Michaels reported that a Google Gemini-powered AI search assistant doubled the conversion rate compared to traditional keyword search on its e-commerce platform, according to Digiday. The craft retailer deployed the assistant to field natural-language questions — "what do I need for a beginner watercolor set" — instead of forcing shoppers to guess keywords like "watercolor paint."

The assistant runs on Google's Gemini model and sits inside Michaels' site search. A customer types a question or describes a project. The AI returns a curated answer with specific products, often packaging multiple items into a starter kit or project bundle. Michaels did not disclose absolute conversion figures or the sample period, but characterized the 2x lift as coming from early deployment data shared with Google.

The mechanism is decision compression. Craft and hobby categories suffer from paradox-of-choice friction: a shopper looking to start resin jewelry faces hundreds of SKUs across resins, molds, pigments, and tools, with no clear entry point. Traditional keyword search returns a grid of products sorted by relevance or price. The shopper still has to decide which six items make a coherent starter set. That cognitive load kills conversion. Gemini collapses the decision tree by answering the implied question — "what do I actually need" — in one structured response. The shopper moves from intent to cart in fewer clicks because the assistant did the product selection work.

The secondary effect is margin mix. When the AI bundles products, it steers the basket toward Michaels' private-label items and higher-margin accessories that a keyword search might bury on page three. The shopper perceives value — a curated answer — while Michaels captures better unit economics than if the same shopper had cherry-picked the lowest-priced items from a keyword grid. The conversion doubles not just because friction drops, but because the basket composition shifts toward purchase-ready assortments.

A small physical-product brand runs the same play without Google's infrastructure by turning product pages into answer pages. Instead of a grid of colors or sizes, write the answer to the question a confused buyer is asking. If you sell camping cookware, the product page for a pot set should open with: "For two people on a weekend trip, you need a 1.5L pot, a frying pan, two bowls, and a compact stove. This set includes all four." That's decision compression in 35 words. Shopify stores can add a simple chatbot using tools like Tidio or Gorgias AI for $25-50/month, trained on five FAQ answers you write yourself. The bot fields "what do I need for X" and returns the same curated answer. The small brand sacrifices some browse — customers who want to pick every component — but converts the larger segment who just want someone to tell them the right answer. Total setup cost under $100 and four hours of writing.

The broader pattern: search is shifting from retrieval to recommendation. Keyword search assumed the customer knew what to ask for and just needed the database to surface it. AI search assumes the customer has a goal but doesn't know the product taxonomy. The brand that answers the goal — not the keyword — owns the conversion. Michaels proved it at scale. A one-person brand can prove it this week by rewriting three product pages as answers instead of specs.