How the figures are made

The sources

A place is covered when its government publishes every routine inspection (not only the ones with violations), keeps the file current, and allows reuse. The snapshots are refreshed once a month, and each page prints the date of the newest record and the date we pulled the file.

What gets counted

1. Inspections, not violations. Some files list one row per violation. Counting rows would make a chain with many restaurants look worse simply for having more of them, so rows are grouped into inspections first, and each inspection counts once.
2. Routine inspections only. Complaint visits, licensing and pre-opening visits, and suspected-illness investigations are left out. They are triggered by something, so they are not a fair sample of how a restaurant usually runs.
3. The first visit only. When an inspector finds a problem, they come back. Those return visits happen only at restaurants that already had a problem, so counting them would count the same problem twice.
4. One yes-or-no finding per place. Each place has its own serious outcome, listed above. An inspection either recorded it or did not. We do not count how many violations, or add up points.

The figure and its range

A chain’s figure is k divided by n: the number of its counted inspections that recorded the finding, over all its counted inspections. The comparison figure is the same sum over everything in the file, chains included. In most places that is restaurants; Austin’s file also holds grocery stores, convenience stores and school kitchens and does not label them, so there it is all food establishments.

The range is a 95% Wilson score interval. In plain terms: the inspections we have are one sample of how a chain’s restaurants perform, and a different few months would have produced a slightly different figure. The range shows how far it could plausibly move. A chain with 3,000 inspections gets a range a few points wide; a chain with 110 gets one ten points wide or more.

centre = (p + z²/2n) / (1 + z²/n) half-width = z · √( p(1−p)/n + z²/4n² ) / (1 + z²/n) with p = k/n and z = 1.96

We use Wilson’s interval instead of the simpler textbook one because it behaves properly with small samples and low rates, and never runs below 0% or above 100%. Sources: E. B. Wilson, “Probable Inference, the Law of Succession, and Statistical Inference,” Journal of the American Statistical Association 22 (1927); Brown, Cai and DasGupta, “Interval Estimation for a Binomial Proportion,” Statistical Science 16 (2001), which recommends it.

A chain needs 100 counted inspections in a place to be charted. Below that, the range is too wide to say anything, so the chain is listed with its count and no bar.

Why chains are not numbered first to last

When two ranges overlap, the records cannot say which chain is really lower. In most places that describes most neighbouring chains. A list numbered 1 to 14 would claim a precision the inspections do not have, so the site draws ranges and marks only three states: clearly under the local rate, too close to it to call, and clearly over it. Tables are sorted by measured rate so they can be read, and say so.

Why places are never compared

Each place counts a different event under different rules: Florida counts inspections that recorded at least one high-priority violation; New York City counts inspections that scored 14 or more points; Chicago counts inspections that ended in a Fail; and Austin counts inspections that scored under 90, below the city's top category. A 58.2% in one place and a 33.6% in another are not two readings of the same thing, so no figure on this site combines two places, and there is no national table.

What does carry across is whether a chain sits under, level with, or over its own place’s rate. The homepage sets those side by side, each column on its own axis, and counts them. It does no arithmetic on the rates themselves.

Matching restaurant names to chains

Government files record whatever name is on the license: “MCDONALDS #11371”, “MC DONALDS RESTAURANTS OF FLA”, “WENDY’S PROPERTIES LLC”. We match them to 34 chains with a table of name patterns, currently version 2026-09-20.3. The full table, and every name it matched, is published so anyone can check it.

A location recorded under two chain names, such as a combined KFC and Taco Bell, is one kitchen. Crediting either brand would be a guess, so it is counted for neither (it still counts toward all restaurants).

If a name is matched to the wrong chain, or a chain’s locations are being missed, write to hello@fastfoodfact.com. Corrections change the table, the version number, and every figure that depends on it.

What the figures cannot tell you

• How well a chain is run, on its own. Chains handle different food. A coffee counter has fewer ways to be written up than a kitchen frying raw chicken. Part of any gap between chains is the menu.
• Anything about one restaurant. A figure describes a chain’s franchised and company locations in one place over one period. It is not the brand everywhere, and not the restaurant near you. For a single restaurant, look it up with the agency that inspected it.
• A fair sample of restaurants. Every place here inspects higher-risk and previously cited restaurants more often. They appear more often in any count of inspections, for chains and for all restaurants alike.
• Restaurants that have closed, where the source drops them. New York City’s file holds only restaurants still in business on the day it is published.
• Differences between inspectors or districts. Inspectors differ, and a chain concentrated in one part of a state is seen by fewer of them. The records do not let us adjust for that.

When a source goes quiet

Inspection portals break, move and stop updating; half of the first ten we checked were unusable. If a source’s newest record is more than 60 days old, its page and its column on the homepage are labelled as out of date. If a monthly refresh fails, the last good snapshot stays up with its own dates on it.

Informational only. These are public records, summarized; they are not a safety rating of any restaurant and not professional advice. Last reviewed: September 2026.