Try asking an assistant for a product like yours, using the words a customer would use. Include a budget, a size or a practical constraint. Then look at what it recommends and which details it gets right.
That small exercise often reveals a more useful starting point than an “AI visibility score.” Perhaps your product is absent. Perhaps it appears with an old price. Or perhaps the assistant can’t work out which variant fits the request.
Those are different problems, and they need different fixes.
Check that the channel is available to your store
Shopify’s agentic storefronts documentation explains how eligible products reach supported AI channels and where merchants manage access. Requirements and checkout behavior vary by channel.
For ChatGPT, Shopify currently describes a discovery flow that sends shoppers to your online store checkout. Check the Agentic section in your own admin and the channel’s requirements before assuming your products are included.
Catalog access is an opportunity to be discovered. It doesn’t guarantee placement in an answer.
Record a useful baseline
Choose five questions your customers might ask before buying. A lamp store could try “a reading lamp for a narrow bedside table” or “a dimmable desk lamp with a replaceable bulb.”
Keep a simple record:
- The exact question and date
- The assistant used
- Products and brands mentioned
- Linked sources
- Incorrect or missing facts about your products
- The page a shopper reaches after clicking
Repeat occasionally, but expect variation. A single answer is a sample, not a stable search position.
Compare that record with identifiable AI referrals and orders in your analytics. The two won’t match perfectly: some recommendations produce no click, and some visits arrive without clear attribution.
Fix missing product facts first
Read your most important product pages as if you knew nothing about the brand.
Can you find the dimensions, materials, compatibility and included parts? Are variants named clearly? Does the description explain who the product suits and any limitations that could change the purchase?
A customer choosing bedding needs dimensions and care instructions more than another promise of luxury. A buyer looking for cookware needs to know whether it works on their hob.
Use support questions to find gaps. Add concise answers where the buying decision happens. Don’t add a long FAQ just to make the page look optimized.
Keep catalog records and visible pages consistent
Prices, availability, variants, shipping and returns need to agree across the places they appear.
Product feeds are another source of information. OpenAI’s feed documentation emphasizes current catalog data for product discovery. For Shopify-managed channels, check Shopify’s catalog setup and mappings rather than assuming you need to build your own separate feed.
Validate Product structured data on representative pages, including variants. If a review app provides markup, check that it describes the reviews actually shown. Avoid duplicate or conflicting output from the theme and apps.
Google’s AI-search guidance does not require special AI schema. Clear FAQ content may help readers, but missing FAQ markup isn’t evidence that your store is losing ChatGPT recommendations.
Give people evidence outside your store
Independent reviews, useful comparisons and credible coverage help customers assess a product. Look at the sources that appear in your test questions to understand what information is available in your category.
Where appropriate, make products available for an honest review and give reviewers the specifications they need. Publish original findings if you have a sound method and something useful to share. Avoid creating numbers just to give a post the appearance of research.
No one can promise that a particular mention will lead to a recommendation.
Check the page after the click
A correct recommendation still needs a usable landing page. Test the selected variant, stock status, delivery details and purchase action on a phone.
Our AI-referred shopper analysis looks at why that landing experience matters. The storefront performance guide covers slow pages and interactions.
If you’re considering llms.txt, read its limitations and setup options before spending on it. It belongs behind accurate product information in the work queue.
Start with ten important products. Fix facts, resolve catalog conflicts and review the landing experience. Keep the prompt record and referral baseline so you can see what changes. Our broader guide to AI citations explains the source side of the same work.