A Shopping campaign can keep spending after its most valuable product disappears. The graph still moves, clicks still arrive, and the budget simply migrates to whatever remains eligible. By the time sales reveal the problem, the dashboard has already spent days pretending the campaign was healthy.
That is why Merchant Center monitoring belongs at the product level. Campaign traffic is a delayed and sometimes misleading proxy for feed health. The operational question is not whether Shopping stopped. It is whether the products the business actually needs to sell are still allowed into the auction.
A living campaign can hide a dead product
A merchant reported finding a GTIN disapproval only after Shopping impressions had been almost absent for three days. That complete failure was visible eventually. A partial failure is harder. If Performance Max redirects budget toward eligible inventory, the campaign total can stay active while a bestseller, high-margin line, or launch product receives nothing.
A 25% fall in eligible products may deserve an alert. So may a single excluded SKU if that product carries a large share of revenue. Percentage thresholds alone treat every item as equally valuable, which is tidy for monitoring and wrong for the business.
The useful alert combines status and commercial importance. Watch the count and share of products that are eligible, limited, pending, and disapproved. Then maintain a smaller priority set for products whose exclusion should wake someone immediately.
Campaign alerts still have a job
A simpler counterargument deserves to survive: some feeds fail almost entirely. In that environment, a zero-spend or zero-impression alert can be enough. It is easy to operate, it catches the common failure, and another layer of monitoring may add noise without improving the response.
The mistake is turning that success into a universal rule. Campaign-level alerts answer whether delivery is still occurring. They cannot tell you whether delivery moved to weaker inventory. Use them as the outer alarm, then use product eligibility to identify what disappeared.
The same hierarchy works for anomaly scripts. A sudden impression fall is a reason to inspect Merchant Center, not proof that the feed is responsible. Product status gives the investigation a sharper starting point.
Do not rebuild the store for a crawler incident
A product-page-unavailable error creates a different temptation: change the storefront immediately. During one reported incident, multiple Shopify accounts saw Merchant Center return HTTP 500 errors for product pages. Checks later cleared affected products without a documented site change. Other accounts appeared unaffected.
That sequence does not prove Google or Shopify was responsible. It does show why a Merchant Center error is not yet a storefront diagnosis. First test the affected product URL as the relevant product crawler, inspect server or CDN logs for its requests, compare affected and unaffected products, and check whether independent accounts are reporting the same pattern.
If the storefront serves the page correctly and the error is spreading across unrelated accounts, request a recheck and observe before changing infrastructure. If the product crawler repeatedly receives an error that the logs confirm, the site has supplied a concrete repair target.
Monitor the failure closest to the money
A practical Shopping control should answer four questions:
- Did total eligible inventory change materially?
- Did any priority product become limited or disapproved?
- Did spend move toward products with weaker commercial value?
- Does the product crawler receive the same page response as a normal visit?
The evidence here cannot establish how common silent partial disapprovals are or which system produced the temporary HTTP errors. It supports a narrower operating rule: monitor the failure where it begins, and verify it before changing the store.
Campaign totals tell you that money is moving. Product eligibility tells you whether it is still moving through the business you intended to advertise.