By Steve Merrill, Founder of WRKNG Digital — June 24, 2026
Shopify Sidekick Just Changed What It Does for Your Customers
Most merchants still think of Shopify Sidekick as a backend assistant. You ask it to write a product description, generate a discount code, or answer a question about your dashboard. That's the mental model baked in from 2023.
That model is outdated.
With Shopify's latest platform update, Sidekick has expanded its role to the customer side of your store. Shoppers can now ask Sidekick questions directly and receive product recommendations in response. Shopify's Sidekick overview describes the assistant as having live access to your catalog to answer questions in real time. What that means in practice: Sidekick is reading your product data and deciding what to recommend to your customers.
Your product data is what Sidekick knows about your catalog. If that data is thin, Sidekick is working blind.
How Does Shopify Sidekick Generate Product Recommendations?
Sidekick doesn't have a separate recommendation algorithm sitting apart from your catalog. It reads your product data and matches it against what a customer is asking.
When a shopper types "good gift for a 10-year-old who loves art," Sidekick scans your catalog for products with titles, descriptions, and tags that carry relevant signals: age context, audience language, use-case phrasing. It's looking for semantic overlap between what the customer wants and what your data says about each product.
If your titles are internal SKU codes and your descriptions say "Great product available in multiple sizes", there's nothing to match. Sidekick either surfaces the wrong product or returns something generic.
The ceiling on recommendation quality is set by the data you put in. That's not a Sidekick limitation. That's how every AI recommendation system works.
What Product Data Does Shopify Sidekick Actually Read?
According to Shopify's product details documentation, Sidekick can access the full set of product fields in your catalog. Not all of them carry equal weight.
Product Title, First thing read. Sidekick treats the title as the definitive statement of what the product is. Vague titles produce vague matches.
Product Description, The richest source of matching signals in your catalog. A 200-word description with specific materials, use cases, audience callouts, and feature details gives Sidekick far more to match against than a two-sentence summary.
Tags, Context signals about what the product is for. Tags like "gift-for-her," "summer-outdoor," or "toddler-safe" help Sidekick understand intent, not just category.
Product Type and Vendor, Categorical filters. Useful when a customer is browsing or comparing within a category. Mostly a secondary signal.
Metafields, Structured attributes: material type, fit, care instructions, age range, size guide. These produce the most precise matches because they're machine-readable and specific. Most stores have them mostly empty.
Variant Data, Size, color, weight, option values. Sidekick can't recommend "the lightweight version" if variant attributes aren't defined.
The Data Problems That Kill Recommendation Quality
Most stores fail this.
I've audited product catalogs for dozens of Shopify stores over the past six months. The same four problems show up every time.
Short descriptions. The average product description in a mid-size Shopify store runs under 60 words. That's not enough semantic content for intent matching. Sidekick is working with one or two sentences and making its best guess.
SKU-formatted titles. Titles written for warehouse management, not for customers or AI systems. "BLK-JCKT-M-047" tells Sidekick nothing about what the product is, who it's for, or what problem it solves.
Empty metafields. Shopify's metafields documentation shows how stores can attach structured custom data to every product in their catalog. Most stores don't use them at all. That's free structured data left sitting on the floor.
Incomplete variant data. A product with 12 variants where half carry generic labels like "Option 1" or blank weight fields is invisible to specific recommendation queries. Sidekick can't match what it can't read.
None of these are technical problems. They're data discipline problems. Fixable with time and a clear priority list.
How Do I Improve Shopify Sidekick Recommendations?
Five things. Do them in order.
Step 1: Rewrite Product Titles for Intent
A good title tells Sidekick what the product is, who it's for, and what it does. "Women's Merino Wool Running Sweater, Lightweight" outperforms "Sweater #4" in every matching scenario. Include material, use case, and target audience where relevant.
Don't stuff keywords. Write it the way you'd describe the product to a customer calling your store.
Step 2: Expand Descriptions to 150 Words or More
Target 150 to 300 words per product. Answer the questions customers actually ask. What is it made of? How does it fit? Who is it best for? What problem does it solve?
That language is what Sidekick scans when a shopper asks a specific question. The more specific and natural your description, the more likely Sidekick surfaces the right product at the right moment.
Step 3: Add Natural-Language Tags
Go through your existing tag library and replace internal codes with intent-based phrases. "STYLE-A4" means nothing to Sidekick. "minimalist-home-decor" or "eco-friendly-gift" does.
Think about what a customer would type into a search bar or ask a friend. Those phrases are your best tags.
Step 4: Complete All Variant Data
Every variant needs a clear title, defined option values, and where possible a measurement or weight attribute. If you sell apparel, add a size_type metafield. If you sell hardware or home goods, add dimensions.
Sidekick can match "small enough to fit in a carry-on" if your products have dimension data. Without it, that query returns nothing.
Step 5: Populate Metafields with Structured Attributes
This is the highest-value change most stores haven't made. Add metafields for: material, care instructions, fit type (for apparel), target age group, and primary use case.
Shopify's standard product metafield schema gives you a starting point. Custom metafields let you go deeper for your specific catalog. The structured data you add today is available to Sidekick immediately.
What This Looks Like Before and After
Earlier this year we ran a product data audit on a home goods store. Average description: 42 words. Metafields populated: 8% of catalog. Tags: mostly warehouse codes pulled from their ERP system.
After updating 200 top-SKU descriptions to 180-plus words, replacing internal tags with natural-language phrases, and populating material and use-case metafields, Sidekick's ability to match customer queries to specific products improved measurably. The "no relevant results" rate dropped. Customers were getting specific product suggestions instead of generic category links.
The starting point for that store was close to zero. It didn't have to be.
FAQ: Shopify Sidekick Product Recommendations
How does Shopify Sidekick recommend products?
Sidekick reads your product catalog, titles, descriptions, tags, metafields, and variant data, and matches that content against a customer's query. The accuracy of that match depends on how much useful signal is in your product data.
What product data does Shopify Sidekick use?
Titles, descriptions, tags, product type, vendor, variant option values, and metafields. Structured metafields give Sidekick the most precise matching signals. Unstructured description text is secondary but still accounts for a large share of the matching work.
How do I improve Shopify Sidekick recommendations?
Rewrite titles to include material and use case. Expand descriptions to at least 150 words. Replace SKU-based tags with natural-language intent phrases. Complete all variant data. Populate metafields with structured attributes: material, fit, target age, and use case.
Does Shopify Sidekick use metafields for recommendations?
Yes. Metafields give Sidekick machine-readable structured data it can match against specific shopper queries with more precision than scanning free-text descriptions. Stores without populated metafields are leaving their best matching signals unused.
What happens if my product descriptions are too short?
Descriptions under 60 words give Sidekick almost no intent signal to work with. Expect weak recommendation accuracy and a higher rate of generic or irrelevant results. Target 150 words minimum per product. That's enough semantic content to produce accurate matches across most query types.
We built a free AI Commerce Readiness audit that shows exactly where your product data is weak and what to fix first. It covers Shopify Sidekick, ChatGPT Shopping, Perplexity, and every AI system reading your catalog right now.
