How we choose
Yumma publishes short guides — the cheapest breakfasts in Leeds, budget student meals in Edinburgh — and each one is a list of real dishes at real places, with prices you can check. This page explains where those lists come from, because a ranked list is a judgement and you deserve to know who made it.
The short version: software chooses the meals, from data already on this site, and writes the sentences around them. It is never asked what a restaurant sells. Every price in a guide links back to the page it was read from.
Last updated 26 August 2026.
Where the prices come from
Nothing on Yumma starts with a guess. Places are found in OpenStreetMap, then our crawler visits each venue's own website and stores the menu pages and PDFs it finds. A language model reads those stored pages and pulls out dishes and prices; it only ever sees evidence we fetched, never its own recollection of a restaurant.
Each published price keeps the URL it was read from and the date it was read. That link is on the meal's page and on every guide that mentions it, so “where did this come from?” always has an answer.
Menus change and ours will sometimes be behind. Where a price has moved since a guide was written, the guide shows the current price and says what it used to say.
How a list is chosen
The choosing is ordinary code, not a language model. A guide starts from every meal already published in that city, and narrows it by whatever the guide is about — a price ceiling, a kind of food, a dietary claim the venue itself published, or distance from a university campus or the city centre.
What survives is ranked by how many readers have given the meal a Yum, then by price, then by how close it is. One dish per venue, so a list is a list of places rather than one menu in ten parts.
- If too few meals qualify, the guide is not published — a top ten padded out to ten is not a top ten.
- A meal that has appeared in a recent guide for the same city is held back from the next few, so the guides do not become the same list retitled.
- No two guides for a city may overlap by more than a small share of their entries.
What the software writes, and what it may not
Once the meals are chosen, a language model is given that fixed list — the dish names, the prices, the venues and what those venues say about themselves — and asked to write the headline, the introduction and a couple of sentences about each entry.
It cannot add an entry, remove one, or reorder them. Anything it writes that mentions a price or a venue it was not given is rejected before publication rather than corrected, on the principle that a piece of writing which invented one fact is not trustworthy about the rest.
Guides are bylined to Yumma, never to a person. Nobody at Yumma has eaten at most of these places, and a made-up reviewer's name would be the cheapest possible way to lose your trust.
What we are not doing
No venue can pay to appear in a guide, or to be placed higher in one. Sponsored content, where it exists anywhere on Yumma, is labelled as such on the item itself.
Rankings are not reviews. We are not telling you a meal is good — we are telling you it exists, what it costs, and where that price was published. The Yum counts that inform the order come from readers, one vote each.
Guides are published city by city from the data in that city. We do not write about places we have no menu for.
If a guide is wrong
Prices move, dishes are withdrawn and places close. Every entry in a guide links to the meal's own page, where anyone can suggest a correction or report the listing — and a venue that verifies its email domain can edit its own prices directly.
Anything else, including a request to be left out of the guides entirely: hello@yumma.co.uk.