Michaels spent six weeks building an AI shopping assistant that now drives a conversion rate more than double that of ordinary site search. The craft retailer is not the story. The story is that Google Cloud’s Gemini Enterprise for Customer Experience has quietly become the default AI shopping-assistant layer for large US retail, with Michaels now the fourth major chain to report hard performance numbers on the same underlying platform in 2026.
What Michaels Actually Built
Michaels unveiled “Ask Mike” on July 21, an AI-powered shopping assistant built on Google Cloud’s Gemini Enterprise for Customer Experience and available on Michaels.com and its iOS and Android apps. Customers describe a creative project in plain language, such as planning a kid’s birthday party or starting a DIY build, and the assistant returns curated product recommendations instead of a list of keyword-matched search results.
The retailer, which operates more than 1,300 stores across 49 states and Canada, said the assistant quietly went live in May and generated nearly 75,000 conversations within weeks, with more than 60% of interactions centered on product discovery. Heather Bennett, Michaels’ president and chief customer officer, said the tool has been used for “everything from party planning to DIY projects.” Paul Tepfenhart, Google Cloud’s global director of retail industry strategy and solutions, credited the build timeline directly: Michaels went from concept to a production-grade AI assistant in six weeks.
The performance numbers are what separate this from a routine chatbot rollout. Shoppers who engage with Ask Mike convert, meaning they complete a purchase after the interaction, at a rate more than double that of shoppers using traditional search. Michaels also said 27% of assistant interactions lead directly to a product click or a cart addition.
Michaels Is Not an Outlier. It Is a Pattern.
Three other large retailers made functionally the same bet on the same Google Cloud product earlier in 2026, each with its own performance claim attached:
The Home Depot
In January, The Home Depot and Google Cloud expanded Magic Apron from a product-page assistant into a full conversational companion built on Google Cloud’s Gemini models and Gemini Enterprise for Customer Experience, adding real-time local inventory lookups and aisle-level product navigation alongside a materials-list generator for professional customers. Jordan Broggi, the retailer’s EVP of customer experience, described the goal as putting “Orange Apron expertise in the pocket of every customer.”
Macy’s
Macy’s launched “Ask Macy’s,” a Gemini-based conversational agent, in March after a December pilot, built with Google Cloud in a compressed four-week engineering sprint once Gemini Enterprise for Customer Experience became available mid-project. The retailer has said customers who use the assistant, which handles product discovery, recommendations, and virtual apparel try-on, spend roughly 4.75 times more than customers who do not.
Ulta Beauty
In April, Ulta Beauty and Google introduced Ulta AI, a shopping assistant on Ulta.com built on Gemini Enterprise for Customer Experience that draws on data from the retailer’s more than 46 million loyalty members to personalize guidance. Google and Ulta paired it with an agentic-commerce integration that lets shoppers complete eligible purchases inside Google’s AI Mode in Search and the Gemini app itself, extending the assistant beyond the retailer’s own site.
Four different retail categories, craft, home improvement, department store, and beauty, converged on the same vendor and the same underlying product inside a single year. That is a platform decision, not a coincidence, and it says more about where Google Cloud has positioned Gemini Enterprise for Customer Experience in retail than any single launch does.
What It Means for the Marketing Leader
Three implications follow from four retailers reporting conversion lift from the same AI layer within months of each other.
First, the shopping assistant is becoming its own channel, sitting alongside site search and category navigation rather than replacing them, and it appears to be earning a materially higher intent-to-purchase rate than either. A marketing leader evaluating on-site AI should treat the assistant as a measurable acquisition and conversion surface with its own funnel, not a customer-service add-on, and should build attribution for it accordingly, as our reporting on brand visibility inside AI answers has already flagged as an emerging measurement gap.
Second, vendor concentration is now a real strategic question. When four competitors in adjacent categories run the customer-facing discovery layer on the same third-party platform, the differentiation shifts from the technology itself to the first-party data and merchandising logic fed into it, which is exactly what Ulta pointed to when it tied its assistant’s personalization to its loyalty program data rather than to the underlying model.
Third, the build timelines matter as much as the results. Michaels shipped in six weeks and Macy’s in four, both citing Gemini Enterprise for Customer Experience as a packaged product rather than a custom build. That collapses the cost and time argument that has historically kept conversational commerce out of reach for all but the largest retail technology budgets, which raises the governance question our coverage of AI agent guardrails inside marketing platforms has been tracking: speed of deployment is outrunning the internal controls most marketing teams have in place for agentic tools.
What To Do Next
Marketing and ecommerce leaders evaluating an AI shopping assistant should ask vendors for a side-by-side conversion comparison against existing site search, not just engagement volume, since that is the metric Michaels, Macy’s, and the others are actually reporting. They should also map what first-party data the assistant will draw on before selecting a platform, since that data layer, not the underlying model, is emerging as the real point of competitive differentiation. Retailers still running static, keyword-matched search should treat this cluster of results as evidence that the category has moved, and should budget a pilot now rather than wait for a fifth competitor to publish the same numbers.
Source: Michaels Press Room
