How we build these pages

Sources

Fuel economy comes from the EPA and Department of Energy dataset behind fueleconomy.gov. Recalls come from NHTSA's recall file, owner complaints from NHTSA's complaint file, investigations from NHTSA's investigations file, and crash ratings from NHTSA's 5-Star Safety Ratings program. Each page shows the date of the files it uses.

Most searched lists

The "Most searched model years" lists on the home and make pages pick model-year pages by US Google searches a month for their recalls and problems, or their MPG, from Google Keyword Planner (September 2025 to August 2026): every year of the 20 most-searched models, and the most-searched years (390 or more searches a month) of the others. Search data only decides which pages these lists show; no figure on any page comes from it.

Matching a model across sources

EPA and NHTSA name vehicles differently, and NHTSA uses one set of names for recalls and another for complaints. We define each model once, list exactly which EPA and NHTSA names it covers, and print those names on every recalls page. Separate models that share a name, such as the Rogue Sport or the Silverado EV, are kept apart; each page says what it includes. NHTSA files some recalls for the light-duty pickups under a name it also uses for the heavy-duty trucks (Silverado and Sierra for 2011 to 2016, Dodge Ram for 2009 to 2011). Such a recall is listed for the light-duty truck unless NHTSA's summary for that model year points only to the heavy-duty trucks, and the page names the ones left out.

Problems owners describe

NHTSA files each owner complaint under one or more components, such as engine or power train. That says where a problem is, not what it is, so we also read the complaint text. For each problem we list, such as oil dilution, stalling, phantom braking or a cracked windshield, we wrote patterns for the ways owners describe it and count the complaints whose text matches, once per complaint. Most patterns look for the description itself (“the oil smelled of gas”, “the brake pedal went to the floor”). A few, such as fuel pump failures, look for the part next to words like failed or replaced, and a few, such as backup camera, wiring, suspension or steering rack problems, count complaints that name the part, since owners seldom name those parts unless something went wrong with them. A match does not count when the owner says it did not happen (“no oil leaks”) or describes a risk (“could cause a stall”), and for parts often named only because a recall covers them, a match that is what the recall is for does not count. Airbags not deploying is counted only in complaints NHTSA files under air bags. Samples of the matches for every problem were read, and the patterns corrected, before publishing. The counts are approximate: some descriptions are missed, and a few complaints that only mention a problem are counted. A year's page lists problems described in at least 3 complaints and at least 1% of that year's complaints. A summary names a problem as the one owners describe most only when at least 5 complaints and 5% of that year's describe it; otherwise it names the area NHTSA files the most complaints under, when at least 5 complaints name it. When a warning light is the most-described item, the summary also names the most-described problem after it.

Investigations

Recalls pages list the NHTSA investigations that name each model year, from NHTSA's investigations file (FLAT_INV), which holds every investigation since 1972 by make, model and model year. An investigation here is a preliminary evaluation, an engineering analysis, a recall query or an equipment query. Each shows NHTSA's subject, the type, when it opened and closed, NHTSA's model names and component, and the first sentence of NHTSA's summary (cut at about 220 characters), with a link to NHTSA's full record.

Defect petitions, which ask NHTSA to open an investigation, are not counted as investigations: a page lists the open ones separately and leaves closed ones out. Audit queries are left out too: they check whether a company, such as a maker, dealer or rental firm, met its legal duties on reporting or recalls, not whether a vehicle has a problem.

A recall NHTSA ties to an investigation is linked only when it is on the same page, since it can cover other model years. Because the file lists only make, model and model year, an investigation into one engine or trim still names the whole model year; its summary says which versions it covers. When an investigation names 20 or more models, the page gives the number: each make and model counts once, however NHTSA spells the make (Mercedes-Benz or Mercedes Benz), and parts listed without a model year, such as air bag inflators, are not counted.

Years to avoid and best years

Each model page ranks the model's years by the owner complaints filed with NHTSA in each year's first three years (for model year Y, complaints filed by the end of Y+2), so every year gets the same time on the road. Years without three full years of complaints are not ranked yet, and a launch or final year whose complaints come to under a quarter of those of the model's typical recent year is left out as a partial year with few sold.

A typical year is the median of the model's ranked years, taken over all of them and over the years since 2016, whichever is higher: owners filed fewer complaints in the early 2010s, and a model's sales rise and fall over its life, so a year has to stand out against both. A year to avoid drew at least twice the complaints of a typical year, and at least 50; the page lists every such year.

Best years come only from the years since 2016: up to three that drew under three quarters of the complaints of a typical year since then (the median of those years), fewest complaints first. A year is left out when an open NHTSA investigation names it (a preliminary evaluation, engineering analysis, recall query or equipment query with no close date; defect petitions and audit queries don't count), when it has a recall telling owners not to drive or to park outside until it is repaired, when it drew under a quarter of the complaints of the years on either side, or when the year before or after is a partial model year; both usually mean few were sold. A do-not-drive or park-outside recall that covers only a variant, such as the plug-in hybrid or vehicles with a tow hitch harness, doesn't rule a year out; the page mentions it instead. The 2020 model year is never picked: its first three years (2020 to 2022) fell in the pandemic, when people drove less, and for many models it drew fewer complaints than the years around it. When no year qualifies, the page says so and why. No best years are picked when a typical recent year drew under 25 complaints.

A model with fewer than four ranked years, or with a typical year under 25 complaints and no year to avoid, gets no verdict. Complaints are unverified and not adjusted for sales, so a verdict shows where owners reported trouble, not whether a particular car is sound: check any car's recalls by VIN before you buy.

Checks before publishing

Every recall in NHTSA's recall file is matched against NHTSA's recalls API. Campaigns the model-level API does not return are looked up one by one by campaign number, and any that NHTSA's own record contradicts are removed (1 so far). Model years with no recalls in the file are checked against the API too. Complaint totals are compared with NHTSA's complaints API and with the counts NHTSA prints on its crash-test records. Every gas, hybrid and diesel rating is checked against EPA's own fuel cost formula (6,502 ratings). A hard error, such as a recall the API lists that the file lacks or a rating that doesn't add up, keeps that model year's pages off the site until it is fixed; smaller gaps are logged for review.

After each build

A separate script re-reads the raw government files and compares the version tables, summary figures, recall lists, investigation lists, complaint counts, areas and described problems, each model's years to avoid and best years, and comparison figures on every page with what the files say. If anything disagrees, or if a new file has fewer recalls or complaints than the last one, the update is not published.

What the figures can and cannot tell you

EPA ratings are lab-based estimates; real-world mileage varies. Complaints are unverified owner reports that rise with sales volume and time on the road. Recalls apply to specific vehicles, so check yours by VIN at nhtsa.gov/recalls.

Updates

Pages are rebuilt from the latest EPA and NHTSA files. Each page names the dates of the files it uses, and a page's own date changes only when what it shows changes.

Pages kept out of search results

Some pages are here for readers but ask search engines not to list them (noindex): one-year MPG pages, since a model's MPG by year page lists every year's ratings; comparison pages; recalls pages of model years with fewer than 25 owner complaints, or fewer than 50 when NHTSA lists no recall for the year, unless a recall tells owners not to drive or to park outside; every recalls page of the newest model year, which has less than a year of owner reports, even one with such a recall; model pages with fewer than 100 complaints in all; automaker pages with fewer than three models on the site; and MPG by year pages of models with fewer than 5 EPA-rated model years.