Attribution needs a change graph
SkuWatch AI Visibility connects each observation to:
- product, variant, and market
- changed field or Theme release
- source evidence and approver
- deployment and verification time
- feed or channel event
- stable prompt IDs
- completed provider runs
- source citations and factual accuracy
- referral or commerce events when legitimately available
Without this graph, a later ChatGPT citation may be celebrated even though the relevant change was unrelated.
Use defined windows
A comparison can report:
Before:
4 canonical citations in 18 completed prompt runs.
After:
7 canonical citations in 19 completed prompt runs.
Material changes:
compatibility page published; feed title corrected; two products
returned to stock; provider model changed during the period.
The evidence supports an association during a defined test set. It does not prove that one edit caused every later recommendation.
Attribution levels
| Level | Defensible statement |
|---|---|
| Temporal | the observation occurred after the release |
| Matched evidence | the answer used the corrected fact or source |
| Controlled comparison | stable prompts and comparable periods changed |
| Commerce link | a legitimate referral or order carried channel attribution |
SkuWatch AI Visibility reports the level earned. It does not replace missing evidence with a confident score.
The completed attribution record returns to Monitor and establishes the next baseline.