EV Battery State of Health: How to Read It

Ask an EV owner about their battery's state of health and most will answer with a range number: it said 272 at the handover and it says 249 now. That is not state of health. It is a prediction the car recalculates from how you have been driving lately, and it will swing twenty miles on a pack that has not changed at all — colder mornings, a roof box, a week of highway commuting instead of city errands.

The real figure is a different thing, stored somewhere else. On a lot of cars it is not on any screen you can reach.

State of health is not the number on your dash

Stripped down, state of health is one comparison: how much energy the pack can store now against how much it stored when new. Ninety-one percent means roughly nine percent of the original usable energy is gone.

Two complications sit under that simple sentence, and they explain most of the arguments in owner forums.

There is no single agreed way to compute it. Some systems measure in ampere-hours — charge the pack can hold. Some measure in watt-hours — energy, which folds in the voltage sag that comes with age. Those methods do not return the same answer on the same pack, and neither is wrong. When a dealer's report and a phone app differ by three points, the gap is often definitional rather than a mistake.

Conditions move the reading. Battery management systems infer capacity from what they observe while you drive and charge. A pack read cold, or one that has not seen a full charge in a month, or one read five minutes after a fast-charging stop, can report differently from the same pack read on a mild afternoon after an overnight AC charge. That is not the instrument lying. It is an estimate, and estimates carry conditions.

Hence the rule everything below follows from: a state-of-health number without a date, an odometer reading and a note on conditions is not a measurement. It is a screenshot.

Where the number actually lives on your car

Four places, and the order is worth keeping — each one costs more effort than the one before it.

Start with the owner's manual, searched rather than read. Download the PDF for your exact model year from the manufacturer's owner site and search it for state of health, battery capacity, capacity gauge, high voltage battery. This is the fastest way to learn whether your car has such a display at all, and it is the only source that describes what your particular gauge indicates. The Nissan LEAF, for one, has long carried a separate capacity gauge beside the charge gauge whose segments drop away as the pack ages — but what a segment corresponds to is set out in the manual for that model year, and it is not a straight percentage. Read that page, not a forum's conversion chart.

Then the manufacturer's app, under service or vehicle health. Some brands surface a battery health item here, sometimes as a test you trigger rather than a live readout. Tesla has offered a Battery Health Test reached through the app's Service section and, on some cars, through Service Mode — but whether it exists on your vehicle has depended on model, model year and software release, with items appearing and disappearing between updates. Any step-by-step walkthrough you find, this one included, is version-specific. Open the menu and see what is on your car today.

A dealer or service-centre battery report is third, and it is the only one of the four that produces paper. Franchise tools read the pack diagnostics directly. If you are booked in for a service anyway, ask for the battery health report, and ask for a copy you can keep. A printout carrying a date, an odometer figure and a shop's name is worth more in two years than any screenshot from a phone.

Which leaves the fourth, and where most people actually end up: a third-party OBD app and a Bluetooth dongle. That one needs its own section, because it is where the caveats live.

For 2026 and later cars one part of that fourth route changes, and it comes from the same California section quoted further down this page. Subsections (c)(2) and (c)(3) of 1962.5 require a ZEV to carry an SAE J1962 "Type A" data link connector, uncovered by any panel, flap or dust cap, and to report the required parameters over SAE J1979-3. Subsection (c)(4) then requires those functions and data to be reachable "without the use of any vehicle manufacturer-specific, user-specific, or tool-specific registration, authentication, authorization, login, password, certification, or other mechanism." On a car built to that, an app is reading a standard parameter rather than a vendor request somebody reverse-engineered.

What an OBD dongle is actually reading

Standard OBD-II grew out of emissions diagnostics on combustion engines. High-voltage pack capacity was never part of that universal set, so it is not something a generic code reader pulls from a documented parameter.

What the popular battery apps do instead is send manufacturer-specific diagnostic requests — vendor parameters worked out by observation, brand by brand and often model by model. That is why these apps make you pick a vehicle profile, why a profile can be excellent on one car and vague on another, and why a new model year sometimes breaks an app until somebody maps it again.

Three consequences, none of which make the apps useless:

• Two apps on one car can report different percentages, because they may be reading different underlying parameters or applying different baselines for "new."
• A number can drift a point between readings on the same day with temperature and state of charge.
• What you are shown may be the manufacturer's own internal estimate — itself an estimate, not a laboratory measurement.

Getting value out of one anyway means not treating it as a verdict. Pick one app, read the car under conditions you can repeat, and record every reading. Four readings over two years from one app tell you something real. One reading from a different app than last time tells you less than nothing.

Worth saying plainly at the point where people start diagnosing themselves: I am not an engineer or a technician, and no figure discussed on this page is an assessment of your battery — what this site is and is not sets out where that line runs.

The 2026 rule that puts SOH in the menu, and its five-point escape hatch

There is a piece of California regulation that almost nothing written about battery health mentions, and it is the most concrete text on this page.

Title 13, California Code of Regulations, section 1962.5 — data standardisation requirements for 2026 and subsequent model year zero-emission and plug-in hybrid vehicles — came out of the Advanced Clean Cars II rulemaking. Everything quoted below is from CARB's final regulation order (PDF), read 17 August 2026. Cornell's LII copy is easier to link and read, but it is a copy; the order is the controlling text.

The number has to be on a screen. Under (c)(6)(A) the vehicle must be able to display the battery SOH parameters to the user "without the use of any tools," and (c)(6)(C) requires that display to be reachable "with no more than 5 selectable screens or submenu selections needed to access the parameter from the home or default display/screen," in alphanumeric form and "converted to standard engineering units." No dongle, no dealer visit, no service menu.

And not only the percentage. The parameter group that display rule points back to, (c)(4)(A)2.a., reads "state of health (SOH), distance traveled since SOH last updated or reset, quantity of battery energy remaining in reserve." The middle item is the one to notice. That is the regulation conceding the argument this page has been making since the second paragraph: an SOH figure is a calculation performed at some past moment, and the driver is entitled to know how far back.

The number has to mean one specific thing. Under (c)(4)(A)4.c. the reported value must be "normalized from 0 to 100 percent" and must correlate "to the usable battery energy for the certification range value as measured in accordance with the ZEV Test Procedure." Aimed squarely at the definitional mess further up.

And it may flatter you, by a stated amount. Same subsection: "The reported SOH parameter shall be no more than 5 percentage points higher than the SOH value that corresponds to the measured usable battery energy." Note which direction that runs. It caps how optimistic the display may be and says nothing about the other way. A screen reading 90 can sit on a measured 85 and comply.

Staleness gets a rule too, a loose one. Subsection (c)(4)(A)4.e. lets a manufacturer recalculate SOH only under conditions robust enough to keep it accurate — after a deep enough discharge and a subsequent charge, for instance — provided an update happens "for vehicles in a test group, on average, at least once every 4,000 miles." An average across a test group, not a guarantee on your car, and up to several thousand miles between refreshes. Which is a regulator's way of saying what the forums say badly: if you want the figure to move, drive the car and charge it fully first.

Who gets this, and when: at least 40 percent of a manufacturer's 2026 model year ZEVs and 100 percent from 2027, per subsection (a)(1). Plug-in hybrids follow the same percentages under (a)(2), but only those certified to earn vehicle values under section 1962.4, and they owe a narrower set of these subsections. A 2026 car may or may not be one of the 40 percent.

Now the caveat, and it is a large one — which is why this section carries a read date. Enforcing standards of this kind requires a federal Clean Air Act waiver, and the EPA waiver decision covering Advanced Clean Cars II (90 Fed. Reg. 642, 6 January 2025) was disapproved by Congress under the Congressional Review Act. H.J.Res.88 became Public Law 119-16, approved 12 June 2025, providing that the waiver rule "shall have no force or effect." The California regulation itself has not gone anywhere — CARB's rulemaking page still records the section as filed with the Secretary of State and effective 30 November 2022.

What that combination means for enforcement on any given day in 2026, I could not settle from a primary source, and news summaries are not good enough for a page that quotes regulation text verbatim. So read the section above as what the text says, not as a promise about your car. One thing holds either way: manufacturers write vehicle software on national timelines, so the screen may well turn up whether or not California can compel it. Your menus will tell you faster than any legal question will.

If your car is older, none of this reaches you directly. It still tells you what a defensible state-of-health number is supposed to look like, which is a useful yardstick for the one your app shows.

So what is normal for the age?

Now the question you opened this page for. With a source, a sample size and a date, because an unattributed percentage here would be worth nothing.

Geotab, which collects telematics from commercial and private fleets, published an analysis of more than 22,700 electric vehicles across 21 models reporting an average of 2.3 percent degradation per year and projecting about 81.6 percent of original capacity retained after eight years (geotab.com/blog/ev-battery-health, checked 17 August 2026). Their method: calculate capacity from energy in during charging and energy out during driving, against the change in state of charge across those transfers.

Two findings from the same analysis are more useful than the headline, and both come with the conditions attached. Vehicles whose charging was heavily DC fast charging — Geotab's group is defined as more than 40 percent of sessions above 100 kW — degraded at roughly double the low-power group, 3.0 percent per year against 1.5, which the same page projects out to 76 percent of original capacity at eight years against 88 percent. And hot climates cost about 0.4 percentage points per year against mild ones.

Attach the conditions to all of it. That is an aggregate across a large, fleet-weighted sample, not a specification and not a prediction for one car. Fleet vehicles are driven differently from private ones, the model mix spans chemistries and thermal designs that do not age alike, and the study's own framing notes that individual vehicles vary substantially around the averages.

There is also a shape that any single annual percentage hides. Capacity loss on lithium-ion packs is not a straight line — generally a faster early phase as the chemistry settles, then a long, much flatter middle. So a two-year-old car showing more than "two years times the average" is not automatically a problem, and you cannot multiply any annual rate out to year ten and expect the answer to hold.

Comparing your reading, with made-up numbers

The arithmetic matters more than any figure I could hand you. The inputs below are invented for illustration, not data.

Say your app reports 91 percent, the car is 3.5 years old with 46,000 miles, and you fast-charge roughly once a fortnight on trips. Nine points of loss over 3.5 years is about 2.6 points a year. Held against the published aggregate, that sits above the average but well inside the spread you would expect from a car that sees regular DC charging, especially given the front-loaded curve.

Now the same 91 percent on a car that is 18 months old with 9,000 miles and has never seen a DC charger. Same reading, entirely different conversation — that one earns a service appointment and a request for the dealer's own report to check against your app.

The reading alone told you nothing. Reading plus age plus mileage plus charging pattern told you something.

One number in this area is not invented, and it is a floor rather than an average. California's ZEV standard requires 2026 through 2029 vehicles to be designed so that at least 70 percent of the cars in a test group retain 70 percent or more of the certification range value over a useful life of 10 years or 150,000 miles, whichever comes first; from 2030 the bar becomes 80 percent, measured on average across the test group (13 CCR 1962.4(d)(2), final regulation order read 18 August 2026). That is a design obligation on a fleet, not a threshold your individual car has to clear. What it gives you is the trajectory a regulator was prepared to call inadequate, which beats a forum consensus for judging the number in front of you.

Log it like a measurement

Whatever source you settle on, record the same six fields every time, somewhere you will still find in three years: date, odometer, the figure, where it came from (app and version, or dealer and work order number), state of charge when read, rough ambient temperature.

Twice a year, and always before any decision that turns on the pack. The trend across five entries is the information. A single figure is noise with a decimal point.

Knowing the figure and knowing what your warranty will do about it are separate problems, and this page only does the first. Capacity coverage lives in its own clause of the warranty booklet, with its own threshold, its own measurement method, and its own rules about whether it survives a change of owner. That clause decides a claim. Your app does not. If a low reading is what brought you here, that booklet is the document to open next, and it will not agree with a phone screen. Three current booklets set three different thresholds, and one of the three is not a percentage at all.

On 2026 and later cars that clause has a figure written under it. Section 1962.8(c)(3) requires the manufacturer to warrant the battery against a state of health — defined by pointing straight back at the 1962.5 parameter above — deteriorating to less than 70 percent, for eight years or 100,000 miles, whichever comes first, across the 2026 through 2030 model years, and less than 75 percent from 2031. Notice what that fastens together. The figure a warranty claim turns on and the figure the dashboard is supposed to show become one measurement, which they have not previously been. The waiver question raised earlier hangs over this the same way it hangs over the display rule.

Start with a search, not a purchase

Before you buy a dongle or book a service appointment, open your owner's manual PDF and search it for "state of health". It takes under a minute, and it splits your situation cleanly in two.

If the search returns a menu path, go there now and write down four things: today's date, the odometer, the figure, and the state of charge it was sitting at. That is entry one. Everything you read afterwards is worth something only when it is held against it.

If the search returns nothing, that is an answer too. The car does not display it, and you are choosing between the manufacturer's app, a dealer report, and an OBD app with all the caveats above. Pick whichever one you will still be able to repeat in two years, then stay with it. Consistency of source beats accuracy of source when the thing you are watching is a trend.

While the manual is open, the charging chapter is worth a bookmark: the AC-versus-DC split running through the degradation figures above is the same one that decides whether a bigger wall box does anything for you at home.

Either way you finish today with an entry in a log rather than a number in an argument.