AI commerce foundations

How to audit a Shopify store for AI readiness

A manual methodology for auditing a Shopify catalog before automating anything — what to check, in what order, what tools to use, and where the gaps usually hide.

10 min read Updated May 10, 2026

The instinct on most catalogs is to start fixing. The instinct on most catalogs is also wrong. A two-hour manual audit, done in the right order, surfaces the gaps in roughly the order an operator should fix them — and almost always exposes one or two issues that would have been wasted effort to bulk-fix until they were named.

This guide is a manual auditing methodology for a Shopify catalog. The work is sequential: each pass builds on what the previous pass found. The audit fits comfortably in one focused afternoon for a catalog under a few thousand SKUs, and the artifacts are copy-pasteable into a follow-up plan.

The walk-through covers six passes, in order:

1. Crawl access
robots.txt + bots

2. Structured data
validity + completeness

3. Catalog hygiene
titles, descriptions, images

4. Feeds
GMC + Microsoft + others

5. Surface checks
real query tests

6. Gap report
prioritized fixes

Each pass should land in 15–30 minutes if the catalog is not pathological. The full pass takes 2–3 hours. The output is a prioritized punch list, not a set of fixes — fixing happens after.

Before you start

Three things to have at hand:

  1. Admin access to the Shopify store (a staff account with view permission is enough; you do not need theme editor access for the audit).
  2. A Google Search Console property verified for the storefront domain. Optional but useful — surfaces crawl errors and rich-results status the manual passes cannot reach.
  3. A second browser profile or incognito window for verifying live pages without admin-session cookies. Some structured-data checks behave differently when viewed under admin context.

A scratch markdown file open in a separate window for taking notes saves time on the writeup at the end. Each pass ends with a short list of findings; the gap report is the merge.

Pass 1 — Crawl access

The first question is whether AI surfaces can fetch the catalog at all. A catalog blocking the major AI crawlers in robots.txt has a visibility problem at the discovery layer, and nothing downstream matters until that is resolved.

Fetch the live robots.txt:

curl -s https://yourstore.com/robots.txt

Look for any of these patterns:

User-agent: GPTBot
Disallow: /

User-agent: ClaudeBot
Disallow: /

User-agent: PerplexityBot
Disallow: /

User-agent: anthropic-ai
Disallow: /

If any major AI crawler is blocked, decide whether that was intentional. Some catalogs block deliberately — IP protection on sensitive content, ongoing licensing negotiations, contractual obligations to partners — and those decisions stand. A different share of catalogs end up blocking inadvertently after copy- pasting a robots.txt from a publisher template or a competitor’s site. The audit’s job is to surface the rule and let the operator decide; the action depends on which case applies.

Also check for the inverse — User-agent: * with Disallow: / or large Disallow: blocks that catch product paths.

Finding template: “Discovery — robots.txt blocks GPTBot and ClaudeBot. No documented reason. Recommend reverting both.”

Pass 2 — Structured data

The second pass is whether the catalog’s structured data is parseable and complete. Pick three to five products spanning the catalog’s variety — bestseller, mid-tier, recent addition, a variant-heavy product, an item from a thin category. The sample matters: a clean bestseller can mask broken variants.

For each sample product:

  1. View source on the live product page. Find the <script type="application/ld+json"> block.
  2. Paste it into the Schema Markup Validator. Note any syntax errors or unrecognized properties.
  3. Run the URL through the Rich Results Test. Note any required-property warnings.
  4. Inspect the structured data manually for completeness. At minimum, look for name, description, image, offers (with price, priceCurrency, availability), brand, sku, and at least one identifier (gtin13, gtin12, gtin8, or mpn).

Common findings on Shopify catalogs:

See Product schema for Shopify for the property-by-property reference.

Finding template: “Structured data — 3/5 sampled products missing GTIN. priceValidUntil set to 2024-12-31 across the catalog (likely a theme template). Variant products render as a single Product with no hasVariant.”

Pass 3 — Catalog hygiene

Once structured data is validated, the next question is whether the content the structured data references is itself good. This pass is about titles, descriptions, images, and attribute density.

For the same sample products, evaluate each on:

Titles. Does the title follow a structured pattern (brand + product type + defining attribute + variant)? Or is it marketing copy (“The Sweater You’ll Reach For”)? Marketing titles surface on marketing-intent queries; structured titles surface on shopping-intent queries. AI surfaces prefer the latter for the shopping context. See Writing product titles for AI agents.

Descriptions. Is the description specific enough that an embedding of it would land somewhere distinctive in vector space? A description that reads “Premium quality. Crafted with care. Order today.” embeds to the same generic location as a thousand other descriptions. A description that names materials, dimensions, use cases, and constraints embeds to a specific location that matches specific buyer queries.

Images. Each product needs at least one image; primary images need to be high-resolution, well-lit, and ideally on a neutral background. Check alt text — Shopify’s image alt defaults to the product title, which is usable, but specific alt text (e.g., “navy blue wool runner from the side”) is better for both accessibility and image-based retrieval.

Attributes. What attributes are present and which are missing? For apparel: material, fit, care instructions, season. For electronics: model number, certifications, technical specs. For home goods: dimensions, materials, weight. Each category has its own attribute stack — see the category-specific guides for the per-vertical reference.

Finding template: “Catalog hygiene — titles are 70% marketing-led, 30% structured. Descriptions average 90 words; attribute density low (no materials, no use cases). Images: alts default to titles, primary images mostly OK.”

Pass 4 — Feeds

If the catalog has a Google Merchant Center feed, audit it. GMC’s Diagnostics report surfaces disapproved items, attribute warnings, and policy issues. The specific things to check:

If the catalog has a Microsoft Merchant Center feed, run the same checks. If the catalog has feeds to TikTok Shop, Pinterest, Meta, or others, check those too.

Finding template: “Feeds — GMC has 47 disapprovals, all for missing GTIN. 312 warnings for missing brand. Feed updated 2 hours ago; freshness is fine. No Microsoft Merchant Center feed exists.”

Pass 5 — Surface checks

The most important pass. Test live queries against the actual AI surfaces. This is what an audit that stops at the technical layers will miss: a catalog that passes every technical check can still be invisible because the structured-data work has not propagated to the surface index yet.

Pick five queries that should plausibly return the catalog’s products:

  1. A branded query (e.g., “Acme Outfitters wool sweater”)
  2. A category + brand query (“Acme wool sweater for women”)
  3. A pure category query (“wool sweater for winter”)
  4. A constrained category query (“wool sweater under $200 for cold weather”)
  5. A use-case query (“warm sweater for hiking in cold rain”)

Run each query on:

Record for each query × surface combination: did the catalog appear? If yes, in what position? If no, what catalogs appeared in its place? Knowing the competitive set is half of knowing what to fix.

The branded query is the floor — if the catalog does not appear for its own brand name on a major surface, the issue is at the discovery or indexing layer (likely Pass 1 or Pass 2). Category queries are the ceiling — appearing in these against established competition takes time.

Finding template: “Surface checks — branded query returns the catalog on ChatGPT and Google but not on Perplexity (likely a schema validity issue affecting Perplexity specifically). Category queries return three named competitors; our catalog does not appear. Use-case query returns no specific products on any surface.”

Pass 6 — Gap report

The last pass is the merge. Take findings from each of the five preceding passes and group them by priority:

Now (this week).

Soon (this month).

Eventually (this quarter).

The structure is intentional: the “now” tier is for issues that are suppressing the entire catalog. The “soon” tier is for issues that are suppressing meaningful subsets. The “eventually” tier is incremental improvement.

Where this audit breaks down

Three known limitations to flag in the writeup:

Tiny catalogs (under 50 SKUs). The sample-of-five from Pass 2 covers most of the catalog. The methodology still works but the gap report is closer to a per-product punch list.

Catalogs with content gated behind authentication. If a catalog requires a login to view products (some B2B catalogs do), AI surfaces cannot crawl past the login wall. The audit reveals the structural issue but cannot reach the products themselves.

Catalogs with very recent migrations. A catalog migrated to Shopify within the last 30 days may still be in transition — old URLs still indexed, new URLs not yet crawled. Some Pass 5 findings reflect the migration, not the steady-state.

What to do after the audit

The gap report is the input to fixing. Most “now” items can be addressed in a single dedicated sprint. The “soon” tier benefits from automation — bulk schema fixes, GTIN sourcing via a service, description enrichment via a tool. The “eventually” tier is ongoing.

The most useful follow-up reads:

Related