Michaels, the $5.3 billion crafts retailer, deployed a Google Gemini-powered AI assistant on its site and saw conversion rates double compared to traditional keyword search, according to Modern Retail. The company did not disclose absolute conversion figures, but the 2x lift marks one of the first documented wins for conversational AI in physical-product retail search.

The assistant works differently from the keyword box. Instead of typing "paint brushes," a customer asks full questions: "What do I need for a beginner watercolor set?" The AI returns a curated list of products with explanations, not just keyword matches. The user stays in a single session, refining the query without starting over. Michaels reported the assistant also increased average order value and basket size, though it did not release those numbers.

The mechanism is simple: conversational AI removes the cognitive load of search. A customer buying physical product rarely knows the exact SKU or category name. She knows the project. Traditional search punishes vague queries with irrelevant results, triggering exits. The AI assistant maps intent to product without requiring the shopper to learn the retailer's taxonomy. Each refinement narrows the set instead of resetting it. The session feels guided, not filtered. Conversion lifts because the path from question to cart is shorter and the match is better.

Michaels is not the only retailer testing this. Instacart, Shopify, and Amazon have all deployed conversational search in the past year. The difference here is the disclosure: Michaels named the 2x conversion lift in public, giving smaller brands a benchmark. The play is now documented and copyable.

Here is how a small physical-product brand runs the same strategy without Google's infrastructure. Start with your email or SMS list. When a subscriber asks a question — in reply, in a DM, on live chat — answer with a short paragraph and a link to three specific products, not a category page. Do not default to "check the site." Write the recommendation as if you were in the aisle with them. Track which conversations convert. After 50 interactions, you will see patterns: the phrasing that leads to purchase, the products that pair, the objections that stop the sale. Codify those into a simple decision tree. Use a low-cost tool like Typeform or Landbot to build a conversational quiz on your site or in your welcome email. Ask the project, suggest the kit, link the cart. You do not need a large language model. You need to stop making the customer translate their intent into your product taxonomy. The lift comes from removing that tax.

The broader pattern is this: search is becoming a conversation, and the brands that script the conversation control the conversion. Michaels proved the mechanism at scale. Now the small brand copies the behavior, not the budget.