Can ChatGPT Retrieve the Right Shopify Product Facts? A Woobles Test Plan

This test uses the public Pierre the Penguin crochet-kit page from The Woobles because it contains both strong buyer evidence and a high-impact commercial conflict.

The store is not presented as a SkuWatch AI Visibility customer. Public facts were last verified July 27, 2026.

The product question

Which beginner crochet kits include video instruction and support left-handed learners?
For every recommendation, cite the official product page and report current availability.

A correct answer needs:

• product and merchant identity
• beginner suitability
• video instruction
• left-handed support
• current stock state
• official product URL

Public storefront baseline

The visible product page communicated:

• beginner positioning
• $30 observed product price in the reviewed context
• sold-out status
• video instruction
• right- and left-handed support
• kit contents
• ages 12 and over
• approximately 4.5-inch finished size
• visible review evidence

The page is not empty or generic. It can answer most of the buying question.

Machine-readable conflicts

Our earlier HTML audit found three JSON-LD blocks and two Product entities. Both Product offers reported InStock while the visible page said sold out. One structured aggregate reported zero while visible review evidence was substantial. The observed brand value Beginner also looked more like skill classification than merchant brand.

This creates three possible answer failures:

1. recommendation reports the product as in stock
2. answer omits useful review evidence
3. product is classified by a malformed brand signal

Stage 1: test public reachability

url='https://thewoobles.com/products/penguin-crochet-kit'

curl -sSIL "$url"
curl -sSIL -A 'OAI-SearchBot/1.0; +https://openai.com/searchbot' "$url"
curl -sSIL -A 'ChatGPT-User/1.0' "$url"

On July 27, our browser-like, OAI-SearchBot-string, PerplexityBot-string, and Googlebot-string requests all returned 200 after resolving to the Canadian-market product path.

This is not proof of legitimate OpenAI crawler traffic because the user-agent string is spoofable. It is evidence that the sampled request path did not block that string.

Stage 2: preserve expected facts

query_id: woobles-beginner-kit-01
canonical_product: Pierre the Penguin Crochet Kit
expected:
  skill_level: beginner
  instruction: video included
  handedness: right and left
  visible_availability: sold out
critical_errors:
  - report product as currently in stock
  - cite only a third-party retailer as the official store

Review the expected sheet immediately before the ChatGPT run because availability can change.

Stage 3: run the ChatGPT observation

Use a clean session. Record what the interface actually displays:

Field	Required record
prompt	exact text
run time	date, time, timezone
locale	country and language
mode/model	only when displayed
answer state	success, refusal, timeout, error
brand mention	exact text
product mention	exact text
visible sources	complete URLs
availability claim	exact wording

Do not rerun repeatedly until a preferred answer appears. Use the scheduled repetition policy defined before testing.

Stage 4: classify the result

Official-domain citation: a visible source resolves to the approved The Woobles storefront domain.

Brand mention only: The Woobles is named but no official source is visible.

Third-party evidence: the product is supported only by another retailer or editorial page.

Completed non-mention: a usable answer recommends other products and omits the approved brand and domain.

Provider failed: no usable answer exists. Exclude it from the mention denominator.

Stage 5: diagnose fact accuracy

If the answer says InStock, save the cited source and timestamp. The storefront conflict makes the root cause ambiguous:

• stale structured Offer may have been used
• a cached page may have been used
• a third-party source may report different stock
• availability may have changed between observations

The report should say which evidence was visible, not assume the provider’s internal path.

Shopify repair

In a duplicate theme:

1. inventory every Product entity
2. identify theme, SEO app, and review app ownership
3. connect Offer availability to selected-variant state
4. align aggregate rating with visible review scope
5. replace classification text in the brand field with the actual brand
6. verify sold-out and in-stock products

Before-and-after test

Repeat the same prompt only after:

• the public fix is verified
• a reasonable recrawl interval has passed
• expected availability is refreshed
• prompt, locale, and scoring rules remain unchanged

A later correct answer is encouraging correlation. It is not proof that one storefront edit controlled ChatGPT.

For the platform-wide procedure, use the complete ChatGPT recommendation guide.