The news hook is a rare "the fix didn't fix it" moment. On 2026-08-20 NHTSA opened Engineering Analysis EA26005 into GM's 6.2L L87 V8 engine, after owners reported engines failing after the recall repair — some on brand-new replacement engines. Rather than echo that single case, we downloaded NHTSA's complaint file and computed a cross-make view of the same phenomenon: which used models draw the most complaints about a defect recurring after a completed recall repair.
Key facts
- Load-bearing finding: 3,457 NHTSA complaints (from 2,240,289 processed, retrieved 2026-08-30) match a strict pattern describing a failure after a completed recall repair.
- Confirmation band, stated honestly: a manual read of a 50-row random sample found ~54% clearly genuine post-remedy vehicle re-failures and ~66% including borderline cases — so the genuine-vehicle floor is roughly 1,870–2,280. The strict count is not "confirmed"; it is a bounded estimate.
- Top models by complaint count (all model years): Ford F-150 110, Jeep Grand Cherokee 104, Ford Escape 66, Ford Fusion 66, Ford Focus 60.
- This is a count, not a rate. It is confounded by model popularity, fleet age and cumulative recall exposure, so it favors old, high-volume, heavily recalled models. It is not a "most likely to re-fail" ranking.
- Emerging signal: the eight GM L87 truck/SUV families (MY2021-2026) hold 75 matching complaints — 73 filed since Jan 2025, 57 in 2026 — the fastest-rising post-remedy cluster in the file.
- What a VIN check proves here: recall presence only. It cannot show remedy status, cannot flag the EA26005 investigation, and cannot map a VIN to the probe scope.
Does a completed recall mean a used car is fixed?
No. "Completed" on a recall record means a manufacturer or dealer logged that a remedy was performed, not that the underlying defect has been eliminated. Owner complaints regularly describe the same failure returning after the recall work, and in some cases after a full part replacement. Treat a completed status — or a seller who simply says the car "was fixed" — as a claim to verify with documentation, not as proof.
There are two separate gaps. First, a repair can be recorded but ineffective: NHTSA's investigation into GM's L87 engine exists precisely because failures kept happening after the prescribed remedy. Second, on a used car you often cannot see remedy status at all — a VIN check and NHTSA's free tool both show whether a recall applies, not whether it was performed. If you are weighing a car with any recall history, our explainer on what does recall completed mean when buying a used car walks through how a "completed" stamp can still leave you exposed.
Which used vehicles draw the most complaints about failing after a recall repair?
By our count of NHTSA complaint narratives, the Ford F-150 (110) and Jeep Grand Cherokee (104) lead, followed by the Ford Escape and Ford Fusion (66 each). These are high-volume, older, frequently recalled nameplates, which is exactly why they top a count. The Jeep Grand Cherokee's history of multiple large recalls is why it appears so high; if you are shopping one, see used jeep grand cherokee problems to check before buying.
The table below is a count of matching complaints, all model years combined. Read it as "where post-repair complaints cluster," never as a per-vehicle failure rate — NHTSA's complaint file has no fleet-size denominator to support a rate.
| Rank | Make & model | Post-remedy re-failure complaints (all MY) | Notable model-year peaks |
|---|---|---|---|
| 1 | Ford F-150 | 110 | 2014 (28), 2013 (22) |
| 2 | Jeep Grand Cherokee | 104 | 2005 (16), 2012 (12) |
| 3 | Ford Escape | 66 | 2008 (13) |
| 3 | Ford Fusion | 66 | 2013 (20), 2010 (14) |
| 5 | Ford Focus | 60 | 2012 (16), 2000 (11) |
| 6 | Ford Windstar | 59 | 2001 (17), 2000 (13) |
| 7 | Ford Explorer | 53 | — |
| 7 | Honda Odyssey | 53 | 2002 (14) |
| 9 | Chevrolet Malibu | 51 | 2009 (12) |
| 10 | Honda Accord | 49 | 2003 (11) |
| — | Chevrolet Silverado 1500 | 44 | 2021-2024 rising (L87) |
| — | GMC Yukon XL (2022 alone) | 15 | 2022 emerging (L87) |
The single most notable modern entry is the GMC Yukon XL 2022 (15) — the only 2020s vehicle in the top model-year tier, and a direct read-through to the GM L87 story below.
What happened with the GM 6.2L V8 (L87) recall — and why is it the warning case?
GM's L87 engine is the clearest live example of a recall fix under scrutiny. NHTSA opened Engineering Analysis EA26005 on 2026-08-20, covering 997,743 MY2021-2026 GM trucks and SUVs with the naturally aspirated 6.2L L87 V8, after logging 499 complaints of engine failure after the recall remedy — 473 after the oil-viscosity remedy and 26 after a complete engine replacement — plus 191 reports of failures in engines built after the original recall's May 31, 2024 build cutoff. (These 499/191/997,743 figures are NHTSA/GM's, attributed as context; our own load-bearing number is the 3,457 above.)
The underlying recall is 25V-274000: 597,571 units, MY2021-2024, across the Chevrolet Silverado 1500, Tahoe and Suburban; GMC Sierra 1500, Yukon and Yukon XL; and Cadillac Escalade and Escalade ESV. The defect is a connecting-rod and/or crankshaft manufacturing problem that can damage or seize the engine; the report was received 2025-04-24 and owner letters mailed 2025-06-18. The remedy is a dealer inspection — passing engines get higher-viscosity 0W-40 oil, a new filter and cap; failing engines are repaired or replaced. Our full breakdown of that specific case lives at gm 6.2 v8 engine failure after recall fix. In our own data these eight families went from roughly two lifetime post-remedy complaints to 75, effectively all in 2025-2026.
Can a recalled engine fail again after the fix — or even after a new engine?
Yes, and the L87 case is the sharpest illustration: NHTSA reports 26 of the 499 post-remedy failures occurred after a complete engine replacement, not just after the lighter oil-viscosity remedy. A recall repair is a manufacturer's best available fix at the time; it lowers risk but does not certify that a given vehicle will never exhibit the defect again.
For buyers, the practical consequence is that "the recall was done" and "the problem is solved" are different statements. This is also why the biggest recalls are worth extra caution — completion is imperfect and remedies can be revisited, as covered in is it safe to buy a used car with an open recall.
How do I verify a recall repair was actually done on a used car's VIN?
Use the authoritative sources first. Enter the VIN at NHTSA's recalls tool (nhtsa.gov/recalls) to see whether any recall is open on that vehicle, then call a franchised dealer of that brand and ask them to confirm remedy status by VIN — dealers can see whether the recall service was performed. Ask the seller for dated repair-order documents that name the recall campaign. These steps, not the recall flag itself, are what tell you a repair happened.
A VIN history report is a useful complement for the surrounding picture. A Zilocar VIN check screens whether a recall such as 25V-274 is present on the VIN — the same visibility as NHTSA's free tool — and adds accident and airbag-deployment records, odometer/rollback checks, salvage and junk-auction records, theft (NICB), ownership history and sales-listing history (past listings, prices, mileage, days on market), which can expose an engine-troubled truck that has been repeatedly relisted or quickly flipped. It does not confirm the recall was remedied and does not track NHTSA investigations — use it alongside the dealer and NHTSA steps, not instead of them.
What a VIN check can and can't tell you here
| Question | VIN check (Zilocar or NHTSA free tool) | Where to confirm instead |
|---|---|---|
| Is a recall (e.g. 25V-274) associated with this VIN? | Yes — recall presence | — |
| Was the recall repair actually performed? | No | Franchised dealer / NHTSA VIN lookup |
| Which remedy was used (oil change vs engine replacement)? | No | Franchised dealer / service records |
| Did the repair later fail? | No | Service records / dealer |
| Is this VIN inside the EA26005 investigation scope? | No | NHTSA ODI |
| Accident, airbag-deployment, salvage-auction, odometer, theft, ownership, listing history | Yes (Zilocar) | — |
By the numbers
- 2,240,289 — complaint rows processed from NHTSA FLAT_CMPL.
- 510,607 — rows mentioning "recall"; 3,457 — strict post-remedy re-failure matches (vehicle-only).
- ~1,870–2,280 — genuine-vehicle floor after manual sampling (54–66% of the strict set).
- 125,873 — loose "recall + recurrence" matches, rejected as the metric: a 50-row sample read out to only ~15% genuine, dominated by parts-unavailable boilerplate.
- 75 — matching complaints on the eight GM L87 families (MY2021-2026); 57 filed in 2026 alone.
Based on 2,240,289 NHTSA ODI complaint records (FLAT_CMPL, file dated 2026-08-29), retrieved and processed 2026-08-30; strict narrative classifier, vehicle-only. Counts are complaint volumes, not per-vehicle rates. Frozen data: post_remedy_failure_ranking_2026-08-30.csv.
Method
Source file. NHTSA Office of Defects Investigation Complaints flat file, FLAT_CMPL.txt extracted from FLAT_CMPL.zip. The archived file is internally dated 2026-08-29; we downloaded and processed it on 2026-08-30. NHTSA republishes this file continuously, so a later download will not reproduce these counts exactly — the frozen copies linked below will.
Universe. · 2,240,289 complaint rows (tab-delimited; field 4 MAKETXT, field 5 MODELTXT, field 6 YEARTXT, field 16 DATEA, field 20 CDESCR free-text narrative). The file stores one row per cited component, so every figure here is a count of records, not of people or vehicles.
Filter. Two steps, both over the CDESCR narrative, case-insensitive. Step 1: 510,607 rows contain the string "recall". Step 2, the strict post-remedy re-failure pattern: keep a row whose text asserts a failure occurring after a completed recall repair — matching (a) "failure occurred after the recall", (b) "after (the) recall (was) repair/repaired/work/performed/completed/done/remedy/fix", (c) "even after (the) recall", or (d) "recall repair/work/remedy/fix (was) performed/completed/done" — and a recurrence token (fail, again, same problem, persist, recur, no longer, broke, blew, seized, tore), unless one of the two strong phrases ("failure occurred after the recall", "even after the recall repair") is present on its own. Rows carrying the parts-unavailable boilerplate ("had not experienced a failure", "parts not available", "not yet available", "exceeded a reasonable amount of time", "parts distribution disconnect") are excluded unless a strong phrase is present.
Cleanup and dedup. The strict pattern returns 3,525 rows. Dropping non-vehicle products (tires, child seats), motorcycles and unknown-year rows (YEARTXT 9999) leaves 3,457 — the load-bearing number. Make and model strings are grouped as a make + model pair with no further normalisation; nothing is dropped for low counts, and all 1,477 make/model/year rows are in the frozen ranking CSV. No de-duplication is performed across ODINO, because the component-row grain is the unit NHTSA publishes.
Aggregation. Counts per make + model across all model years, and separately per make + model + model year.
The check most likely to reverse this, and its result. A narrative regex can match text that does not actually describe a post-repair failure, so we hand-read random samples of both candidate metrics. The loose set — "recall" plus a recurrence token, no exclusions, 125,873 rows — read out to only about 7-8 of 50 genuine (~15%), dominated by Takata "parts not available" and "had not experienced a failure" boilerplate. That loose metric is rejected and is reported only as a cautionary upper bound. The strict set read out to about 27 of 50 clearly genuine post-remedy vehicle re-failures (54%) and about 33 of 50 including borderline and collateral cases (66%). So the manually confirmed genuine-vehicle floor is roughly 1,870 to 2,280. The strict 3,457 is a bounded estimate, not a confirmed count — roughly 34-46% of its rows are noise, while the pattern simultaneously misses paraphrases no regex catches. Both figures are published so anyone can re-read the narratives themselves.
Known limits. (1) NHTSA complaints are self-reported and unverified. (2) This is a count, not a rate — the file carries no fleet-size denominator, so the ranking is confounded by model popularity, fleet age and cumulative recall exposure, and skews to old, high-volume, heavily recalled nameplates. It is not a "most likely to re-fail" ranking. (3) Counts are records, not vehicles or owners. (4) The classifier reads owner narratives, not repair orders: it cannot confirm that a recall remedy was in fact completed on the vehicle described. (5) Complaint volume also tracks how much press a defect received, so an investigation or a class action can lift a nameplate's count independently of engineering. (6) The GM L87 cluster (75 matches) is recent and still accumulating; it is an emerging signal, not a settled total.
Frozen data. · post_remedy_failure_ranking_2026-08-30.csv (1,477 make/model/year rows with counts) and post_remedy_failure_complaints_frozen_2026-08-30.csv (all 3,457 filtered rows with their full narratives, so the classifier can be audited row by row).
Try a VIN check before you trust a "fixed" claim
Before buying any used car a seller calls "recall-fixed," run the VIN through NHTSA's free tool and a franchised dealer to confirm the actual repair — and use a Zilocar VIN check for the history those sources don't show: accident and airbag-deployment records, salvage and junk-auction records, odometer/rollback checks, theft, ownership and sales-listing history, specs, NHTSA/IIHS safety ratings and market valuation. It confirms recall presence, not remedy status, so pair it with the dealer and NHTSA for the full picture.
