McDonald's Review Drivers

29,259 people told Google about their McDonald's.

Every light here is one of their reviews, from 40 restaurants across the US. Scroll, and they'll show you what goes wrong.

Scroll

Nobody reviews an average burger

27% of reviews give one star. 32% give five. The middle is thin: people write when a visit was great or awful.

1★ through 5★, one dot per review.

Half the happy ones say one word

"Excellent." "Good." "Great." About half of 4 and 5-star reviews (the dimmed lights) mention nothing specific.

Only 23% of 1-star reviewers stay that quiet. Unhappy customers explain themselves, which is why the rest of this story is about them.

The lady at the drive-thru window was so rude and angry (for no reason) that I cancelled my order and drove away.

1★ · Kissimmee, FL · ≈2020

It's the service, not the food

Each cluster is a topic. Further right means more reviews mention it. Lower means those reviews rate lower.

Food comes up a lot but floats near the top. Staff and speed sit far right and low: each pulls the average down by more than a tenth of a star.

1-star reviews sit at each cluster's core. A review that mentions several topics is drawn once, in its rarest topic, so cluster size is not the topic's share.

Rare, but brutal

Wrong orders and closed or broken equipment come up in only 5–7% of reviews. When they do, more than half are 1-star.

12:43 am on a Sunday. They were closed. 24 hours is not accurate, please update the hours.

1★ · New York, NY · ≈2022

Many availability complaints are about listed hours that are wrong. Fixing a store listing costs almost nothing.

Same menu. Very different stores.

Every cluster is one restaurant, with its 1-star reviews at the center. The bigger the red core, the worse the store.

Miami Beach: 62% one-star. A Manhattan store on 7th Avenue: 12%. Same food, same prices, different management.

And it's getting worse

The share of 1-star reviews bottomed out at 22% in 2020. By 2023 it was 32%.

Years are estimated from labels like "3 years ago", so each can be off by about one. 2023 covers January to mid-June.

Had a drive thru order for EIGHT. Every meal was correct and picture perfect.

5★ · Orlando, FL · ≈2023

Good looks like this: fast, correct, friendly. The lights still on are the 5-star reviews.

Now find what's hurting a store

Filter by store, year and topic. Every number, chart and quote below updates as you go.

It's the service, not the food.

Find what's hurting a store

Pick a store, a year range or a topic. Click any dot, bar or store to filter by it.

REVIEWS 29,259 STORES 40 YEARS 2011–2023 AVG 3.18★ SOURCE Google Maps via Kaggle (2023)
Topic

Which topics cost the most stars?

Share of reviews that mention a topic, against how much lower those reviews rate than reviews that don't. Curves mark equal total drag on the average.

Table view

How often it comes up

% of reviews in the current selection. Click to filter.

Average rating by store

Circle size = number of reviews. Colour = average rating compared with all stores.

Which stores are real outliers?

% of 1-star reviews against review count.

Table view

Rating mix by year

Share of reviews at each star rating. Years outside your range are faded.

Complaints that travel together

Of reviews mentioning the row topic, % that also mention the column topic.

What customers say

Rating

How this was built

Data and cleaning

    Topic tagging

    How to read the numbers

    • Rating impact = average rating of reviews that mention a topic minus the average of reviews that don't. It shows association, not cause: an angry customer may mention everything.
    • Drag = share × impact. The overall average would be this many stars higher if reviews mentioning the topic rated like the rest.
    • Funnel limits use Spiegelhalter's random-effects adjustment. Stores differ far more than chance alone allows, so plain binomial limits would flag most stores. The adjusted limits ask which stores stand out even after allowing for that spread.
    • In the story, each review is one dot. A review that mentions several topics is drawn once, in its rarest topic; the dashboard counts it under every topic it mentions.
    • Stores with fewer than 50 reviews (4 stores) count toward totals but are left out of store charts. In filtered views a store needs 30 matching reviews to be plotted.

    Rebuilt from a Tableau version

    • The earlier dashboard used CrowdFlower's 1,471 negative-only reviews. Its "dissatisfaction score" was annotator agreement, not customer sentiment. This version uses real star ratings.
    • Topics keep the original categories (RudeService → Staff & service, SlowService → Speed & wait, OrderProblem + MissingFood → Order accuracy, BadFood → Food quality, Filthy → Cleanliness, ScaryMcDs → Safety, Cost → Price) and add Availability & hours.
    • The radar chart and word cloud are replaced by charts that can be read to scale and filtered.