Who Actually Opens a Car Recall? The 6% NHTSA Pushes Are the Ones That Cover Millions of Cars

Key facts

• 10,403 distinct vehicle recall campaigns were filed between 2015-01-01 and 2026-09-17, covering 417,165,016 potentially-affected units. (Units at launch, not cars still on the road — see Limits.)
• 9,682 (93.1%) are flagged MFR, the manufacturer opened them. 656 (6.3%) are flagged ODI, they came out of NHTSA's Office of Defects Investigation. 65 (0.6%) are flagged OVSC, a compliance-test finding.
• Median size: 322 vehicles manufacturer-filed, 26,479 investigation-flagged. An 82x gap between the middle campaign of each group.
• Those 6.3% of campaigns carry 36.8% of every recalled vehicle in the period — 153,460,621 units.
• 17 of 17 Part 573 filers with at least 10 campaigns in both buckets show the larger median on the investigation side. Ratios run from 2x (Daimler Trucks) to 135x (Daimler Vans).
• Remove every Takata and airbag-inflator campaign and the gap survives at 47x — 14,895 against 318.
• The field is not in NHTSA's public recalls API, which returns 16 fields per campaign and no initiator. It exists only in the bulk flat file.

Is a car recall voluntary or does the government order it?

Almost every recall is filed by the manufacturer, and the small minority that is not is where the mass sits. NHTSA's recalls flat file carries a field — number 14, INFLUENCED_BY — that its own published record layout defines as "Recall Initiated By (MFR/OVSC/ODI)". MFR is the manufacturer. ODI is the Office of Defects Investigation, the unit that opens preliminary evaluations, engineering analyses and recall queries. OVSC is the Office of Vehicle Safety Compliance, which runs the standards testing.

Across campaigns filed 2015-01-01 to 2026-09-17:

Read the last column first. The typical recall a carmaker opens by itself covers 322 vehicles — a production window of a few days, a bad batch of one part. The typical recall that comes out of a federal investigation covers 26,479.

Why are investigation-driven recalls so much bigger?

Because the two processes are answering different questions. A manufacturer that finds a defect in its own warranty data usually knows which vehicles were exposed: the supplier lot, the plant, the build dates. It files a narrow campaign because it can prove the boundary.

An investigation starts from the other end — owner complaints spread across model years, plants and trims, with no production record tying them together. By the time it closes, the population that cannot be excluded is the population that gets recalled. That is how you get Ford's 1,392,935-vehicle recall of 2015-2017 F-150s for unexpected transmission downshifts (NHTSA campaign 26V237000): NHTSA opened a probe on 2026-01-30, and Ford filed on 2026-04-07 across three model years of one nameplate.

September 2026's largest campaign of any kind works the same way. Jeep's 328,381-vehicle rear coil spring recall (26V562000, Chrysler 89D) is the third campaign for the same defect on that platform, and autoevolution reports it was opened after NHTSA contacted FCA US over roughly 20 complaints of springs separating on vehicles that had already been repaired under the 2023 campaign. The flat file flags it ODI. Nothing on NHTSA's consumer page for that recall says so.

This is the same mechanism we measured from a different angle last week: recall width tracks what the filer can prove, not how likely your particular car is broken.

Does the pattern survive the obvious objections?

We ran the three checks most likely to reverse it. Two confirmed it. One partly bites, and it is reported here rather than buried.

Check 1 — is it just Takata? Airbag-inflator campaigns are enormous and heavily investigation-driven, so they could be carrying the whole result. Removing every Takata and airbag-inflator campaign leaves 9,558 manufacturer-filed and 460 investigation-flagged campaigns. Medians: 318 and 14,895. The gap narrows from 82x to 47x and does not close.

Check 2 — is it just the big-fleet manufacturers? If investigations concentrated on Ford and Chrysler, who build high-volume vehicles, the gap could be a fleet-size artefact. Comparing inside each manufacturer removes that entirely:

17 of 17. Every filer with enough campaigns on both sides shows the same direction.

Check 3 — has it always been true? No, and that is the finding's real limit. Before 2015 the gap does not exist. In 2011 the median investigation-flagged recall was 305 vehicles against 340 manufacturer-filed (0.9x); in 2012, 0.3x; in 2014, 0.1x — investigation-flagged recalls were smaller. The gap opens in 2015 and runs 3x to 400x every year since.

It is tempting to read that as the post-2014 enforcement era arriving. We cannot support that from this file, because how often the flag is applied changed at the same moment: NHTSA flagged 122-159 campaigns a year as investigation-driven in 2011-2014, then 12-75 a year from 2015. When both the count and the size of a category shift together, a coding-practice change and a policy change look identical in the data. So the durable claim here is the cross-sectional one — within the 2015-2026 era, and within each manufacturer — not a story about enforcement getting tougher.

Can you check this against anything outside the one field?

Yes, and it holds. NHTSA's public recalls API carries NHTSAActionNumber, the identifier of the investigation a recall is associated with. The two systems are populated separately, so agreement between them is real corroboration.

We drew a random sample of 40 investigation-flagged and 40 manufacturer-filed campaigns from the 2015-2026 set and queried the API for each:

Zero of 40 is the number that matters. The flag is not noise.

What does this change when you are buying a used car?

It should make you less impressed by a recall's headline number, in either direction. A 1.4-million-vehicle campaign is not evidence that the car in front of you is defective; it is evidence that nobody could narrow the group. A 300-vehicle campaign is not reassurance either — it may be the one that names your exact build week.

The question that actually decides a purchase is narrower: is this VIN still open? That is free to answer at nhtsa.gov/recalls, and it is the check to run before money moves.

What a VIN check can and cannot tell you here

A Zilocar VIN check can show that an open recall exists on a specific vehicle and what component it covers, alongside the things a recall record cannot reach: accident and damage records including airbag deployment, odometer and rollback history, junk and salvage auction records, theft records, ownership and sales-listing history, specs, NHTSA and IIHS ratings, and valuation.

It cannot tell you who opened the campaign — that field is in none of NHTSA's consumer-facing products, only in the bulk file this analysis reads. It also cannot tell you whether the repair was performed on that VIN, cannot show open NHTSA investigations that have not yet produced a recall, and does not report a vehicle's legal title brand, which is NMVTIS and the state DMV's job. For remedy status on your exact car, the authorities are NHTSA's VIN lookup and a franchised dealer.

Method

Source. NHTSA Office of Defects Investigation recalls flat file, FLAT_RCL_POST_2010.txt inside FLAT_RCL_POST_2010.zip. Retrieved 2026-09-22; the archive member's own timestamp is 2026-09-22 07:02 UTC; tab-delimited, 29 fields, no header row, 245,388 rows, 15,258 distinct campaigns. The newest Part 573 receipt date present is 2026-09-17, which is the freshness assertion the script makes before computing anything.

Universe (N). Rows deduplicated to distinct CAMPNO, because the file carries one row per make/model/year within a campaign; POTAFF is taken once per campaign and never summed across its rows. Restricted to RCLTYPECD='V' and to RCDATE (field 16, the Part 573 receipt date) in 2015-01-01 through 2026-09-17. That is N = 10,403 campaigns, 417,165,016 potentially-affected units.

The grouping field. Field 14, INFLUENCED_BY, defined in NHTSA's published record layout as "Recall Initiated By (MFR/OVSC/ODI)". It takes exactly three values across the whole file (plus one blank, outside our window).

Statistic. Medians, not means. Both distributions are severely right-skewed — the manufacturer-filed mean is 26,424 against a median of 322 — so a mean comparison would describe the tail rather than the typical campaign.

Reversal checks, all three reported above. (1) Every Takata/airbag-inflator campaign removed, identified by "TAKATA" or "INFLATOR" in the defect summary or component string: gap 82x to 47x, survives. (2) Repeated within each Part 573 filer with at least 10 campaigns in both buckets: 17 of 17 same direction. (3) Pre-2015 period: the gap does not exist and the flag's application rate changes at the same point, so the time-series reading is not supported and only the cross-sectional claim is made.

Independent validation. A random sample (seed 22) of 40 ODI-flagged and 40 MFR-flagged campaigns queried against api.nhtsa.gov/recalls/campaignNumber: 28 of 40 and 0 of 40 respectively carried a non-empty NHTSAActionNumber. Zero fetch errors.

Limits.

• POTAFF is the manufacturer's launch-time estimate of units potentially affected. It is not a count of cars still on the road, still unrepaired, still in the US, or ever actually defective. This file can support "how much exposure was opened" and cannot support "how many dangerous cars are out there".
• RCLTYPECD='V' means vehicle, not car: it includes motorcycles, buses, heavy trucks, motorhomes and trailers. We did not restrict to cars and light trucks here because the comparison is between two groups drawn from the same universe, but readers should not treat 417,165,016 as a passenger-car figure.
• Manufacturer grouping uses MFGTXT, the Part 573 filer, not the badge on the car. One campaign can span several makes.
• 2026 is a partial year: Part 573 receipts through 2026-09-17 only.
• Three filers in the by-manufacturer table are not passenger-car makers (Daimler Vans, Daimler Trucks, Forest River); they are shown because they met the campaign-count threshold, and the direction is the same with or without them.

Reproduce it. The frozen dataset is who-opens-a-car-recall-nhtsa-investigation-data-vin-check-2026-2026-09-22.csv, tidy long format: one row per (cut, key, initiator flag) with campaign counts, unit totals, medians and the count of million-vehicle campaigns, for the era total, the ex-inflator reversal check, each year 2011-2026, and each manufacturer.

FAQ

Are car recalls voluntary or mandatory? Almost all are filed by the manufacturer. In NHTSA's own recall file, 93.1% of the 10,403 vehicle campaigns filed since 2015 carry the MFR flag; 6.3% carry ODI, meaning the recall came out of a federal defect investigation, and 0.6% carry OVSC. The split matters because the 6.3% are far bigger: a median of 26,479 vehicles against 322.

How do I find out whether NHTSA pushed a recall or the carmaker opened it itself? Not from the consumer tools. NHTSA's public recalls API returns 16 fields for a campaign and none is the initiator. The flag lives only in the bulk recalls flat file, field 14. The usable proxy on the public API is NHTSAActionNumber: in our 40-campaign sample of investigation-flagged recalls, 28 carried an investigation number such as EA15001, against 0 of 40 manufacturer-filed ones.

Does a bigger recall mean the car is more dangerous? No. Recall size measures how wide a population the filer could not rule out. An investigation-driven recall is usually wide because the agency pushed the boundary outward; a manufacturer-filed one is often a few hundred vehicles from a known production window.

What is the median size of a car recall? 322 potentially-affected vehicles manufacturer-filed, 26,479 investigation-flagged, for campaigns filed 2015-01-01 to 2026-09-17. Remove every Takata and airbag-inflator campaign and it is still 47x — 14,895 against 318.

Does this hold for every carmaker or just the big recall filers? Every one we could test: 17 of 17 Part 573 filers with at least 10 campaigns in both buckets, from 2x to 135x.

Does a VIN check tell me why a car was recalled, or who opened it? It tells you a recall exists on that VIN and what component it covers. It does not tell you who opened the campaign, and it does not tell you whether the repair on your car was done. For remedy status, use NHTSA's VIN tool and a franchised dealer.

Were the biggest recalls of 2026 investigation-driven? Several were. The two largest investigation-flagged 2026 campaigns are Ford's 1,392,935-vehicle F-150 transmission downshift recall, filed after a NHTSA probe opened 2026-01-30, and Chrysler's 1,076,999-vehicle Jeep Wrangler and Gladiator power-steering-pump fire recall. September 2026's largest campaign of any kind, Jeep's 328,381-vehicle rear coil spring recall, also carries the flag.

Why does this matter when I am buying a used car? Because a wide recall is weak evidence about the specific car in front of you. The question that decides the purchase is whether this VIN is still open, which the free NHTSA VIN tool answers.

Sources

• NHTSA ODI Recalls flat file (FLAT_RCL_POST_2010.zip) — the primary dataset for every number on this page. Last verified 2026-09-22.
• NHTSA RCL record layout / data dictionary — defines field 14 INFLUENCED_BY as "Recall Initiated By (MFR/OVSC/ODI)". Last verified 2026-09-22.
• NHTSA recalls API, campaign 26V562000 — the Jeep Grand Cherokee rear coil spring recall record, 328,381 units, Part 573 received 2026-09-01; also the field list showing no initiator is published. Last verified 2026-09-22.
• NHTSA recalls VIN lookup — the free open-recall check by VIN. Last verified 2026-09-22.
• The Fast Lane Truck — Ford recalls 1.3 million 2015-2017 F-150s under NHTSA transmission investigation — independent confirmation of the probe-then-recall sequence on campaign 26V237000. Last verified 2026-09-22.
• autoevolution — 328,381 Jeep Grand Cherokee SUVs recalled over rear coil springs that can detach — independent confirmation of campaign 26V562 / Chrysler 89D at 328,381 units, and of NHTSA having contacted FCA US over post-repair separation complaints. Last verified 2026-09-22.
• Frozen dataset for this analysis (CSV) — every figure above, reproducible. Last verified 2026-09-22.