Two measurements, not one visibility score
SkuWatch AI Visibility separates storefront readiness from sampled AI visibility. Readiness describes public, repeatable signals a merchant can inspect: reachable pages, crawler directives, canonical destinations, structured product facts, and visible supporting evidence. Visibility describes what a defined AI surface returned for a specific buyer question at a recorded time.
A store can have strong readiness and still be absent from one answer. It can also appear in an answer while its product data contains important gaps. The two measurements inform each other, but they must not be merged into an unexplained ranking score.
What the storefront scan checks
A public scan starts with URLs a shopper can open without an account. It records response status, redirects, canonical destinations, crawler directives, agent-facing guidance, and machine-readable product facts. Product checks compare visible facts with JSON-LD rather than treating the presence of a script as proof that the data is correct.
Each finding should state:
- Observed: the exact public value or missing signal.
- Expected: the condition used by the check.
- Where to fix: the Shopify field, theme output, or public page involved.
- Evidence: URL, timestamp, field path, and relevant value.
A pass means the defined signal was observed at scan time. A warning requires human review. A failure means the expected public signal was absent, inaccessible, or contradictory. It does not mean an AI platform has penalized the store.
How AI visibility sampling works
Visibility monitoring begins with a controlled set of buyer questions. A useful prompt identifies a product category, use case, and meaningful constraint. SkuWatch AI Visibility records the prompt, timestamp, market and language context, provider or endpoint identifier, visible answer, and available sources.
Results are aggregated as sampled observations across completed runs. A provider error, timeout, or unavailable answer is excluded from the denominator rather than silently counted as a failure to mention the store.
Sampling is directional, not universal. Prompt wording, location, model updates, retrieval indexes, personalization, inventory, and source availability can change an answer. SkuWatch AI Visibility therefore does not claim to show what every user sees in a private conversation.
Evidence levels for a store appearance
- Domain verified: a known storefront domain or canonical URL appears in observable source evidence.
- Brand mention: the answer names the brand, but no known storefront URL is available.
- Inferred mention: the interface suggests an association that cannot be directly verified.
- Not observed: the completed result contains no matched domain, URL, or brand string.
- Not tested: no valid result was collected.
These states must remain distinct. An inferred mention cannot be presented as a verified citation.
How changes should be evaluated
Record a baseline before changing the storefront. Make the smallest useful repair, verify the public output, and rerun the same prompt set after the change. Keep the timing and platform context visible. A change followed by a higher appearance rate is an observation worth investigating, not proof of a causal ranking factor.
What public research will include
Every SkuWatch AI Visibility research report will state its store and prompt coverage, collection period, provider context, evidence definition, exclusion rules, and material limitations. Reports and comparisons will not be published until the underlying data can be inspected and reproduced.