Methodology

How the Eat Score works

The Eat Score is yeet.food's own score from 0 to 100, and higher is better. It is not a Just Eat rating: Just Eat has no Eat Score. It rewards consistently positive ratings and penalises poor experiences: occasional low ratings have little effect, while a substantial negative share receives a stronger penalty. The 1-star band weighs slightly more than the 2-star band. It is an index, not a probability of enjoying your order.

Just Eat's rating and ours

A median tells you about the middle rating. A takeaway can have a median of 5 while a large minority of its customers rate it 1 or 2 stars. The Eat Score is designed to show that minority.

What goes in

Just Eat publishes each restaurant's rating as a count of ratings in five buckets, from 1 star to 5 stars (half-star ratings are grouped into these buckets). Just Eat also reports a larger total that includes ratings outside the counted sample. The Eat Score uses only the counted ratings, as Just Eat's rating does. It uses no written reviews.

The current snapshot was taken on 2026-10-07: 70,941 UK takeaways with 6,964,719 counted ratings, out of 39,530,692 in total.

Listings not taking orders. Just Eat keeps some listings that have stopped trading, often still showing opening hours. We leave a listing out of maps, search, rankings, top lists and area figures when it was offline when we captured its ratings and had no new order rating for 180 days before that capture. Offline alone is not enough: many takeaways are switched off for an evening and back the next day. This snapshot leaves out 4,995 listings. Their profiles stay online with a note, and they return automatically once they are taking orders or being rated again.

The calculation

1. Count the ratings. Let N be all counted ratings and c_i the count in band i. Band 1 contains ratings of 1 and 1.5; band 2 contains 2 and 2.5; band 3 contains 3 and 3.5; band 4 contains 4 and 4.5; band 5 contains 5. Band 4 receives 90% positive credit and band 5 receives 100%. Band 3 contributes to N, but earns no positive credit and no additional negative-rating penalty.
2. Adjust small samples. Let p_i be the typical UK takeaway's share of ratings in band i, frozen as described below. The adjusted proportion is q_i = (c_i + 5 × p_i) / (N + 5). This is equivalent to adding 5 reference ratings and can raise or lower the observed shares.
3. Weight negative feedback. Negative severity S = 1.10 × q₁ + 0.95 × q₂. These are proportions, not percentages. The 1-star band contributes slightly more than the 2-star band.
4. Apply the growing penalty. Eat Score = 100 × (0.90 × q₄ + q₅) × exp(−4 × S²). Here exp(x) means e raised to the power x. The squared severity makes rare negative feedback inexpensive, while widespread negative feedback has a much larger effect.
5. Round once. Calculate with full precision, then round to one decimal place within 0–100. The formula does not subtract and floor negative scores at zero. Very small values can still round to zero. Listings without usable counted ratings are excluded.

The reference distribution is the typical takeaway's mix of ratings: the average of each takeaway's band shares, with each of the 48,645 UK food venues that had at least 20 counted ratings on 26 September 2026 weighted equally. From band 1 to band 5 it is 15.5%, 5.4%, 9.2%, 9.6%, 60.4%. Weighting takeaways equally stops the busiest venues, which tend to be rated lower, from dominating it. The same reference is used for every area and future refresh, so an unchanged set of ratings keeps its score.

The 5-rating adjustment, positive credit weight 0.90, negative weights 1.10 and 0.95, and penalty strength 4 are editorial choices, not parameters proven to predict future orders. Moving feedback from band 1 to band 2 reduces the penalty; moving it into a positive band also earns positive credit. The score rewards positive feedback rather than neutrality, so an entirely middle-band distribution can score below a mixed positive/negative one.

Updated on 7 October 2026: the small-sample adjustment now adds 5 reference ratings rather than 10, in the typical takeaway's mix rather than a mix pooled across all ratings. The pooled mix gave every takeaway about two imaginary 1-star ratings, which penalised new takeaways for low ratings no customer gave. Established takeaways move by about a point.

Updated on 27 September 2026: band 4 now receives 90% of the positive credit of band 5, giving excellent feedback a modest advantage over good feedback. The growing penalty replaces the previous adjusted-positive-minus-adjusted-negative formula. It separates distributions previously tied at zero while making occasional negative ratings less influential. Historical research previews retain their original formulas.

Worked examples

These are made-up takeaways, calculated by the same functions as the site. Displayed components are rounded; the final calculation uses full precision.

A busy takeaway. 170 counted ratings: 128 (75%) in buckets 4–5 and 27 (16%) in buckets 1–2. Its adjusted positive share is 75.1% and adjusted poor share is 16%. All 27 negative ratings in this example are in band 1. Applying the band-4 credit weight and growing penalty gives an Eat Score of 66.4.

A new takeaway. 5 counted ratings, all in buckets 4–5. After adjustment, its Eat Score is 80.5. This is limited evidence, not evidence of poor service. Check the rating count and the raw distribution alongside the score.

Middling versus poor. With 500 ratings, 80% positive and 20% middling gives 79.9. Keeping 80% positive but changing the other 20% to band-1 feedback gives 65.8. This difference reflects the aim of avoiding disappointment.

Top lists

The “Highest Eat Scores” lists on location pages include only takeaways with more than 15 counted ratings, the same group as the area figures, so a handful of ratings cannot top an area. The map, search and each takeaway's profile still show every Eat Score.

“Many 1–2★ ratings”

A takeaway gets the “many 1–2★ ratings” label when it has at least 20 counted ratings and more than 15% of them are in the 1–2 star buckets, which is more than about one rating in seven. It is yeet.food's editorial label and threshold, not Just Eat's.

Rating gap: where the methods disagree

Rating gap = |Just Eat rating × 20 − Eat Score|. We convert Just Eat’s 1–5 rating to 20–100 and take the absolute difference from the Eat Score. The index ranges from 0 to 100 points. For example, Just Eat 5/5 and Eat Score 60/100 produce a Rating gap of 40.0 points.

Higher means greater numerical disagreement, not worse food. This is our chosen comparison index: the two methods measure different things, and a five-star median does not mean every customer was happy. The gap can run in either direction; we show both original ratings alongside it.

“Most disagreed” lists and Rating gap sorting include only venues with at least 15 counted ratings and both scores available. We rank by the gap, then by counted ratings when gaps tie. Smaller samples can still have a gap shown on their profile, but are excluded from these rankings. This minimum reduces the influence of very small samples; it does not guarantee reliability.

The comparison uses the same current counted rating aggregates as the Eat Score and Just Eat rating. The historical review archive is not used. No location-specific rescaling is applied: a restaurant has the same gap everywhere.

Area figures

Every location page opens with the same short set of figures. They use takeaways with more than 15 counted ratings (16 or more) and a Just Eat rating, and are only shown where at least 10 takeaways qualify.

• Show a perfect 5: Just Eat’s rating is exactly 5 out of 5. A rating of 4.75 is not counted as 5.
• A plain average of the same ratings (shown on each takeaway as its Expected rating, to one decimal): the mean of each takeaway’s individual order ratings in the same 12 months as Just Eat’s counted ratings, rounded to the nearest whole star (4.5 and above counts as 5). We found those individual ratings in Just Eat’s public reviews feed and kept an average only where their number matched Just Eat’s counted ratings to within 2% (or 2 ratings). We publish one average per takeaway, never the individual ratings or comments. The figure is quoted only when averages exist for at least 95% of qualifying takeaways. It is not used in the Eat Score or in rankings.
• 1 in N ratings was the lowest possible: across the five-star takeaways, the share of counted ratings in Just Eat’s lowest band (1 or 1.5 stars).
• Many 1–2★ ratings: the five-star takeaways that also carry the “many 1–2★ ratings” label. The “5★ but many 1–2★” list shows them, largest share of 1–2★ ratings first.
• Reviews that go into today’s ratings: counted ratings as a share of the lifetime review totals Just Eat shows beside its stars, across every mapped takeaway in the area.

Why compare with a plain average? Just Eat’s rating is the median order rating of the last 12 months. Because most ratings are a perfect 5, a median shows 5 whenever just over half a takeaway’s orders were rated perfect, whatever the other ratings were. A plain average of the same ratings shows how much that choice matters. Neither is the “true” quality of a takeaway.

What the Eat Score is not

• It is not Just Eat's rating, a mean rating, or a prediction that your order will be good.
• It does not detect fake ratings, and it cannot see why a rating was given. People who submit ratings may differ from those who do not; Bayesian smoothing does not remove that selection bias.
• It is a dated snapshot of historical ratings, not a live inspection. Restaurants change.
• The buckets group half-star ratings, so “4–5 star” is not an exact count of individual four- and five-star ratings.
• The reference distribution, its 5-rating weight, the growing penalty and band weights and the “many 1–2★ ratings” threshold are editorial choices. Different choices would rank some takeaways differently.
• Promotions play no part: a stamp card or discount does not change a takeaway's Eat Score.

Independence

yeet.food is independent data journalism about Just Eat's ratings. It is not affiliated with Just Eat. The ratings come from Just Eat's public ratings data at the snapshot date. Made by Ashley Hindle.