I Didn't Track My Mileage: How to Reconstruct a Log the IRS Will Accept

GigOdo Team · Published July 17, 2026 · Every figure sourced to the IRS unless noted

Can you still deduct mileage you didn't track?

Usually yes - but not by guessing. The tax code bars the IRS and the courts from accepting a round-number estimate for vehicle deductions. What it does accept is a reconstruction: a dated, mile-by-mile rebuild tied to real evidence such as odometer readings, platform records, and delivery history. This guide walks through that rebuild step by step.

The stakes justify the effort. At 2026 rates, every 1,000 business miles you can document is worth $725 to $760 off your taxable income. A delivery driver who put 12,000 working miles on the car is sitting on roughly a $9,000 deduction - if the paper trail supports it. The rest of our Mileage & Records series covers keeping the log; this post is the rescue plan for the year you didn't.

Why you can't just estimate: Section 274(d)

Vehicle expenses live under the strictest substantiation rule in the tax code. Section 274(d) says no deduction is allowed for a car's business use unless you substantiate the amount of mileage, the date and destination, and the business purpose of each use. An honest-sounding estimate satisfies none of those elements.

This is why the famous Cohan rule doesn't save you. In Cohan v. Commissioner (1930), a court held that judges may estimate ordinary business expenses when records are thin. Congress later overrode that rule for cars: because a passenger vehicle is "listed property," the strict rules of Section 274(d) apply, and The Tax Adviser confirms courts cannot estimate those expenses no matter how credible the taxpayer is. The regulation, 26 CFR 1.274-5T, requires every element to be substantiated by adequate records or sufficient corroborating evidence.

IRS Tax Topic 510 puts it plainly: "The law requires that you substantiate your expenses by adequate records or by sufficient evidence to support your own statement." A number with nothing behind it is neither.

What an acceptable record looks like

Per IRS Publication 463, an adequate record is an account book, diary, log, trip sheet, or app record showing the date, mileage, destination, and business purpose of each business use - kept at or near the time of the trip. Pub 463 treats a log updated weekly as timely.

Reconstruction lives in the gap the rules leave open. Pub 463's incomplete-records provision lets you prove an element with your own written statement containing specific information plus "other supporting evidence that is sufficient to establish the element." The regulations set the bar: after-the-fact records need corroborating evidence with a high degree of probative value to approach the credibility of a contemporaneous log. One honest caveat - the formal "reasonable reconstruction" right in the regulation covers records destroyed by events beyond your control, like a fire or flood. A never-kept log leans on the incomplete-records path instead, which is why every rebuilt entry must tie to outside evidence.

Step 1: Anchor the year with odometer readings

Start by bounding total miles. Oil change invoices, inspection reports, tire receipts, and repair orders almost always record the odometer with a date. Two readings that bracket the year - say, a December service and the following January's - establish how many miles the car drove in total. Nothing you claim can exceed that ceiling.

Pull every service record you can find: the shop can usually reprint invoices, dealerships keep service history, and some states record odometer readings at inspection or registration. Even a dated photo of the dashboard helps. These anchors do two jobs: they cap the claim at a believable total, and they show an auditor your numbers grew out of the car's actual history rather than a target deduction.

Step 2: Pull what your platforms already recorded

Every platform holds some record of your driving, but coverage varies widely. DoorDash emails a yearly mileage estimate; Uber's tax summary reports online miles; Instacart and Amazon Flex report nothing at all. Whatever your platform provides is a verified floor for the rebuild - a number a third party will stand behind - never the final total.

DoorDash's help center says mileage estimate emails go out by January 31 to Dashers who dashed by car, covering on-delivery mileage only - accept to drop-off. Miles between offers and repositioning are absent, and EntreCourier's testing found the estimate ran at least 10% below actual driving even on deliveries, with untracked between-order miles typically adding 30-50% more. We break the whole document down in how to read your DoorDash annual summary.

Uber's tax summary is broader: it includes all online miles - waiting for a trip, en route, and on-trip - though miles driven off-app to reposition still aren't counted. Instacart provides no mileage figure to shoppers, and the Amazon Flex app navigates routes without reporting miles, which is why our DoorDash and Amazon Flex pages tell drivers to treat their own log as the primary record.

Step 3: Rebuild the days from delivery history and bank records

Now fill in the days. Your delivery history - order timestamps, restaurant and customer neighborhoods - shows exactly when and roughly where you worked. Weekly pay statements prove which days you drove. The goal is a day-by-day log, each entry showing the date, miles, area driven, and purpose, each backed by a document.

Bank and card statements corroborate from the outside: deposits confirm active weeks, and fuel purchases place you on the road on specific dates. For miles, map your typical patterns - home to your usual zone, average delivery distances from the trip history - rather than inventing one flat daily number. A rebuilt Tuesday that reads "March 3, 41 mi, dinner rush, Northside zone, DoorDash deliveries (12 orders per app history)" is substantiation. "About 40 miles every workday" is an estimate wearing a log's clothes.

Step 4: Use sample periods the way Pub 463 allows

You don't necessarily need to rebuild all 52 weeks at the same depth. Publication 463 lets an adequate record kept for part of a year support the whole year - if you can demonstrate with other evidence that the sampled period is representative of the rest.

Pub 463's own example: records for the first week of each month showing 75% business use, plus invoices showing the business ran at the same rate all month, are sufficient for the year. For a driver, the parallel is rebuilding a few weeks per quarter in full detail, then showing pay statements with similar order counts and earnings across the other weeks. If you started logging properly partway through the year, those logged months are your strongest sample - provided your platform records show the unlogged months looked the same.

What auditors and the Tax Court reject

Reconstructions fail in predictable ways: round numbers, identical entries repeated week after week, totals with no dates or destinations, and logs that first appear during the audit. The Tax Court has seen every version, and the opinions read like a checklist of what not to hand an examiner.

In Velez v. Commissioner (T.C. Memo. 2018-46), a taxpayer produced two mileage logs created two days before trial, built from a calendar he never entered into evidence. The court rejected logs made "several years after the relevant vehicle use," denied the deduction, and upheld a negligence penalty. In Eze v. Commissioner (T.C. Memo. 2022-83), the taxpayer's calendars listed identical trips on identical dates two years running, with odometer figures prepared during the exam - every vehicle deduction was denied.

The lesson isn't that reconstruction is hopeless; it's that reconstruction must look like evidence, not like arithmetic working backward from a wish. Varied daily miles, real dates, named zones, and documents behind each claim are what separate the two.

The 2026 math: two rates, two halves

2026 is the rare year with a split rate, so a reconstruction must divide miles by half-year before multiplying. Miles driven January 1 through June 30 deduct at 72.5 cents each under IRS Notice 2026-10; miles from July 1 through December 31 deduct at 76 cents under Announcement 2026-11.

Date-stamping matters more than usual this year: 10,000 reconstructed miles are worth $350 more if they fall in the second half. Your log's dates decide which rate applies - one more reason undated totals don't survive. Full background on both rates is in our 2026 IRS mileage rate guide.

The real fix: start the contemporaneous log today

Reconstruction is a rescue, not a system. Every month you log properly from now on is a month you never rebuild, and a logged second half doubles as the representative sample supporting a reconstructed first half. The IRS standard is a record kept at or near the time of driving - an app satisfies it automatically.

That's what GigOdo is for: automatic trip detection with a purpose note on every drive, free with no trip cap, and totals computed at the correct split rate all year. Come filing time, the report pack hands your preparer a log instead of a shoebox. Start it before your next shift, then rebuild the past on your own schedule.

FAQ

Sources: IRS Publication 463; IRS Tax Topic 510; 26 CFR 1.274-5T; IRS Notice 2026-10; Announcement 2026-11 (IRB 2026-29); The Tax Adviser on the Cohan rule; Velez v. Commissioner, T.C. Memo. 2018-46 (Taxpayer Advocate appendix); CPA Practice Advisor on Eze v. Commissioner, T.C. Memo. 2022-83; DoorDash Dasher Guide to Taxes; Uber tax summary help; EntreCourier mileage testing. This article is general information, not tax advice.