Example deliverable
Sample report
This is a practice audit of a real public store. The store’s name, product names, web addresses, and product codes have been removed. The score, findings, and fix list are from that practice audit. Some checks that need a store login are marked unverified.
AI Commerce Readiness Audit (sample)
Prepared for: a mid-size home goods brand (identity withheld)
Storefront: a public Shopify direct-to-consumer site selling blankets and related home/outdoor textile goods
Prepared by: LammyRu
Date: 22 November 2026 (Asia/Manila)
Audit type: Sample report, based on a real practice audit of public pages only. Brand name, product names, URLs, identifiers, and other uniquely identifying details have been removed or generalized so this document can be published as an example of LammyRu’s audit deliverable.
Overall score: 79 / 100
How to use this report
This audit reviews whether AI shopping agents — including ChatGPT shopping, Google AI Mode / Gemini shopping, Perplexity, and agents that speak documented agentic checkout standards — can find, understand, and buy from the store.
It is based on publicly reachable pages and files only. We did not log in, submit forms, start a checkout, or contact the business. Findings that require Shopify admin, Google Merchant Center, Search Console, or WAF logs are marked unverified.
Scores follow the LammyRu playbook (v1.0). Area F (page speed) could not be measured in this run (the official PageSpeed Insights API returned HTTP 429). That 5% weight was dropped and the other weights were rescaled to 100%.
1. Executive summary
This is a mid-size Shopify direct-to-consumer home goods brand (about 140 products / 200 variants in the public catalog snapshot). From the public internet, it already looks like a store that Shopify has wired for agentic commerce: a public Universal Commerce Protocol profile is live at /.well-known/ucp, /llms.txt and /agents.md tell agents how to use it, official policy URLs exist, and product pages carry rich Product / Offer markup with brand, SKUs, and valid GTINs. Prices and stock status agreed across the page, structured data, and public product APIs, including out-of-stock items that still showed a price.
The gaps that would still trip an agent are about conflicting or incomplete machine-readable facts, not a missing storefront. Return and shipping rules are split across official /policies/* pages and richer custom pages, and they do not fully agree (return window starting on purchase vs delivery; exchanges offered vs “no direct exchanges”). The structured MerchantReturnPolicy is only a URL plus the country name “United States” — not the ISO country code and not the required return-window properties Google needs for a return-details enhancement — and there is no OfferShippingDetails markup at all. A second Product JSON-LD graph on the same pages uses Shopify numeric IDs as an mpn. GTINs are on the page and in /products/{handle}.js, but they are omitted from the public /products.json catalog, which is a feed-quality risk if anyone builds a Google or OpenAI feed from that endpoint.
Overall score: 79 / 100 — usable, with clear high-impact fixes. Completing return/shipping markup, reconciling policy text, and making identifiers consistent across every export would move this into the mid-80s or better. Whether the store is actually listed in ChatGPT or Google AI Mode was not verified (that needs Shopify admin and Merchant Center).
2. Scorecard
| Area | Weight (original) | Score | Notes |
|---|---|---|---|
| A. Crawler and agent access | 12% | 90 | Catalog is allowed; Shopify agent instruction files are present and useful. |
| B. Structured data (Product / Offer) | 18% | 72 | Required Product/Offer fields are present, but return/shipping markup is incomplete and a second graph conflicts. |
| C. Identifiers and variants (GTIN, brand, SKU) | 15% | 80 | Brand and SKUs are complete; sampled GTINs are valid, but /products.json drops barcodes and a second graph misuses mpn. |
| D. Price and availability consistency | 12% | 90 | Page, schema, and public APIs agreed, including out-of-stock items. Checkout/Merchant Center not checked. |
| E. Product page content quality | 10% | 80 | Specific titles, real descriptions, and variant names; specs could be more structured. |
| F. Page speed and technical performance | 5% | N/A | Official PageSpeed Insights API returned 429. Not scored. |
| G. Policy machine-readability | 13% | 58 | Policies exist, but official vs custom pages conflict, and schema does not encode the rules. |
| H. Agentic checkout readiness | 15% | 88 | Public UCP profile and Shopify agent signals are strong; admin eligibility is unverified; no merchant-hosted ACP document (not a mandatory fail). |
| Weighted overall | 95% scored, rescaled to 100% | 79 | (12×90 + 18×72 + 15×80 + 12×90 + 10×80 + 13×58 + 15×88) / 95 = 79.3, rounded to 79 |
3. Top 5 fixes (impact and effort)
| Rank | Fix | Impact | Effort | Why it matters |
|---|---|---|---|---|
| 1 | Complete MerchantReturnPolicy and add OfferShippingDetails (and nest them on the Offer where possible). Use applicableCountry: "US", returnPolicyCategory plus merchantReturnDays, and the actual return fee. |
High | Medium | Agents and Google merchant listings read markup before they read a long policy page. Today the markup only has a URL and the country name “United States,” which does not match Google’s required two-letter ISO code, and shipping is not marked up at all. Official: merchant listing structured data. |
| 2 | Make Settings → Policies the source of truth and align the custom Returns and Shipping pages with them. | High | Low | /llms.txt, Organization markup, and Shopify Agentic Storefronts all point at /policies/*. Those pages currently disagree with the custom returns and shipping pages on exchanges and on whether the 60-day window starts at purchase or delivery. An agent will randomly pick one. |
| 3 | Remove or fix the second Product JSON-LD graph that sets mpn to a Shopify numeric ID; add brand on each variant Product; add priceValidUntil while a sitewide sale is running. |
High | Medium | Two graphs on one page can confuse parsers. An mpn must be the manufacturer part number, not a platform ID. Official: Merchant Center identifiers and OpenAI’s feed schema (“do not invent a value to replace a missing GTIN”). The real SKU already exists on the variant node. |
| 4 | Ensure GTINs are in every catalog export used for Google Merchant Center or an OpenAI product feed — do not build a feed from /products.json alone. |
High | Low–Medium | On-page JSON-LD and /products/{handle}.js already have valid GTINs for every sampled variant. The public /products.json endpoint omitted the barcode field for every variant in the snapshot. A feed built from that endpoint would look like a store with no GTINs. |
| 5 | In Shopify admin (screenshare or screenshots): confirm Sales channels → Agentic, Shopify Catalog eligibility, Google Merchant Center connection (needed for Google AI Mode / Gemini), and that the WAF allows shopping crawlers and ChatGPT’s cloud browser. | High | Low | Public signals look eligible (US selling, policies present, products have title/image/price > 0, UCP profile live), but listing and account standing cannot be proved from outside. Official: Agentic storefronts, Selling on Google AI Mode and Gemini. |
4. Detailed findings by area
A. Crawler and agent access — 90 / 100
URLs checked: the storefront robots.txt, llms.txt, agents.md, sitemap.xml, an agentic-discovery sitemap, and the product / page / collection / article sitemaps linked from the index (about 140 product URLs and a comparable number of pages and collections).
Observations
robots.txtreturned 200,text/plain. TheUser-agent: *group allows the catalog and disallows the usual Shopify private paths (/admin,/cart,/checkout,/orders,/accountexcept login). Those disallows are expected and were not treated as defects.- There is a second group only for
adsbot-google. There are no specific rules forOAI-SearchBot,GPTBot,ChatGPT-User,Google-Extended,PerplexityBot,Googlebot, orStorebot-Google. Those crawlers therefore inherit the*rules, which allow product and policy pages. Official meanings of those tokens: OpenAI crawlers, Google common crawlers, Perplexity crawlers. /llms.txtand/agents.mdboth returned 200 as Markdown (several thousand characters). They identify the store, point to the Shop skill, document UCP discovery at/.well-known/ucp, link the four official policy URLs, and point agents at/sitemap.xmland/products/{handle}. This is more useful than a missing file, even though it does not use the optional blockquote summary in the llms.txt v2 proposal.- The agentic-discovery sitemap contains a single URL: the storefront’s
/agents.md. - Sampled product pages had a correct canonical and no
noindexrobots meta. - Whether the store’s CDN/WAF blocks verified ChatGPT-User / OAI-SearchBot IP ranges is unverified. We did not hit a captcha on ordinary GET requests for the pages listed.
Recommendations
- Keep allowing OAI-SearchBot, PerplexityBot, and Googlebot on product and policy pages.
- Treat a GPTBot or Google-Extended disallow as a training-policy choice, not as a shopping-discovery fix, if that is ever added.
- Confirm WAF / bot-fight settings against OpenAI’s cloud browser access and published IP lists, in admin, with the merchant.
B. Structured data (Product / Offer) — 72 / 100
URLs checked: five public product pages, including a core in-stock blanket, a multi-size variant product, an accessory, and two out-of-stock items, plus the homepage, a collection page, and policy pages for Organization-level markup.
In this published sample, product titles are generalized:
- Blanket style A — solid dark colorway (in stock, single size)
- Blanket style A — printed colorway (in stock, three sizes)
- Accessory — muted solid colorway (in stock)
- Blanket style A — gradient colorway (out of stock)
- Packable mat — striped colorway (out of stock)
Official testers: Not run in the practice pass that produced this sample. Before a client-facing audit is delivered, LammyRu pastes the same product URLs into the Rich Results Test and the Schema Markup Validator. The findings below are from extracted JSON-LD compared to Google’s required and recommended properties.
Observations
- Each sampled product page includes a
ProductGroupwithname,image(multiple URLs),brand.name(the store’s brand),url,category,aggregateRating, andhasVariantProductnodes. Each variantProducthasname(including the size),sku,gtin12, an image, and a nestedOfferwithpricegreater than zero,priceCurrency: "USD",availability(https://schema.org/InStockorOutOfStock),itemCondition: https://schema.org/NewCondition, and a varianturl. That satisfies the merchant-listing required Product/Offer properties in Google’s merchant listing documentation. - In-stock example: an in-stock solid-color blanket had a unique SKU, a GTIN-12, an offer price in USD, and
InStock, and the Open Graph product price matched the offer. - Out-of-stock example: a gradient blanket and a striped packable mat were both marked
OutOfStockand still exposed a price, which is what Google asks for when an item is out of stock. - Multi-variant example: a printed blanket exposes three variant Products (a one-person size, a two-person size, and a compact throw), each with its own GTIN-12, SKU, and price.
- Return markup is incomplete. Organization-level
hasMerchantReturnPolicyis essentially a policy URL plus"applicableCountry": "United States". Google requiresapplicableCountryas a two-letter ISO 3166-1 alpha-2 code (US) and, for the return-details enhancement,returnPolicyCategory. A 60-day window would also needmerchantReturnDays. Offers do not nesthasMerchantReturnPolicy. - No
OfferShippingDetailswere found on the sampled product pages, despite a public shipping policy (a free-shipping threshold for contiguous domestic orders, short processing plus several days in transit, international shipping currently unavailable — on the custom shipping page). - A second Product / ProductGroup graph appears on the same pages (likely a theme or app duplicate). It repeats price and availability but sets
mpnto a Shopify numeric ID. Those are not manufacturer part numbers. The real SKU already lives on the first graph. The first graph is the complete one. - Variant
Productnodes in the first graph do not repeatbrand(it is on theProductGrouponly). Ratings on several colorways of the same blanket family share the same high rating and the same review count of several thousand, which may be intentional family-level reviews; this was not independently verified. - Active prices are sale prices (see area D).
priceValidUntilis not set.
Recommendations
- Keep the variant-level
Product+Offergraph; delete or correct the duplicate graph; never output a Shopify ID asmpn. - Add complete return and shipping markup per Google’s property lists; use
USnotUnited States; encode the stated refund-return fee accurately (and free returns for store credit, if that remains the policy). - Add
brandon each variant Product andpriceValidUntilfor the sale. - Re-test the sample URLs in the official testers.
C. Identifiers and variants — 80 / 100
Observations
- Public catalog snapshot via the unofficial
/products.jsonendpoint: about 140 products, 200 variants. Every product had a title, at least one image, a non-empty description, and a vendor (almost entirely the store’s own brand, plus one partner-branded item). Every variant had a price greater than zero and a non-empty SKU. Dozens of products had more than one variant, typically a size option. - Barcodes are missing from
products.json: the variant objects in that endpoint did not even include abarcodekey. - Barcodes are present elsewhere.
/products/{handle}.jsand on-pagegtin12included a GTIN for every variant in a 15-product / 22-variant sample. All 22 GTINs had a valid GS1 check digit (UPC-12), and each size of a multi-variant product had its own GTIN. - Brand in JSON-LD matches the product vendor for those items. That meets Merchant Center’s brand rule for branded goods (brand).
- The duplicate JSON-LD
mpnvalues (Shopify IDs) are a defect; they should not be submitted to Google or OpenAI as an MPN (OpenAI feed schema).
Recommendations
- Use
/products/{handle}.js, the Admin/Catalog APIs, or on-page JSON-LD — notproducts.json— as the GTIN source for any feed. - Confirm in Merchant Center (unverified here) that variant GTINs are populated and that no identifier warnings exist.
- Keep one stable
item_idper variant (the SKU is a good candidate) and a sharedgroup_id/item_group_idfor size variants, with titles that already include the size.
D. Price and availability consistency — 90 / 100
Observations
- For every sampled product, the visible/Open Graph price, JSON-LD
Offer.price, andproduct.jsprice agreed, in USD. - A sitewide percentage sale was advertised in the header. Public compare-at prices matched that discount on every sampled item (including an out-of-stock blanket). JSON-LD correctly exposes the active sale price, which is what Google wants for merchant listings.
priceValidUntilis missing, and an OpenAI feed would also want regularpriceplussale_priceas separate fields — unverified whether any such feed exists. - Out-of-stock items were consistently
available: falseinproduct.jsandOutOfStockin JSON-LD, with the price still present. A small number of variants in the catalog snapshot were unavailable; the rest were available. - We did not open checkout and did not see a Merchant Center feed, so landing-page vs checkout vs feed match is unverified.
Recommendations
- Add
priceValidUntil(and, in any OpenAI or Google feed, a proper sale-price field plus the regular price) for the duration of the promotion. - When the sale ends, confirm schema and feeds move back to the regular price on the same day.
E. Product page content quality — 80 / 100
Observations
- Titles are specific (product family + colorway) and variant Product names append the size. Title length in the catalog snapshot averaged a few dozen characters, well under OpenAI’s 150-character aim.
- Every product in the snapshot had an HTML description. Plain-text length typically ran several hundred characters. The core blanket description explains the use case, materials, and features. That is enough for an agent to describe the item; a structured spec table (dimensions, weight, fill, care) in HTML would be even better.
- Sampled pages had multiple product images on the platform CDN, and they looked crawlable.
- Helpful supporting pages exist (a size guide, an FAQ, and a warranty page). The FAQ page did not expose
FAQPagestructured data in the extract.
Recommendations
- Add
FAQPagemarkup on the FAQ, and put key dimensions/weight in text (not only in images) on product pages. - Keep legal or warranty disclosures in the first part of the description, per Shopify’s ChatGPT guidance (first 6,000 characters).
F. Page speed and technical performance — N/A
Tool: The official PageSpeed Insights API returned HTTP 429 Too Many Requests on 22 November 2026 for the homepage (mobile and desktop) and one in-stock product page (mobile).
This area is not scored. The 5% weight was removed and the overall score was rescaled. Product HTML and JSON-LD were present in the initial GET response, which is the more important agent-readiness signal.
Recommendations
- Run https://pagespeed.web.dev/ on the homepage and one product page, mobile and desktop, and attach the scores to a client-facing version of this report.
G. Policy machine-readability — 58 / 100
URLs checked: official /policies/refund-policy, /policies/shipping-policy, /policies/privacy-policy, /policies/terms-of-service, and /policies/contact-information, plus custom returns, shipping, and FAQ pages, and the policy links inside /llms.txt.
Observations
- Official Shopify policy URLs that Agentic Storefronts require (Terms, Privacy, Return/refund) do contain real text, not an empty theme shell. Privacy and Terms are substantial. The official refund policy has a real body. The official shipping policy is thin (only a few hundred characters). The official Contact information page is effectively empty (a title only).
/llms.txtcorrectly lists the four official policy URLs. Organization JSON-LD points the return-policy URL at/policies/refund-policyand the shipping-policy URL at/policies/shipping-policy.- The official refund policy and the custom returns page do not fully agree:
- Official refund policy: returns within 60 days of purchase, also refers to a night-trial guarantee, lists reasons the store “will accept exchanges,” notes a fixed fee deducted from a refund, and free returns for exchanges and store credit. It also discusses international returns.
- Custom returns page: returns within 60 days of delivery, “we do not offer direct exchanges” (return and place a new order), the same refund-return fee, store-credit returns free, international orders not eligible, and a multi-year warranty with a different term for older orders.
- Official shipping policy vs custom shipping page: the official page only covers order-processing time and a request to email about changes. The custom shipping page is the useful one: a free-shipping threshold for contiguous domestic orders (with some state/territory exclusions), typical transit time, an expedited option, and a statement that shipping outside the country is currently unavailable.
- Structured return/shipping data does not encode any of those facts (see area B). An agent that only reads markup learns almost nothing; an agent that reads both policy URLs may give the shopper opposite answers about exchanges and about when the 60-day clock starts.
Recommendations
- Decide the legal rules once (purchase vs delivery; exchanges or not; international). Put the complete, current text on the official
/policies/refund-policyand/policies/shipping-policypages, then make the custom pages a short summary that links there. - Fill or remove the empty Contact information policy.
- Encode the agreed rules in
MerchantReturnPolicyandOfferShippingDetails(countryUS, return window, fee vs free store-credit returns, free-shipping threshold, transit times, no international if that remains true).
H. Agentic checkout readiness — 88 / 100
URLs checked: /.well-known/ucp on the custom domain and the myshopify.com host, versioned UCP profiles linked from it, /.well-known/acp.json (404), and GET-only requests to the advertised MCP path (404, expected for a JSON-RPC POST endpoint). We did not call tools or create a checkout. Platform signals included the Shopify response header, Shopify-style sitemaps, and the public products API.
Observations
- Platform: Confirmed Shopify from public headers, sitemaps, product APIs, and a
myshopify.comhost in the UCP profile. - UCP:
GET /.well-known/ucpreturns public JSON, no authentication, advertising protocol version2026-08-25and also supporting earlier dated versions documented in Google’s UCP guides. Services include shopping over MCP (on themyshopify.comhost) and an embedded transport. Advertised capabilities include checkout, cart, fulfillment, discount, order, catalog search/lookup, and a permalink checkout path. Payment handlers include Google Pay, a card handler, and Shop Pay. This matches Shopify’s documented behavior that storefronts advertise UCP, including 2026-08-25, at/.well-known/ucp(Shopify changelog, 4 September 2026; agent profiles; Google’s public-profile rule at Prepare your UCP profile). Whether Google AI Mode already negotiates the2026-08-25version is uncertain (Google’s merchant UCP guides still highlight2026-04-08and2026-01-23). /llms.txttells agents to discover UCP on the custom domain but then to connect to an MCP URL on that same custom domain. The profile itself advertises the MCP endpoint on themyshopify.comhost. Agents should trust the profile’sendpoint. This is a small documentation inconsistency, not proof that UCP is down.- Shopify Agentic / Catalog, public signals only: the store is not passworded; products in the snapshot have a title, image, and price greater than zero; Terms, Privacy, and Refund official policy pages have content; shipping pages indicate selling to the United States, with international shipping currently unavailable. ChatGPT via Shopify requires selling to US customers (Selling on ChatGPT); Google AI Mode / Gemini requires US, Canada, or Australia plus a Merchant Center account (Selling on Google AI Mode and Gemini). US selling appears to be met. Plan, account standing, Catalog eligibility, Agentic opt-out, Supplemental Terms acceptance, and Merchant Center connection are unverified.
- ACP:
GET /.well-known/acp.jsonreturned 404. That path is described in an ACP RFC marked proposal / unreleased at research time, so this is not a failed mandatory check. No merchant-hosted ACP checkout API was discovered. For a Shopify store, ChatGPT shopping may already be handled by Shopify Catalog plus in-app or tab checkout without a custom ACP implementation.
Recommendations
- Fix the MCP URL in
/llms.txtso it matches the UCP profile (or advertise both). - Confirm Agentic, Catalog, and Merchant Center in admin (fix 5).
- Do not commission a custom ACP implementation solely because
acp.jsonis missing.
5. 30-day action plan
Week 1 — correctness
- Align official
/policies/refund-policyand/policies/shipping-policywith the custom pages (or vice versa). Decide purchase vs delivery, and exchanges vs repurchase. - Expand the official shipping policy with countries, the free-shipping threshold, and transit times.
- Draft complete
MerchantReturnPolicyandOfferShippingDetailsmarkup.
Week 2 — identifiers and feeds
- Remove the duplicate JSON-LD graph and the Shopify-ID
mpn. - Point any Google / OpenAI feed at a source that includes variant GTINs and sale-price fields.
- If Merchant Center exists, review identifier and availability diagnostics (needs the merchant).
Week 3 — agent entry points
- Align
/llms.txtMCP endpoint with the UCP profile. - Confirm robots.txt and WAF still allow shopping crawlers and ChatGPT’s cloud browser.
- Add
FAQPagemarkup if the FAQ is staying.
Week 4 — channels and measurement
- Screenshare: Sales channels → Agentic, Catalog eligibility, Google Merchant Center connection, optional ChatGPT Business profile claim.
- Re-run the Rich Results Test and PageSpeed Insights on the same sample product URLs and the homepage; record before/after.
6. What we could not verify
- PageSpeed Insights / Lighthouse scores (API HTTP 429).
- Official Rich Results Test and Schema Markup Validator results (not run in the practice pass).
- Google Merchant Center feed health, identifier warnings, and shipping/return settings.
- Shopify admin: plan, account standing, Catalog eligibility, Agentic channel opt-in/out, Supplemental Terms, and Merchant Center connection.
- WAF allowlisting for OAI-SearchBot, ChatGPT-User, or ChatGPT cloud-browser signatures.
- Whether the store’s products actually appear in ChatGPT, Google AI Mode, Gemini, Perplexity, or Copilot today.
- Whether Google already negotiates UCP version
2026-08-25with this storefront. - Checkout-page price, tax, and shipping quotes (we did not start a checkout).
- Inventory quantities beyond the public
availableboolean. - Whether family-level
aggregateRatingvalues reused across colorways of the same product family are review-guideline compliant — they may be legitimate shared-product reviews; this was not audited against review snippet rules in depth.
7. Scope and method
- Playbook version: 1.0
- Sample: 5 product pages (plus supporting
product.json 15 products / 22 variants), a public catalog snapshot of about 140 products, official and custom policy pages, homepage and a collection page, and the discovery files listed above. - Access: public pages only, read-only.
- We did not: log in, submit forms, start a checkout, contact the business, place an order, or call UCP/ACP checkout-create or payment APIs.
8. Next services (optional)
These are optional follow-ons. They are not required to accept an audit. Cleanup is quoted after the audit. A custom assistant is a custom quote.
Catalog and Feed Cleanup
A good fit here: GTINs already exist on the storefront but would be dropped by a /products.json-based feed; sale prices need a correct regular-price + sale-price mapping; titles and variant grouping are already in decent shape and would not need a full rewrite.
Custom AI Agents
Only after policy text and identifier consistency are fixed. Shopify already exposes a UCP MCP endpoint on this store; a custom agent may be unnecessary for basic “shop this catalog” behavior, and would need a clearer use case (support, warranty claims, B2B, and so on).
Not sold from this sample audit: a custom UCP/ACP implementation. The platform already publishes UCP. A custom protocol build is not the fix for the gaps we found.
9. Walkthrough
In a real engagement, a 30-min call is included. We walk through the scorecard, the top 5 fixes, and the 30-day plan, and leave time for questions.
Questions to confirm on the call
- Which policy is legally authoritative if official and custom pages disagree?
- Do you already use Merchant Center, ChatGPT Business Hub, or Sales channels → Agentic? Can we review screenshots?
- Which of the top 5 fixes do you want to tackle first, in-house or with LammyRu?