Reading owner complaint data without fooling yourself

Anyone can search the federal complaint database. Almost nobody gets anything useful out of it, because the agency shows you filings one at a time in date order — and a single complaint is an anecdote from a stranger. The finding only appears when you count thousands of them by component, and then it appears very clearly.

This guide is about the four ways that count goes wrong, all of which we hit while building the aggregation on this site.

A complaint count is not a fault rate

This is the mistake that matters most, and it is almost irresistible. NHTSA publishes no sales figures. So a car that sold two million units and a car that breaks constantly produce the same shape of number, and nothing in the data distinguishes them.

A 2019 Lexus ES shows 34 complaints. A 2015 Toyota Camry shows 401. Read as reliability that is a twelvefold difference and a clear verdict. It is nothing of the kind: the Camry is one of the best-selling cars in the United States and the ES is a low-volume luxury saloon. The ratio is mostly a sales chart.

What it does show, when you count by component

Within one vehicle the proportions are meaningful, and they are often startlingly specific. We ran the aggregation across twelve nameplates picked to try to break it. Eleven resolved; ten produced a result that was both concentrated and recognisable. Nobody involved knew these stories in advance — they fell out of counting.

Each of these is public knowledge only through owner forums, class actions and word of mouth. None of it required curation. The data was always there; it just needed adding up.

Below a certain number of filings, ranking is noise

A 2019 Mazda MX-5 Miata has 11 complaints on file, and its most-reported component has three. You can sort that list and produce a “top” component. It would mean nothing — three people is not a pattern, and with a different three the order changes.

This site refuses to rank below 25 filings, or where the leading component has fewer than five. Under those thresholds it reports the total and declines to name anything. The ranked list is not built at all rather than built and hedged, because a number on a screen next to a component name will be believed no matter what the caption says.

Missing data is not a clean record

NHTSA's coverage has holes. Its own model catalogue lists the Alfa Romeo Giulia for 2017 through 2019 and then stops — nothing for 2020 to 2022. Ask the catalogue whether that car exists and the answer is no. Ask the recall endpoint about a 2020 Giulia directly and it returns nine campaigns.

A tool that used the catalogue to decide whether to ask would report that car as having no recalls. That is not a small error; it is the most dangerous direction to be wrong in, because “nothing found” is what a person acts on.

The agency's own vocabulary drifts

Complaints carry a component label chosen at filing time, and the same fault is not always labelled the same way. The Outback's cracking windshield is filed under both VISIBILITY and VISIBILITY/WIPER. Counted separately, the windshield came second behind the battery drain. Counted together it leads — 387 against 362.

The obvious explanation is wrong, and we checked: these are not renames. Both labels are used concurrently across the whole 2012–2026 span, so the choice between them carries no information at all. Merging them is honest. Hiding the merge would not be, so where this site combines two agency labels it shows you both.

There is a subtler version. Some NHTSA categories contain a comma — FUEL SYSTEM, GASOLINE and SERVICE BRAKES, HYDRAULIC are single categories, not two. Split on every comma and they shatter into halves that get counted apart. Over 10,788 filings we validated the separator: a bare comma divides categories, a comma inside a name is followed by a space. Getting that wrong put the wrong component first on two of the eight vehicles we tested — the Civic's fuel system was split three ways and reported behind steering when it actually leads, 145 to 105.

What a complaint actually is

• Unverified. A member of the public filled in a form. Nobody inspected the vehicle. It is not a finding against the manufacturer and not a safety rating.
• Self-selected. People file when angry, after a recall makes news, or when a forum thread tells them to. Coverage of a defect can move for reasons that have nothing to do with the defect.
• Not a recall. Recalls are manufacturer campaigns with a remedy attached. Complaints sometimes precede one and often do not.

None of that makes them useless. It makes them evidence about where a given model draws attention, which is genuinely worth knowing before you buy one — and it is not a verdict on whether any particular car is safe. For that, an inspection by a mechanic and a VIN-level recall check are the tools, and neither of them is this website.

Look it up yourself

Everything above comes from data you can read directly. The aggregation is the work; the records are public.

• NHTSA vehicle safety — the agency's own front door
• NHTSA recall lookup — by VIN, which is the only way to speak to one particular car