How we decide a rate is good

What is a good cashback rate?

A cashback rate is good when it beats what that store usually pays on that same portal. Every shopping portal (a site that pays you cashback, miles or points for starting your purchase there) will tell you today's rate. Few tell you whether it's worth acting on. For example, when Nike shows 8% on TopCashback, the question that matters is whether 8% is good for Nike on TopCashback.

So each chip compares a rate only with its own history on that portal, never against another store. The optional Rank by value view, described below, is the one place rates from different programs sit side by side, and it uses point values you can edit. For a worked case, see Is 10% cashback actually good?, which runs this comparison on Nike.

The chip next to the rate

Each rate carries a short chip saying how today's number compares to that store's own past on that portal. There are five, and they are the whole scale:

• Highest yet - the top of the range we have recorded for this store on this portal. One label, two scopes: either the highest we have ever tracked for it, or the highest in the past year (or since we started tracking it, if that is shorter). Hover the chip and the tooltip says which one you're looking at. We deliberately don't split the label, because with histories this young the distinction tells a reader nothing they can act on.
• Near high - close to the top of that range, not quite at it.
• Above usual - higher than this store normally pays here.
• Typical - about what it usually pays. Not a bad day, just an ordinary one.
• Below usual - lower than usual for this store on this portal.

A small flame rides on Highest yet only when it means the highest we have ever tracked, and only once we have tracked that store on that portal for 30 days or more. A rate at a one-year high sitting below an older peak never flames, and neither does a young history, however high the rate is today.

Each chip has a four-segment meter beside it, so the scale reads at a glance without relying on color. The meter is coarser than the labels on purpose: it groups the chips into four steps rather than five, so "Near high" and "Above usual" share a step. Hovering tells you the rest: which comparison the chip made, and the store's usual range in that unit ("Usually 4–6%").

We used to print a percentile instead ("96th"). It was accurate and almost nobody could read it, so it's gone.

Three details keep the chips accurate:

• They're time-weighted. A rate that sat at 4% for 80 days counts for 80 times as much as a 10% flash that lasted a day. A one-day spike can't distort the year.
• A rate has to have actually risen to claim a high. A rate that has never moved is trivially sitting at its own maximum, and calling that a high would put "Highest yet" on most of the catalogue. So a rate only reaches the top two chips if it has genuinely spent time below where it is now.
• They need real history. For the first week after we start tracking a store on a portal, we show no chip at all. A "high" needs something to be high against, and we would rather show nothing than manufacture a claim.

What the verdict box does

The verdict is the one-sentence answer at the top of a store's page. It turns those stats into plain English, leading with the strongest fact it can state: the highest we've ever tracked beats the highest in the past year, which beats "higher than usual," which beats "the usual rate." It names the portal and the program every time, and it appends the caveats that matter, such as an "up to" headline that only applies to some categories, or a stale reading we haven't confirmed recently.

One rule inside it is worth spelling out, because it looks like a bug and isn't. When a headline "up to" rate is higher than any category the portal actually publishes, we rank that rate by the best category it really reaches, and the verdict then drops the history claim entirely and just says what the store pays right now. Saying "3% is the highest yet" would be a claim about a number the history never held.

The rules we apply

A few rules we hold ourselves to, because a claim you can screenshot is a claim we have to defend:

• Each rate keeps its unit. We show miles as miles and points as points, never silently converted into dollars. Each rate appears in the unit its portal states: airline portals usually pay miles, some now pay points, card portals pay points, and cashback portals pay cash. 3 miles per dollar (3 mi/$) has no fixed cash value until you choose what a mile is worth to you. The optional Rank by value view does exactly that, using your own point values (our defaults stand in until you set yours), with the assumption shown and editable.
• Stale stays visible, badged. If a portal hasn't been confirmed within its stale window, the rate gets a stale badge with the last-confirmed time, and it stays ranked. For most portals that window is about 2.5 times the normal check interval. For our most-checked stores it is at most twice the interval, and a few portals have longer windows we set by hand. Hiding the data would make the page tidier, and the reader would know less.
• Provenance on hover. Every number shows the portal's exact wording, character for character, on hover, and the rate's details show when we last checked it.
• "Up to" means up to. Tiered headline rates keep their up to badge, and the fine print doesn't disappear into a bigger font.

Where the data comes from

We read each portal's own store pages on a schedule, usually every 6 to 24 hours. The portal data status page lists every portal we track and when it was last confirmed, including any we check by hand, which are labeled as such. If something looks off on a store you know well, use the Report this rate button on that store's page: reports go straight to us for review.

The short version: we do the comparison a careful person would do with a spreadsheet and plenty of patience (and we show our work). To see the chips in practice, browse today's Highest yet list, or set an alert for a store you buy from on your alerts.