How to approach it
Inspect a product with and without reviews. Compare visible counts and aggregate scores with rendered markup. Check for competing theme and app output and validate representative URLs.
Inspect a real, rendered product page after reviews have loaded. A theme and a third-party reviews app can each emit a product entity. Identify which output corresponds to the reviews shoppers can see. Compare the exact product identifier and the period spanned by the visible reviews before correcting any aggregate.
Worked example
If a page shows eight reviews while markup says 80, trace app synchronisation or duplicated output. Revalidate after the visible and machine-readable counts agree.
How to verify the change
Sample a product with reviews, one without reviews and a variant with a different review policy. Test results again after an app or theme update. Keep an example URL of each type in your maintenance checklist. An eligible rich result can still be absent from a particular search, so treat validation as a QA measure.
If the same reviews are intentionally shown across variants, describe the grouping on the page so shoppers understand what the aggregate represents. Where products differ materially, test whether the review display still describes the item being considered.