stays silent on /quiz and /results so it never interrupts the funnel mid-flow. That bug cost me ~15% of quiz completions before I caught it. Personalization Logic Each quiz answer maps to a vector of attributes (interests, style, budget, relationship). Products in the catalog have matching tags. Ranking is a weighted score: score = tag_overlap * w1 + budget_match * w2 + occasion_match * w3 No ML model needed — a few hundred products and clean tagging is enough to feel personal. Amazon Affiliate Integration Every product link runs through getAmazonUrl() which: Checks if the URL already has a tag= param — if so, replaces it with ours Otherwise appends ?tag=gifthive08-20 Falls back to an Amazon search URL if no product URL exists export function getAmazonUrl(gift: Gift) { const AFFILIATE_TAG = "gifthive08-20"; if (gift.amazonUrl) { return /[?&]tag=/i.test(gift.amazonUrl) ? gift.amazonUrl.replace(/([?&])tag=[^&]*/i, `$1tag=${AFFILIATE_TAG}`) : `${gift.amazonUrl}${gift.amazonUrl.includes("?") ? "&" : "?"}tag=${AFFILIATE_TAG}`; } return `https://www.amazon.com/s?k=${encodeURIComponent(gift.name)}&tag=${AFFILIATE_TAG}`; } Every ASIN in the catalog is real and verified, so clicks register in the Associates dashboard. Conversion Optimization A few things that moved the needle: Exit-intent modal with a 15-second arm delay so it doesn't fire on bounce-and-leave Social proof toast ("12 people found a gift in the last hour") in gentle mode on results Countdown bar that creates urgency without being sleazy Dark mode matching the user's system preference — warm palette instead of pure black Deployment Deployed on Cloudflare Pages via wrangler. The default *.pages.dev domain works fine, but some startup directories (like BetaList) reject it as "free hosting" — something to keep in mind if you're planning a launch there. What's Next A/B testing CTA copy Localized quiz for non-US markets A "gift recipient profile" save feature Try It GiftHive is live at https://gifthive.pages.dev — I'd love feedback, especially on the quiz flow and the quality of recommendations. If you're building something with a similar funnel, the biggest lesson was: route-aware components beat global components. A social proof toast that fires on every page feels spammy; one that only fires on /results feels like proof. Happy hacking!