What You Get Back From a Free 10-SKU Catalog Enrichment Sample

A free 10-SKU catalog enrichment sample is a proof unit, not a teaser brochure. You send ten live product URLs (or ASINs). You get back structured attribute rows — typed fields, extracted specs, and the gaps that stayed blank because we could not verify them — that you can compare against your Shopify admin or Amazon listing template before you commit to a full catalog pass. The point is not “trust our process.” The point is to see whether the enriched data is accurate enough that you would publish it.

This is a GigaCommerce method how-to for operators in the $100K–$10M band who keep hearing that AI shopping needs structured catalogs and need a concrete artifact to evaluate. It sits under Catalog Enrichment for AI — the foundation work that Brand Agents, Rufus, and off-site assistants all read. It is not a replacement for the full catalog enrichment playbook; it is the smallest unit of that playbook you can run without boiling the ocean.

Why a 10-SKU sample beats another audit deck

Most mid-market catalogs already have some product data. Titles exist. Descriptions exist. A handful of options or bullets exist. What is missing is the layer machines reason over: named attributes with discrete values, consistent units, and category schemas that match how shoppers actually ask questions. That gap is why a store can rank in classic search and still fail when a Brand Agent or ChatGPT tries to answer “will this fit a Queen mattress?” or “is this dishwasher-safe?”

A full-catalog proposal without a sample forces you to buy on faith. A sample flips the order: you see the work product first. You can open your Shopify product editor, look at metafield definitions, and ask whether the returned material_composition, fit, or care_instructions values match the physical product. On Amazon, you can open the category listing template (flat file) for that product type and check whether the sample fills recommended columns — not only the red “required” ones — that filters and Rufus lean on.

That coverage band is the commercial reason the sample exists. If roughly half of the fields that matter are empty or stuck in prose, ten SKUs are enough to prove the pattern without waiting for a 2,000-SKU project plan.

What we ask for — and what we refuse to invent

The intake is deliberately boring: name, email, store URL, and ten product URLs (or a paste of ASINs / handles). We want products that already exist on a live channel so the “before” state is public and checkable.

Good SKUs for a sample

• Heroes by revenue, not orphans — if the sample only proves you can enrich a clearance SKU, you learned nothing about the money.
• At least two SKUs with known pain: apparel with fit ambiguity, home goods with dimensions buried in copy, accessories with “works with” questions, gift bundles with opaque contents.
• One SKU that looks “fine” in the admin — so you can see whether “fine for humans” still fails machine readability.

What we will not fill

We leave a field blank when we cannot point to a source: the physical product, a manufacturer spec sheet, packaging, or a measurement. Invented attributes are worse than empty ones for AI shopping — an agent that confidently states the wrong material or compatibility creates returns and trust damage. That rule is the same discipline used in serious Amazon attribute work: blank is recoverable; wrong is expensive.

What you get back in five days

Turnaround on the free sample is five business days when the ten URLs are valid and the products are publicly viewable. Delivery is usually a structured table (Notion or spreadsheet) plus short notes per SKU — not a 40-page strategy PDF.

1. Before / after attribute inventory

For each SKU you see:

• Fields that already existed as structured data
• Fields extracted from description, bullets, or image captions (prose-trapped)
• Fields still empty because no trustworthy source existed
• A recommended category schema row (aligned to templates like those in the attribute templates library)

A typical mid-market “before” on a home-goods PDP might show four populated options (color, size) and a marketing paragraph that mentions dimensions once. A usable “after” for agent readiness often lands in the high teens to low twenties of relevant fields for that category — materials, dimensions with units, care, use occasion, and compatibility — not fifty vanity columns. Treat those counts as a pattern check, not a leaderboard.

2. Publish-ready field values

Values are typed the way Shopify metafields and Amazon flat files expect them: booleans as true/false, dimensions with units, materials as structured strings, not poetic fragments. Where your store already has metafield definitions, we map into those keys when you share the definition list; otherwise we propose a clean key set you can import.

On Amazon, the checkable reference is the category listing template itself. Seller Central’s flat file for a product type commonly exposes dozens of attribute columns; only a minority are marked required, but browse filters and assistant answers lean heavily on the recommended set. The sample should make clear which template columns your ten ASINs would newly populate.

3. Schema and feed notes (not a full site rebuild)

The sample is catalog work, not a theme rewrite. You still get notes on Product / Offer gaps that would block clean structured data — missing GTIN where you have one, price without a clean currency, availability as free text — so you know whether publishing attributes alone is enough or whether PDP schema needs a follow-on pass. Deep schema scoring is a sibling job (see structured-data posts in Commerce GEO); the enrichment sample flags it, it does not replace it.

4. A gap log you can act on this week

Beyond the ten rows, you get a short pattern list: “dimensions only in prose on 7/10,” “no dishwasher_safe boolean,” “bundle contents not itemized,” “compatibility absent on accessories.” That log is what you take to your team — or to a scoped Catalog Enrichment engagement — without another discovery workshop.

1. 1

   Send ten live product URLs

   Prefer revenue heroes plus two messy SKUs. Include platform (Shopify / Amazon) and category.

2. 2

   Review the returned table against the physical product

   Every non-blank field should be verifiable. Flag anything you would not publish.

3. 3

   Decide: publish, request corrections, or stop

   If the sample is wrong, stop. If it is right, publish those ten and scope the next batch by category.

How to judge the sample in thirty minutes

Do not “feel impressed.” Run a publish test.

1. 1Accuracy check. Pick three fields per SKU at random. Confirm against the product in hand or the manufacturer sheet. One invented field fails the vendor.
2. 2Machine-readability check. Ask whether a Brand Agent or Rufus could answer a real shopper question using only the new fields — no description paragraph. If the answer still requires prose, the enrichment is incomplete for that question class.
3. 3Channel check. On Shopify, paste values into metafields (or a staging product). On Amazon, map into the flat-file columns for that product type. If mapping takes heroics, the schema keys are wrong.
4. 4Honesty check. Count blanks that are labeled “needs source.” A good sample has some. A sample with zero blanks on thin source material is suspicious.

If you are under that band, the sample should feel immediately usable in Shopify admin. If you are far above it, the sample still proves quality — scale mechanics are covered in enriching a 10,000-SKU catalog.

What the sample is *not*

• Not a Brand Agent install. Enrichment feeds agents; it does not configure conversation flows or Copilot Checkout. For the agent layer, see Agentic Commerce Setup after the catalog floor is honest.
• Not a Commerce GEO citation panel. Filling attributes improves extractability; it does not guarantee ChatGPT will name you tomorrow. Citation measurement is a separate loop.
• Not a license to invent peak-season facts. If you are racing into Q4, the sample should prioritize corrections and gift/compatibility fields — the same sequencing as peak-season catalog mistakes — not a race to fill every optional column.

How the sample scales into real catalog work

If you approve the ten rows, the next step is not “enrich everything.” It is:

1. 1Lock a per-category attribute schema (start from shopper questions — see product attribute schema design).
2. 2Enrich the next revenue cohort with the same keys and verification rule.
3. 3Add compatibility relationships where shoppers shop by fit — covered in compatibility data for AI agents.
4. 4Gate new SKUs so intake cannot skip required fields.

That is the same sequence as the playbook, with the sample as week-zero proof. Throughput pricing on the SKU page exists so large catalogs are scoped by volume after quality is proven, not before.

Who should send a sample this week

• Shopify merchants preparing Brand Agents or on-site AI search who suspect metafield coverage is thin
• Amazon sellers who fill required flat-file cells but leave recommended attributes empty
• Teams stuck arguing PIM vs metafields before they have proven they can maintain a schema at all
• Anyone who has been pitched “AI catalog optimization” without seeing a single enriched row for their own products

If you already run a disciplined PIM with high coverage, a sample may only confirm you are fine — that confirmation is still useful before you spend on agent tooling.