# Michaels AI assistant doubled search conversion by surfacing project kits, not just SKUs *Google Gemini turned browse into bundles — and the unit economics improved even before the second click.* By **Jenny Huang Goodman MPA MSc MHSA, Principal** — The Stash Edge, Hako Shikin LLC. Published 2026-07-27. Canonical: https://www.pops4.com/stash/articles/michaels-2026-07-27t09-5 Subject: Michaels Tags: conversion, ai, search, bundling, product discovery, chatbot --- Michaels reported that its new Google Gemini-powered AI assistant doubled the conversion rate of traditional on-site search, according to Digiday. The craft retailer deployed the assistant to answer natural-language queries — "what do I need for a beginner macramé project" — and return grouped product sets instead of isolated SKU results. The result: shoppers who engaged with the assistant converted at **twice the rate** of users who ran a standard keyword search. The mechanics are straightforward. A visitor types a project or problem. The assistant, built on Gemini's conversational model, returns a curated list of items that solve the entire task: rope, scissors, mounting hardware, an instruction PDF. Michaels skips the multi-query slog — search beads, search wire, search clasps — and puts the whole kit in front of the customer in one interaction. The assistant surfaces SKUs the retailer already stocks; no new inventory required. The conversion lift stems from three compounding factors. First, the assistant collapses decision fatigue. A shopper who lands on a category page with **40 yarn SKUs** faces choice paralysis. The assistant shortlists **three relevant items** based on the stated project, cutting cognitive load and shortening time to cart. Second, the bundle structure raises average order value. A single-SKU search might yield one $8 purchase. A project kit — paint, brushes, canvas, primer — pushes the basket to $45 before the customer realizes they are buying four items. Third, the assistant answers the unasked question: do I have everything? Traditional search cannot do this. A keyword hit returns matches; it does not audit completeness. The assistant does, which means fewer abandoned carts from incomplete orders and fewer post-purchase support tickets. A small physical-product brand can run the same play without licensing Gemini. Start with a **project lookup table**: a spreadsheet that pairs common customer questions to product bundles you already sell. If you ship camping gear, map "first overnight trip" to tent, sleeping bag, headlamp, stakes. If you sell baking supplies, map "sourdough starter kit" to flour, jar, scraper, thermometer. Load this table into a simple chatbot interface — Typeform, Landbot, or a custom Shopify app using OpenAI's API at **$0.002 per interaction**. When a visitor asks a question, the bot matches intent to a row in your table and returns the bundle as a single add-to-cart action. Track which questions arrive most often and expand your table weekly. Budget: **$150/month** for the chatbot platform, **$50/month** for API calls at **1,000 interactions**. If your average order is **$60** and you convert **10% of chatbot users** vs. **5% of search users**, you net an additional **$300/month per 1,000 visitors** after costs. The assistant also opens a second revenue line: post-purchase project suggestions. After a customer buys the starter bundle, the same conversational interface can propose a **next-level project** that reuses some items and adds **two new SKUs**. Michaels has not disclosed whether it runs this loop, but the architecture supports it. For a small brand, this is a **triggered email** three days after delivery: "Loved the beginner kit? Here's the intermediate version" with a **one-click upsell**. The conversion rate on these follow-ups typically sits at **8-12%** because the customer has already validated interest and received the first shipment. The broader pattern is this: search is now a product configurator, not a keyword router. Brands that treat it as such — surfacing solutions instead of matches — capture intent earlier and convert it faster. Michaels proved the unit economics at retail scale. A direct brand can prove it in a weekend with a spreadsheet and a $20 API key. ## The takeaway Map customer questions to product bundles, surface them through a simple chatbot, and double search conversion without new inventory. --- ## Publisher **Hako Shikin LLC** — Virginia Beach, Virginia. Founded 1997. ASI 217876 · DUNS 18-204-6339. Principal and author: **Jenny Huang Goodman MPA MSc MHSA**. - Author: https://www.huanggoodman.com/about - LLM context: https://www.pops4.com/stash/llms.txt - MCP endpoint, for AI agents: https://mcp.pops4.com/mcp - Client dashboard: https://dashboard.pops4.com/ - Catalogue: 70,000+ products, 200+ brands