How we rate coffee
Cuppd scores coffee with a 1–5 community rating, averaged without weighting, on top of a catalogue collected from roasters' own product pages. This page documents exactly how that works, including the parts that are AI-assisted.
How is a coffee's score calculated?
A coffee's score is the plain arithmetic mean of every rating it has received, on a 1 to 5 scale, rounded to two decimals. There is no weighting, no decay, no adjustment for who left the rating and no editorial override. Ten ratings of 4 and one of 5 produce 4.09 — nothing more complicated is happening.
A rating is one whole number from 1 to 5 per person per coffee. Anyone with an account can leave one, and changing your rating replaces your previous one rather than adding to it, so no single account can move an average more than once.
How many ratings does a coffee need before a score is shown?
The average appears as soon as one rating exists, but a coffee needs at least three ratings before Cuppd publishes it as a machine-readable star rating for search engines. One person's opinion is a data point, not a verdict, and presenting it to Google or an AI assistant as an aggregate rating would overstate what we actually know.
Most of the catalogue is unrated. Those pages exist so the coffee is discoverable and its details are documented — not because we are claiming it is good.
Where does the coffee data come from?
Every coffee's origin, farm, altitude, variety, process, roast level and tasting notes come from the roaster's own product page, collected by an automated crawler. The roaster is the primary source; Cuppd does not invent attributes, and where a roaster does not publish a field, the field stays empty rather than being guessed.
The crawler checks robots.txt before every visit and only reads product pages. A roaster that asks not to be crawled is not in the database. Catalogues are re-checked monthly for new releases.
Which parts of a coffee page are AI-generated?
The prose summary at the top of a coffee page is AI-assisted: a model writes it from the roaster's own product copy and the structured cupping data, and it carries a visible “AI-assisted description” label wherever it appears. Nothing else on the page is generated.
Ratings, review text, tasting notes, and every attribute in the details table are either entered by a person or taken verbatim from the roaster. A generated summary never influences a score.
How are the toplists ordered?
Rated coffees come first, highest average score down, and unrated coffees follow. Every coffee in the catalogue is eligible — there is no filter by roaster country or by whether the coffee has a product image.
Personal recommendations in the For You feed are ranked from your own past ratings: we look at the attributes shared by the coffees you scored highly and rank similar coffees up. They are not sponsored, and no roaster can pay for placement in them.
Does Cuppd make money from these ratings?
Some outbound links to roaster webshops are affiliate links, which may earn Cuppd a commission on a purchase. Affiliate status has no effect on a coffee's score, its position in any toplist, or whether it appears at all — the ranking code has no access to commercial data.
What does Cuppd deliberately not do?
We do not run professional cupping sessions and do not publish SCA scores. Cuppd is a community rating platform for home coffee drinkers, not a certification body. Where a coffee has a competition result, that comes from the competition, and we say so.
We do not remove low ratings on request, from roasters or anyone else, and we do not fill empty data fields with plausible-sounding guesses.
Found something on a coffee page that looks wrong? Tell us — every coffee page also has a report button for exactly this.