NHTSA Complaints and TSBs: Finding Known Faults

Put two 2014 SUVs through NHTSA's complaint database and the totals look like they are about to settle something. The Jeep Grand Cherokee comes back with 2,374 complaints. The Ford Escape comes back with 2,076. Both queried 28 August 2026.

That near-tie is worth nothing, and the reason is easy to skate past. NHTSA counts complaints. It does not count cars. There is no sales or registration figure anywhere in that database, so the total is a tally divided by an unknown, which does not stop it reading like a score. What separates two cars sits underneath the total: which parts keep coming up, at what mileage they let go, and whether anyone at the agency thought the pattern was worth opening a file on.

Three files, three different questions

The federal defect data comes in three pieces, and people blur them constantly.

Complaints are what owners typed in. NHTSA does not verify them. A complaint is a statement that someone believes something failed, sometimes written the day it happened and sometimes years later. The agency's web form is one of fifteen intake codes listed in the file documentation, alongside congressional referrals, insurance companies, consumer letters and Early Warning Reporting.

Manufacturer communications are what the factory told its dealers. Technical service bulletins live here, and so do service campaigns and warranty extensions.

Investigations are what the regulator did about it. This is the only one of the three where a government engineer looked at the pattern and wrote down a conclusion.

All three are free, and all three exist twice over: as a search box on nhtsa.gov, and as machine-readable copies you can count. Every figure below came from the second kind, pulled on 28 August 2026 from api.nhtsa.gov and static.nhtsa.gov. The search boxes answer one car at a time, and nhtsa.gov itself came back with HTTP 403 that day, so every number here came out of the downloads rather than off a page. Use the browser forms for one car, the downloads when you are comparing several.

A complaint count with no denominator is not a signal

The complaint endpoint needs no key and no registration:

https://api.nhtsa.gov/complaints/complaintsByVehicle?make=jeep&model=grand%20cherokee&modelYear=2014

For that 2014 Grand Cherokee it returned 2,374 records. Broken out by the component tags NHTSA assigns, they land like this: electrical system 482, power train 473, engine 408, seats 338, then 271 tagged only "unknown or other", service brakes 268, air bags 209, seat belts 187. One hundred and forty-nine records carry the crash flag and 40 the fire flag. One hundred and sixteen report at least one injury, 138 injuries between them. None report a death.

Then the part that changed how I read these. Sorted by the date each complaint was filed rather than by model year, the same 2,374 records run: 65 in 2013, 312 in 2014, 270 in 2015, 182 in 2016, 129 in 2017, 176 in 2018, then back up to 255 in 2019 and 318 in 2020 — the busiest single year, six years after the car was built. A design flaw that was there in year one does not get worse in year six. Recalls, news coverage and an aging fleet pushed people toward the complaint form. That curve is a record of attention as much as of failure.

The component breakdown is the part that survives contact with reality. It is the same question a mechanic would ask, which is what breaks on these.

When two official copies of the same file disagree

The same endpoint, asked for a 2014 Toyota Camry, returns 14 complaints. Fourteen, for one of the most common cars in the country, and the oldest of the fourteen was filed on 30 September 2025 — ten in 2026, four in 2025, nothing before that.

The daily complaint flat file, downloaded the same morning, holds 455 distinct complaints for a 2014 Toyota Camry, spread across filing dates from 2013 to 2026. Neighboring model years agree between the two sources exactly: 2015 is 268 either way, 2017 is 148 and 148, 2018 is 735 and 735. Only 2014 splits, 14 against 455.

Matching the fourteen by ODI number against the file shows what the split is not. All fourteen are in the 455. The API is not returning different complaints, it is returning a recent slice of the same ones. Beyond that I cannot tell you what happened, and neither, in public, can NHTSA: the complaint file's own documentation carries a standing notice dated 28 June 2021 that a system update caused "discrepancies between the contents of the complaints file," that the agency is still working to resolve it, and that website lookups are unaffected. Whether this Camry gap is that known issue, a model-string problem, or something else, is not something the published data answers. Treat both figures as claims with a source and a date attached rather than as the number.

The habit that survives is small and worth having: an implausibly good number is a question about the data, not an answer about the car. Fourteen complaints on a mass-market sedan should send you to a second source before it sends you to the seller.

A related trap sits next to it. The model string has to match what the database calls the car, which is why api.nhtsa.gov/products/vehicle/models?modelYear=2014&make=toyota&issueType=c is worth calling first. It lists CAMRY and CAMRY HV as separate entries, and the equivalent Jeep list returns twelve rows for six distinct models because several are repeated. Query the wrong string and you get a silent zero rather than an error.

Mileage at failure is the column the JSON drops

The API's complaint record has twelve fields: ODI number, manufacturer, crash and fire booleans, injury and death counts, incident and filing dates, a partial VIN, components, summary, product. The flat file has fifty-one. Field 18 is MILES, the vehicle's mileage when it failed. Field 19 is the number of occurrences, 23 the purchase date, 24 whether the complainant was the original owner. Fields 50 and 51, state of incident and vehicle operator, were added on 30 April 2026.

Mileage is the one that matters when you are about to buy a car with 90,000 on it. Below, the 2014 Grand Cherokee complaints that report a mileage above zero, counted once per complaint per top-level component, from the file downloaded 28 August 2026:

Counted once per complaint instead, 1,507 of the 2,374 carry a mileage at all, and their median is 50,000. Read down that table and the powertrain and brake complaints mostly sit behind a 90,000-mile car. The seats column is sitting directly on top of it. That is a different conversation with a seller than "this model has a lot of complaints."

Two cautions. Owners round, which is why 75,000 and 18,000 land so tidily and why a median should be read as a neighborhood rather than a number. And the file is big: 370 MB zipped, 1.59 GB of tab-separated text, 2,239,860 rows covering 1,615,196 distinct complaints. It is a scripting job, not a spreadsheet job.

The bulletin index is where somebody else's money sometimes hides

Manufacturers hand their dealer bulletins to NHTSA because a regulation says so. 49 CFR 579.5, read through the eCFR on 28 August 2026 against Title 49 as of 1 August 2026, reaches communications about "any defect in its vehicles or items of equipment... whether or not such defect is safety-related," and paragraph (a) sweeps in "warranty and policy extension communiqués and product improvement bulletins" by name. Paragraph (d) sets the clock: a copy must be submitted "not later than five working days after the end of the month in which it is issued." The index runs about a month behind the dealer network, which is what the one worked example below shows.

The downloads sit under static.nhtsa.gov/odi/ffdd/tsbs/ in five-year chunks, TSBS_RECEIVED_1995-1999.zip through TSBS_RECEIVED_2020-2024.zip. There was no 2025-2029 chunk on 28 August 2026, so the most recent twenty months of bulletins are reachable only through the site's own search.

The 2020-2024 file alone holds 2,406,749 rows. For the 2014 Grand Cherokee it lists 111 distinct bulletin IDs, weighted toward equipment (29), electrical system (26) and engine and engine cooling (24). The 2014 Camry has 76 in the same window; the 2014 Escape, 26. That is not a reliability ranking. A bulletin count measures how much a manufacturer writes down and files, which varies enormously by brand.

What you are hunting for is the small share that pays. From the same file: document 23-036-19, issued 18 December 2019 and received by NHTSA on 21 January 2020, "Leather Instrument Panel (IP) Cover Delaminating - X81 Warranty Extension," covering 2011 to 2014 Grand Cherokees. An extension like that attaches to the vehicle rather than to the person who bought it new, so a later owner is not excluded by not being the first one — but it is the manufacturer's own program, with its own age and mileage cut-offs, and a 2019 extension may well have run out. The bulletin tells you the program exists; only the manufacturer or a franchised dealer can tell you whether it is still open on a given VIN.

The field that would let you filter for these directly is "Communication Type," added in the May 2024 revision of the file format. In that 2020-2024 chunk, every value other than "Service Bulletin/Repair Instructions" appears for the first time in May 2024: Service Campaign, Warranty Program/Extension, Emissions and Over The Air account for 22,901 rows between them, plus 18,707 tagged only "Other," all from the last eight months of the file. Everything older is unclassified — 23-036-19 itself, warranty extension and all, is filed as a plain service bulletin — so for anything before mid-2024 you are reading titles, not filtering a column.

PE, EA, and the letters that say how serious NHTSA got

The investigations file is the small one: 4.3 MB zipped, 154,367 rows, and only 5,344 distinct actions since 1972. By prefix, PE 2,329, EA 1,933, DP 505, RQ 244, SQ 159, EQ 85, AQ 41. Twenty-two had been opened in 2026 by 28 August, against 37 in all of 2025 and 58 in 2024. These are rare events, which is what makes one on your shortlisted model worth reading in full.

A Preliminary Evaluation is the first stage; if it survives, it is upgraded to an Engineering Analysis, and the summaries say so in those words. A DP is a defect petition filed from outside, and a granted one can open an EA of its own.

The 2014 Grand Cherokee carries a textbook chain. ODI opened PE15-030 on 20 August 2015 "to investigate 14 complaints alleging incidents of rollaway after intended shifts to Park" in 2014 and 2015 Grand Cherokees fitted with the monostable electronic gearshift. Fourteen complaints was the trigger. That PE closed on 5 February 2016, two days after EA16-002 opened. FCA filed a defect information report on 22 April 2016 covering, in the words of the EA's own closing summary, approximately 811,586 vehicles, the EA closed on 24 June 2016, and the campaign number recorded against it is 16V240000. Eight months from a complaint cluster to a recall.

Not every file ends that way. PE19-014 opened on 9 September 2019 on 128 complaints that active head restraints were deploying with no crash behind them, thirteen alleged injuries known at the time. Its close date is 25 February 2026 and its campaign number field is empty. Six and a half years, no recall attached to it in this data. That is exactly the failure you want reflected in the offer, because nobody else is going to pay for it.

Any campaign number you do find here then has to be checked against the actual car. The recalls endpoint lists 19 campaigns touching the 2014 Grand Cherokee across all builds; which of them apply to one VIN, and whether they were ever performed, is a separate lookup covered in checking a VIN for open recalls.

What none of these three files can see

They cannot see the car. That is not an oversight to work around, it is the design. The public complaint record stores field 15 as VIN CHAR(11), eleven characters of seventeen, enough to identify the build and deliberately short of identifying the vehicle. No query here answers whether the car parked in front of you has the problem.

They cannot see anything nobody reported. Complaints are self-selected by people annoyed enough to file one, which is another reason shape beats size. Repairs done quietly under warranty are invisible here, and so are the sales figures that would turn a count into a rate.

What they do is narrow the question. You arrive knowing that on this model the seats and the transmission are where the money goes, that failures cluster around a specific odometer reading, and that one investigation closed with no remedy attached. That is a list for the person you hire to look at the car, and it turns the pre-purchase inspection from a generic once-over into six specific things to check. If one of the six turns up, you already have the mileage distribution and the bulletin number in hand when you sit down with the repair-or-replace arithmetic.

The database tells you what usually goes wrong. Only the car tells you what already has.