By Steve Merrill | May 29, 2026
Most of the conversation about AI and Shopify has focused on external discovery, getting products into ChatGPT, Perplexity, Google AI Mode. That conversation matters. But something happened inside Shopify recently that's worth paying attention to separately.
Third-party AI agents can now plug into Shopify Sidekick and manage product discovery on your own storefront. Not just surface recommendations, actively adjust which products shoppers see, when, and in what order, based on real-time behavior signals.
Nosto launched Agentic AI Workflows for Shopify Sidekick in May 2026, making it possible for commerce teams to guide personalization and product surfacing through conversational AI natively inside the Shopify assistant. This is a different layer than the external AI commerce stack most merchants are focused on right now.
What Are Agentic AI Workflows Inside Shopify?
Standard Shopify merchandising is rule-based. You configure which products show on collection pages, in what order, under what conditions. Rules don't change unless you change them. They don't adapt to what a specific shopper is doing in real time.
Agentic discovery is different. The AI monitors session behavior, purchase signals, inventory levels, and catalog attributes simultaneously, and adjusts what gets surfaced as the shopper moves through your store. A shopper browsing gift-related content gets different products surfaced than someone going deep on a specific category. No manual rule update required.
According to Nosto's announcement, merchants using agentic workflows can now direct product discovery through natural language inside Sidekick: "Promote our new spring line to returning customers" or "Suppress out-of-stock variants from collection pages." The AI executes those adjustments across the storefront in real time.
What Does Your Catalog Data Need to Feed This?
This is where most merchants will find the gap. Agentic systems make decisions based on the data available in your catalog. If that data is incomplete, inconsistent, or written for marketing impressions rather than structured decision-making, the agent can't route effectively.
It's the same principle that drives external AI shopping recommendations. Shopify's product data documentation outlines the fields agents use, but the difference between good and bad agentic output almost always comes down to whether those fields contain useful signal or filler.
Specific things that matter:
Brand name in every product title. Agents use this to understand catalog relationships. Without it, products look unconnected to each other and to the brand's broader identity signal.
Accurate, specific product descriptions. Not "elevate your routine with our premium formula." Something like: "Daily face wash for combination skin, oil-control formula, fragrance-free, dermatologist-tested." Agents use attribute language to route, not brand copy.
Consistent taxonomy and tags. If "women's shoes" is tagged 14 different ways across your catalog, the agent can't route collection discovery reliably. Standardize before enabling agentic workflows.
Real inventory signals. Agentic systems are increasingly inventory-aware. Promoting out-of-stock products in surfacing decisions is a fast way to build bad shopper experience data that reduces the agent's effectiveness over time.
How Is On-Site Agentic Discovery Different From ChatGPT or Perplexity?
Different surface, different optimization. They depend on each other but aren't the same thing.
ChatGPT and Perplexity discover your products from outside your store, from your product feed, schema markup, structured data, and external content about your brand. A shopper asks an AI assistant for product recommendations and your products appear (or don't) based on how readable and complete your external data is.
On-site agentic discovery happens after the shopper is already on your store. It determines what they see in search results, collections, and recommendation carousels, and adjusts in real time based on behavioral signals that external AI doesn't have access to.
Both matter. The mistake is treating them as competing priorities. Your product data quality drives both outcomes. Shopify's own Sidekick documentation makes clear that the quality of catalog data is the first dependency for any AI-assisted merchandising, internal or external.
Who Benefits Most From This Right Now?
Stores with 50-500 SKUs are the clearest immediate fit. Large enough to have meaningful routing decisions, where the agent can meaningfully differentiate what gets shown to whom. Small enough that the catalog can be audited and cleaned without an enterprise-scale data project.
Stores with 10 SKUs don't need this. Stores with 10,000 SKUs need it more but also have more infrastructure to set up first.
If you're in the middle, the place to start is the same regardless: audit your product data before you connect any agentic system to it. An agent running on bad data produces bad discovery outcomes fast, and at scale. Get the data right first.
What Should Merchants Do This Week?
Three things, in order:
First: Pull your products.json and review your top 50 SKUs for attribute completeness. Are titles structured? Are descriptions specific? Are metafields populated? This takes an hour and will tell you how ready your catalog actually is.
Second: Check your Shopify Agentic settings (Settings > Apps and sales channels) and confirm your catalog is actively syncing. This is the data layer all third-party agentic integrations access. It should be active and current.
Third: Pick one surface, your main collection page or your search results, and look at what's currently being surfaced to first-time visitors. If it's generic best-sellers with no behavior personalization, that's the gap an agentic workflow closes. That's where the test starts.
Frequently Asked Questions
What are Nosto's Agentic AI Workflows for Shopify Sidekick?
Nosto's Agentic AI Workflows allow AI agents to plug into Shopify Sidekick and automatically manage product discovery and personalization on your storefront. Instead of manually setting merchandising rules, the AI monitors behavior signals and adjusts which products get surfaced to which shoppers in real time.
How does AI-managed product discovery differ from standard Shopify merchandising?
Standard Shopify merchandising is rule-based: you set which products appear where, in what order. AI-managed discovery is behavior-adaptive, the system adjusts rankings and surfacing in real time based on session signals, purchase history, inventory levels, and catalog attributes. It responds faster and doesn't require manual rule updates.
What product data do AI agents need to make good merchandising decisions?
AI agents need complete, structured product data: brand names in titles, accurate and specific descriptions, populated metafields, consistent category tags, and available inventory signals. Aspirational marketing copy doesn't help agents make decisions. Specific, attribute-rich data does.
Does this work for small Shopify stores or only large ones?
Agentic AI workflows are most immediately useful for stores with 50-500 SKUs, large enough to have meaningful routing decisions, small enough that the catalog can be cleaned and standardized without enterprise infrastructure. Very small catalogs don't need it yet. Very large ones need it but require more setup.
How is on-site agentic discovery different from ChatGPT or Perplexity discovery?
ChatGPT and Perplexity discover your products from outside your store, from your product feed and schema data. On-site agentic discovery happens inside your store, adjusting which products shoppers see when they're already browsing. Both depend on good product data, but they operate on different surfaces and require different optimization approaches.
Want to know how your store's AI readiness stacks up?
Check Your Store's AI Readiness →
