Presence is automatic. Selection is not.
Shopify syndicates merchant catalogs to AI shopping surfaces through Shopify Catalog and agentic storefront integrations. You do not have to submit anything for your products to be eligible. Eligibility is not the bottleneck any more. Shopify's own documentation says as much: eligible products are automatically discoverable by AI channels through Shopify Catalog (Shopify Help Center).
Selection is. When someone asks an assistant for the best waterproof hiking boots for wide feet under $200, the assistant is running a filter. It has to establish four things: that the product is a hiking boot, that it is waterproof, that a wide fit exists, and that it costs under $200. Each of those is a lookup against a field.
If the width only appears in marketing prose, or the product category is blank, or the price feed is stale, the product fails the filter silently. No error, no visible penalty. It simply is not in the answer. Nothing below is a trick — it is filling in fields.
You do not have to take our word for how widespread that failure is. Adobe Analytics data, reported by TechCrunch on 16 April 2026, found that around 34% of retailer product pages cannot be properly accessed by AI, and that roughly a quarter of retailer homepage content is not optimized for LLMs (Adobe Analytics data, reported by TechCrunch, 16 April 2026). That is roughly a third of retail product pages failing before any of the field-level work below is even reached. Adobe's own write-up is not currently reachable, so that figure is quoted as TechCrunch reported it rather than from the primary source.
The same Adobe Analytics data, reported by TechCrunch on 16 April 2026, put AI traffic to US retail sites up 393% in Q1 2026 year over year. In March 2026 that traffic converted 42% better than non-AI traffic, with revenue per visit 37% higher. The analysis draws on over 1 trillion visits to US retail sites plus a survey of 5,000+ US respondents. Read the growth number carefully: a 393% rise says the channel is growing fast, not that it is large yet, and the conversion gap may partly reflect who is already using assistants to shop rather than the channel itself.
The field-by-field reference
Work down this table for any product you want recommended. The examples use a hiking boot, but the pattern transfers to anything.
This is not a list we invented. Shopify's documentation for agentic storefronts sets out what actually travels when a product is syndicated — title, description, options, images, price, availability, and other key attributes (Shopify Help Center). Every item on that list has a row below. A field left blank in your admin is a field that arrives blank on the other side.
| Field | Why an assistant needs it | What good looks like |
|---|---|---|
| Title | First and strongest signal of what the product actually is. A brand-only name is unmatchable. | Product type plus the distinguishing attributes: Trailform Ridge 6 Men's Waterproof Hiking Boot — Wide Fit, Nubuck Leather |
| Description | Where use case, materials and suitability live. This is what answers the vague half of a shopper question. | What it is, what it is made of, what it is for, who it suits, what it does not do. Specs written as facts, not adjectives. |
| Product category | How your catalog gets classified. A blank category means the product may not be considered for a category-shaped question at all. | Set from the Shopify Standard Product Taxonomy, at the most specific node that fits — hiking boots, not simply footwear. |
| GTIN / barcode | The global identifier that lets the same product be matched and reconciled across sources. | The 12–14 digit code from the manufacturer, entered per variant. Empty is better than guessed. |
| Brand / vendor | Needed to attribute a mention to you. No brand, no credit. | The real brand name on every product: Trailform. Not blank, not a placeholder, not your supplier's internal code. |
| MPN | Manufacturer part number. Useful for exact-match when a shopper names a specific model. | The manufacturer's own code per variant: TF-RG6-WD-M10 |
| Images + alt text | Alt text is machine-readable description. Multiple angles support visual matching. | Several images, each with descriptive alt text: Brown nubuck waterproof hiking boot, side profile, wide fit |
| Price & availability | Assistants filter on constraints like under $200 and routinely exclude out-of-stock items. | Accurate, current, and synced. Inventory that lags reality gets you recommended and then bounced. |
| Attribute metafields | Material, colour, size, fit, compatibility, dimensions and care are what attribute-heavy questions filter on. | Structured values, not prose: material: full-grain nubuck leather, width: wide (2E), weight: 1.2 kg per pair |
The Catalog Readiness Hub scans your catalog and scores each product against exactly this rubric, then tells you per product which of these fields are missing. It also drafts a suggested fix for each gap — but nothing is written to Shopify until you review and approve it.
Before and after: title and description
Most catalogs fail on these two fields, and both fail the same way — the copy is written to be evocative rather than informative.
Title
Weak: The Ridge 6 — Built For The Trail
An assistant reads that and learns nothing. There is no product type, no gender, no material, no waterproofing, no fit. It cannot be matched to a boot query because it does not say it is a boot.
Strong: Trailform Ridge 6 Men's Waterproof Hiking Boot — Wide Fit, Nubuck Leather
Brand, model, audience, product type, the key functional attribute, the fit, the material. Every one of those is a filter it can now pass. It reads slightly less romantic on a collection page and that is a trade worth making.
Description
Weak: The Ridge 6 is our best-selling boot. Premium quality, built to last, and the perfect gift for the adventurer in your life. Order yours today.
Premium quality is not an attribute. Built to last is not a spec. There is nothing here to match against.
Strong: The Ridge 6 is a men's waterproof hiking boot for multi-day trail use on rough ground. The upper is full-grain nubuck leather with a seam-sealed waterproof membrane and a gusseted tongue. It has a 5 mm lugged rubber outsole and a moulded midsole with ankle support, and weighs about 1.2 kg per pair in US 10. Available in standard (D) and wide (2E) widths, US 7–14. Suited to hikers who need ankle support and dry feet in wet conditions; it is heavier than a trail runner and not intended for fast-and-light use.
Note the last sentence. Saying what a product is not for is useful — it stops you being surfaced for questions you would lose anyway, and it gives the assistant a reason to recommend you confidently for the ones you win.
Write for attribute + use case + constraint
Shopper questions to an assistant are long and natural. They are almost never one keyword. They usually stack three things:
- Attribute — waterproof, wide fit, nubuck, 40 oz, dishwasher safe
- Use case — multi-day hiking, commuting, gifting, a small kitchen
- Constraint — under $200, in stock, ships to Canada, available in size 12
Your product data should be able to answer each part independently. Attributes belong in metafields and variants. Use case belongs in the description, stated plainly. Constraints come from price, inventory and variant data being accurate right now.
A useful test: take the three questions you most want to win, break each into its parts, and check whether a machine reading only your product record could confirm every part. If any part is only implied, it is not there.
Variants and metafields
Anything that genuinely differs between purchasable options — size, colour, width, capacity — should be modelled as a variant, not described in a sentence. A boot that says available in wide in the body copy but has no wide variant will not survive a wide-fit filter, and a shopper who is sent there will bounce.
Everything else that is a fact about the product belongs in a metafield with a consistent key and a consistent value format across the catalog. Use the same units, the same spellings, and the same vocabulary every time. 2E in one product and wide/2E in another and Wide fit in a third are three different values to a machine.
Pick the attribute set that matters for your category and apply it everywhere. For footwear that might be material, width, waterproofing, weight, closure and care. For a water bottle: capacity, material, insulation, lid type, dishwasher safe, dimensions. Six well-populated fields beat twenty half-populated ones.
Structured attributes are not a dead end inside your own admin. Shopify documents Catalog Mapping for metafields and metaobjects, which is the mechanism by which custom structured data is carried through to AI channels (Shopify Help Center). At the other end, OpenAI's commerce specifications describe structured product feeds used for discovery and checkout in ChatGPT, covering pricing, availability and seller context (OpenAI, product feeds). Both ends of that pipe expect fielded values. Prose does not survive the trip, which is why the same fact written in a paragraph and written in a metafield are not equivalent.
Consistency beats depth on one product
It is tempting to perfect your hero SKU. Resist it. A retrieval system builds confidence from the shape of the whole catalog, and a store where a quarter of products have categories and the rest do not is unreliable to draw from. Coverage is worth more than a single immaculate product page. A sane order of work:
-
Fix the blanks that block classification
Product category and brand/vendor, everywhere, first. These are the cheapest fixes and the most damaging gaps.
-
Rewrite titles for your top sellers
Product type plus distinguishing attributes. This is the single highest-leverage copy change in the catalog.
-
Populate the attribute set
Decide the metafield keys for each category once, then apply them consistently rather than improvising per product.
-
Then go deep on descriptions
Materials, use case, who it suits, what it is not for. Slowest work, so do it after the structural gaps are closed.
Check you are not blocking the crawlers
Product data quality is irrelevant if the crawlers cannot reach your site. AI crawlers are allowed or blocked through robots.txt — GPTBot, PerplexityBot, ClaudeBot and Google-Extended among them. Blocking them removes you from the answers those systems build from crawled web content. Check what your file currently allows before you invest hours in metafields.
The working checklist
Per product, top to bottom:
- Title states the product type, not just the model name
- Title includes the one or two attributes that distinguish it from your other products
- Product category set from the Shopify Standard Product Taxonomy, at the most specific node that fits
- Brand / vendor filled in with the real brand name
- GTIN / barcode entered per variant where one exists — left empty rather than guessed where it does not
- MPN entered per variant where the manufacturer provides one
- Description states what it is, what it is made of, what it is for and who it suits
- Description contains specs as facts, with units — no premium quality, no perfect gift
- Every purchasable difference exists as a variant, not only as a sentence
- Attribute metafields populated using the same keys and value formats as the rest of the catalog
- Multiple images, each with descriptive alt text
- Price current and inventory accurate
- Same standard applied to the whole catalog, not just the hero SKUs
None of this guarantees a mention. AI answers vary from run to run for the same question and the same catalog — that variance is real and worth measuring rather than wishing away. Cuebase tracks it two ways: Agent Channel Analytics uses a Shopify Web Pixel to capture AI-referred traffic and conversion by source, and Share of Voice sends real shopper prompts to ChatGPT, Perplexity, Gemini, Claude and Google AI to see how often your store comes up versus competitors.
FAQ
Do I need GTINs if I make my own products?
If you manufacture your own goods and nobody else sells them, there is no existing GTIN to look up. You can buy a prefix from GS1 and issue your own, which is worth it if you plan to sell through marketplaces or retail partners. If you do not, leave the barcode field empty rather than inventing a number. A fabricated GTIN can collide with a real product and cause the wrong item to be matched. Focus instead on the fields that carry the weight for a single-source product: product category, brand, a specific title, a complete description, and attribute metafields.
Does image alt text really matter for AI shopping assistants?
Yes, because alt text is the machine-readable description of the image. It is text a retrieval system can index, and it is often the only place a visual attribute like colourway, cut, or shape gets written down in words. It also serves accessibility, which is a good enough reason on its own. Write what is actually in the frame plus the identifying attribute, not a keyword list.
Will AI rewrite my product descriptions badly?
It can, if you let it write unsupervised. That is why Cuebase never writes to your store on its own. The Catalog Readiness Hub generates a suggested fix for each flagged gap, and you review every suggestion before anything is applied to Shopify. Suggestions are built from the product data you already have rather than invented, and anything that reads wrong for your brand can be edited or rejected. Treat generated copy as a first draft that saves typing, not as a publishing decision.
How many products do I need to fix before it makes a difference?
There is no threshold that guarantees anything, and AI answers vary from run to run even with identical data. What is reliable is the direction: a catalog where every product has a category, a brand, a specific title, real attributes and accurate stock is a catalog a retrieval system can use, and a half-filled one is not. Start with the products you actually want recommended, get those complete, then work outward until the whole catalog is consistent.
Should I block AI crawlers in robots.txt?
Blocking GPTBot, PerplexityBot, ClaudeBot or Google-Extended removes your site from the answers those systems produce from crawled web content. If your goal is to be recommended by AI assistants, blocking them works directly against that. Check what your robots.txt currently allows before you spend time on product data, because a blanket block undoes the work.
Sources
Every external figure quoted above is linked to its original publisher with the date attached, so you can check it and see how current it is. Where a publisher's own page is not reachable, we have named who reported the number rather than implying we read the primary source.
- Adobe Analytics data, reported by TechCrunch. AI traffic to US retailers rose 393% in Q1, and it's boosting their revenue too. 16 April 2026. Based on over 1 trillion visits to US retail sites plus a survey of 5,000+ US respondents.
- Shopify Help Center. Shopify Catalog and product discovery for agentic storefronts. Covers the attributes syndicated to AI channels and Catalog Mapping for metafields and metaobjects.
- OpenAI. Product feeds and commerce specifications. Structured catalog data for discovery and checkout in ChatGPT. No last-updated date is published for this page.
See which fields are missing, product by product
Cuebase scans your Shopify catalog against this rubric, scores every product, flags the exact gaps and drafts the fixes — which you review before anything is applied. Catalog readiness scoring and AI channel analytics are on the free plan.
Add to Shopify →