Why your restaurant rating dropped on a delivery app

A rating that fell on one delivery app and not another is usually arithmetic rather than cooking, and restaurant chains lose weeks retraining kitchens over it. Deliveroo averages your last 400 ratings, Uber Eats weights the past 90 days, Wolt uses six months on a 1 to 10 scale, and Just Eat takes the median instead of the mean. The window decides how fast a bad fortnight leaves. Only Bolt Food publishes a route to have a rating struck from the average altogether.

Why did the rating drop on one delivery app and not on another?

Because the same complaints are fed into four different formulas, and a rolling count behaves nothing like a rolling calendar. Deliveroo states that “The rating shown on the restaurant list is an average of the last 400 ratings you have received”, so a busy site replaces its whole rating in weeks while a quiet site carries a bad month for most of a year. Uber Eats measures on a clock instead: the metric “is based on your average customer rating, weighted for the past 90 days”, using “only ratings submitted from completed orders”. Volume changes nothing there.

Just Eat breaks the intuition entirely, because it does not average at all. It states that “The overall rating score is calculated by using the median of the ratings you’ve gained from your customers”, and that a median “prevents a few extremely high or low ratings from unfairly skewing your overall score”. A median sits still through a bad month and then moves a whole point in a week, so anyone reading it as a smooth average will call the fix a success days before it arrives.

Wolt runs two ratings on the same venue. The public 1 to 10 rating covers six months and is recalculated daily, while the rating inside Best of Wolt, which decides a venue’s level, is built from the last two complete calendar months. A kitchen fixed in mid August can lose a Best of Wolt level on 1 September while its public rating barely moves.

Why did my restaurant rating drop when nothing changed in the kitchen?

Ratings on these platforms measure the delivery experience, not only the food, and a large share of what lands on a restaurant’s score was decided after the food left the building. Late arrival, a courier problem, a missing item picked in error, an order the platform cancelled. The kitchen can have an unchanged week and the number can still move, which is why looking for a cause inside the kitchen usually finds nothing.

Which delivery apps let a bad rating be removed?

Two, and they work differently. Bolt Food is the only platform here that documents a merchant initiated correction to the average. Its Merchant Portal guide says a partner “can dispute a rating if you disagree with a rating given by a customer”, names courier performance and delivery time as typical grounds, and states the outcome: “If your dispute is accepted, the rating will be excluded from your average.” A Bolt Food rating is therefore partly a function of how diligently a chain files disputes, which is true of no other platform here.

Just Eat gives grounds rather than an outcome. Its Partner Hub reviews screen offers seven named reasons to report a comment, two of which cover most of what restaurants consider unfair: that it “Mentions meal or item that was not part of the order” and that it “Mentions system issues outside of restaurant/partner control”. DoorDash offers neither, only a private reply inside a seven day window.

What does a rating drop cost besides the number itself?

On four platforms it costs access to promotion, and the thresholds are published. Deliveroo reassesses Marketer access three days before the first of each month and requires, among other criteria, that “Star rating must be 3.8 or more”. A site that misses one criterion loses Adverts and Offers “for a full month”, and the check runs per site, so one branch can be advertising while another two streets away cannot. Careem sets a hard floor for paid campaigns: “Only outlets rated 4.0 and above can run ads.”

Just Eat attaches its rating to a status rather than a tool. Local Legend requires four criteria held “for two consecutive quarters”, including a “Customer food review rating of 3+ stars” and “Time offline at 10% or below”, and the terms list “Priority in giveaways, competitions and marketing campaigns” among the benefits. Glovo folds the rating into a ranking factor and defines a bad one narrowly: “Customers give bad ratings whenever they experience that their order has missing or wrong products or if the meal tastes bad or has a bad quality.” Lateness is measured separately there.

Snoonu is the only platform we have found that warns about a specific habit in advance and puts a number on it, showing “Putting higher than preset delivery times may result in ratings lower than 3.5!” at the moment a branch pads its own prep time.

Which delivery apps have no restaurant rating at all?

Three, and an operator convinced their rating fell on one of them is reacting to something else. Ninja’s restaurant portal has no ratings screen, no reviews list and no average anywhere in it, only “Order Compensations”, each with a named type and an appeal clock. ToYou documents exactly one rating system and it points the other way, “a rating system for ToYou’s Representatives, through which the customers, Merchant as well as the Support and the Fleet Teams, evaluate the Representatives’ performance”. On ToYou the merchant scores the courier.

Keeta is a fourth case. Reviews exist there, but its published API covers Basic, Order, Store and Menu and contains no ratings resource, so a chain integrated with Keeta reads store status automatically and still opens an app branch by branch to see what customers said.

Why does a rating fall after a week when the restaurant was closed?

Because an outage removes the good scores that would have diluted the bad ones. A rating window keeps running whether or not you trade, and a dark listing contributes nothing to it. On Deliveroo, where the window is a count of 400, nothing enters to push the old complaints out, so a site that lost three hours on Friday carries last month’s reviews longer than the site that traded through. The mean Deliveroo interruption in our July 2026 UAE panel ran 3 hours 10 minutes, which is a whole service.

The Saudi version is a frozen star rather than a falling one. A branch card that has quietly stopped trading keeps its last average on display and collects nothing new, which reads as stability in a monthly report. Across our July 2026 Saudi panel the mean incident ran 12 hours 24 minutes, and 5 hours once persistently offline listings are excluded, so that long tail is largely cards with no business behind them. A rating that has not moved in six weeks is the first thing to check on HungerStation, and why we keep two downtime bases is at our methodology. Whether being offline enters any rating formula at all is answered on does being offline hurt your rating.

What should a chain actually do with this?

Pick the lever that exists on each platform rather than running one rating programme across all of them. On Bolt Food the highest return action is filing disputes. On Just Eat it is reporting reviews on the seven grounds while remembering that a median moves late. On Deliveroo it is protecting the 3.8 floor before the monthly reassessment rather than after it. On DoorDash, where the formula is withheld, the store’s own rating history over time is the only measurement available.

All four depend on knowing what the number was last week, which no platform hands you as a series. Because a rating window keeps running through an outage while the orders do not, a rating history is only readable next to an availability history for the same branch and the same dates, and that pairing is what Kitchain (kitchain.co) records from the customer side. Background on the individual platforms is under kitchain.co/aggregators.