A brand's own reference price overstated the discount about five times as often as it understated it.
Of 501 products where both a reference price and Dipp's own recorded history point to a real discount, the reference price implied a discount at least 10 percentage points steeper than the recorded history would in 129 cases — against 24 the other way.
Dipp already prefers its own record — which is why most products get neither number.
3,633 of 4,688 tracked products get no discount claim at all today: either there is no credible reference price, not enough recorded history, or — for a recognised brand — the current price sits at or above what Dipp has typically seen it sell for, which the site treats as disqualifying regardless of what the reference price claims. Of the rest, 198 are shown against a reference price and 857 against Dipp's own recorded typical price — already the majority, by more than four to one.
When a like-for-like comparison is possible, the two bases mostly agree — but not always in the shopper's favour.
Restricting to the 501 products where both a reference price and a recorded typical price independently imply a real discount — the only cohort where "which basis is more generous" is a fair question — the two agreed within 5 percentage points 291 times. Where they diverged by 10 points or more, it went one direction far more often than the other:
A reference price overstating the discount is the pattern a crossed-out price the product never actually sold at produces. This is not proof of that in any individual case — a genuine, temporary reference price is indistinguishable from a stale one by this method alone — but the asymmetry itself (26% versus 5%) is the reason Dippstates which basis it used on every product page rather than printing one number and letting a shopper guess.
How this study was calculated
Computed 2026-09-13 by running the same functions the site itself uses to decide what a shopper sees — getComparisonBasis and getObservedMedianPrice from the live codebase — against every product in the catalogue at that moment, not a separate analysis written for this report. If the site's comparison logic changes, this snapshot does not update itself and would need recomputing to match.
This is a cross-sectional audit of one moment, not a month of daily observations like the August price index. "Recorded history" requires at least three logged checks, matching the threshold the site itself uses before it will name a typical price; a trusted brand priced at or above its own recorded median is excluded from the recorded-price comparison entirely, matching the site's own rule that no discount is claimed in that case. The 501-product "both comparable" cohort is a subset of the 963 with both signals present — the remainder had a reference price or a typical price technically computable, but the current price did not sit below at least one of them, so no discount comparison was possible from that side.
Coverage is not a census of Amazon.ca. It reflects products Dipp tracks, and department mix is uneven — Dipp's own department taxonomy was substantially corrected earlier this month, so a department breakdown of this cohort would describe the correction more than the underlying pattern and is deliberately not included here. This is independent research. Amazon did not commission, review, or endorse it.