Body Shop Quality Control Checklist: Post-Repair Checks With AI

Start from your own comebacks. Give an AI assistant the last thirty returns and your repair stages, ask for one short yes/no checklist per stage (panel fit, paint, refit, electrics and scans, cleaning, photos), then edit every line against the manufacturer's repair information and add a named sign-off box that must be signed before the car moves on.

The AI drafts and tidies the checks. It cannot see paint, feel a gap or know the tolerance for a particular model, so every figure comes from the manufacturer's repair information or your paint supplier, and every tick comes from a person who looked at the car. A checklist nobody signs is a poster; the signature is what makes it quality control.

Follow me on Instagram@sagnikteaches

A comeback costs more than the check that would have caught it

Count before you check. For one month, write down every car that returns, or is stopped at final inspection, for something the shop should have caught: a headlamp that flickers, a paint nib the customer found at home, a trim clip rattling in the boot. Those notes are the raw material for everything below, and they keep the list honest because every line traces back to something that really happened.

Connect on LinkedInSagnik Bhattacharya

The stakes are easy to sum with illustrative figures: 100 cars handed over in a month, 5 returns, and about three hours per return across the strip-down, the fix, the re-clean and the phone calls. That is 15 hours a month. The same 15 hours would pay for nine extra minutes of checking on every car (100 cars at 9 minutes is 900 minutes). A stage-by-stage list does not add nine minutes at the end. It moves most checks to the stage where a fix takes five minutes instead of fifty, and you keep the monthly count going to see whether it is working.

Subscribe on YouTube@sagnikteaches

Turn your last thirty comebacks into stage checks

Strip names, plates and VINs from the notes first and refer to each job by its number; the assistant does not need to know whose car it was. Then paste the notes under a prompt that stops it inventing numbers.

You are helping a small collision-repair shop write post-repair quality checks.
Below are 30 short notes about cars that came back or were stopped at final inspection. Names and plates are removed.

1. Group the notes by the stage that should have caught the problem: panel fit, paint and finish, refit and trim, electrics and scans, cleaning, photos and paperwork.
2. For each group, give the number of notes and write up to four checks. Each check must be one action a technician can do in under a minute and answer yes or no.
3. Do not invent measurements, tolerances, torque values or manufacturer requirements. Where a figure is needed, write "per manufacturer repair information".
4. Return a table with the columns: stage | notes | check wording | how to verify.

NOTES:
[paste the 30 notes here]

A run like this returns something close to the table below (illustrative; the counts add up to the 30 notes).

Then edit. Three corrections come up almost every time:

• Split rows that mix two defects. If the draft files dirt nibs and colour mismatch together under "finish", separate them. The fix, the stage and the person responsible are different.
• Delete any figure you cannot find in the repair information. A confident-looking "gap 3 to 5 mm" is exactly the line to remove or replace with "per manufacturer repair information".
• Reword vague checks into yes/no actions. "Ensure quality finish" cannot be ticked honestly. "No dirt nibs visible on repaired panels under the inspection lamp" can.

Store the finished master list and this prompt in a ChatGPT or Claude Project, so nobody has to paste the list from memory each time. If you want the wider picture of intake, handover and quality steps, AI-drafted shop SOPs covers the neighbouring procedures.

Panel fit: gaps, flushness and what closes properly

Gaps and flushness (how level a panel edge sits with its neighbour) are where a car looks "off" to a customer who cannot say why. The reference is the manufacturer's repair information for the model. Where it gives no figure for a seam, the opposite side of the car is the twin to compare against.

Notice that two items are signed by someone other than the person who did the work. That is deliberate, and the rules for it come further down. The AI's job at this stage is trimming: ask it to keep only the rows that match the operations on the repair order, so a paint-only job does not carry the wheel-torque line and a wheel job does not carry the hose test.

Paint match and finish: what a camera cannot judge

Paint checks are visual, and colour is the hardest to judge. A phone photo records condition (a run, a scratch, a nib) but cannot approve a match, because the camera adjusts white balance and every screen shows colour differently. The decision needs two people looking at the car itself, in daylight at the yard door where possible and again under the booth or inspection lamp.

Blending is the normal way to get an invisible match: the new colour is faded into the neighbouring panel so a slight difference is not noticed. Trade guidance in FenderBender describes a blendable match as the industry standard for matching colour, and recommends a spray-out card (a test card sprayed under the same conditions, with the same gun, air pressure and clear coat) to check the formula before the car is painted. That card is evidence, so it travels with the job. The checks that follow it:

1. Colour on the repaired panel against its neighbours. View at a normal standing distance and again close up, in daylight and under the lamp, with the spray-out card to hand.
2. Blend zone. No visible edge where new paint fades into old, on both sides of every seam. A colour can be right and still look wrong beside an unblended panel.
3. Texture. The surface "orange peel" (the faint dimpling in a sprayed finish) should resemble the adjoining panels, not look glassier or rougher.
4. Defects. Dirt nibs, runs, sags, solvent pop (tiny pinholes), sanding scratches showing through, polishing swirls.
5. Masking lines and overspray. No ridge at the tape edge; no mist on glass, rubbers, lamps or trim.
6. Clear coat carried to the panel edges. Paint makers' warranty conditions may depend on this, so confirm what yours says.
7. Aftercare note. How long to wait before washing or waxing comes from the paint supplier's data sheet, never from the AI's memory.

The last item is a good place for AI. Paste the supplier's data sheet and ask for a four-line customer note in plain words, then compare it with the sheet line by line. Summarising text you supplied is something an assistant does reliably; recalling cure times it was never given is not.

A completed paint sign-off in the job record might read like this (illustrative):

Paint, job J1043: colour checked in daylight at the yard door and under the inspection lamp; blend not visible on the rear door or rear quarter; spray-out card kept in the job folder; texture matches adjoining panels. Two dirt nibs on the upper quarter edge flatted and polished. No overspray on rubbers or lamps. Painter [initials] 14:20. Final inspector [initials] 15:05.

Electrics, scans and calibration before the keys go back

A car can leave with perfect panels and still have an airbag warning light, a dead parking sensor or a driver-assistance camera aimed at the wrong thing. Driver-assistance systems (ADAS: lane-keeping cameras, radar behind bumpers, sensors that steer or brake) make the electrics stage a safety stage. I-CAR's repairer news calls post-repair calibration essential for many of these systems to work properly, and says that while a post-repair scan may be enough for some systems, many need far more steps than hooking up a scan tool and clearing codes. The procedure to follow is the vehicle maker's, and I-CAR's pages link to the makers' own position statements on pre- and post-repair scanning, which is the place to start for the model in front of you.

The generic electrics checks (all exterior lights on low and high beam, indicators, brake and reversing lamps, horn, wipers and washers, windows, central locking, mirrors, warning lights at ignition and after a short drive) are the same on every job. What changes with the repair order is the extra line to add:

The one checklist line to write carefully reads: "Calibration or scan requirement looked up in the manufacturer's procedure: required, not required or not applicable. If required, report attached." Ask the AI to flag repair-order lines that might trigger a calibration so a person looks them up, but never let it conclude "no calibration needed". A negative has to come from the maker's document. I-CAR's own article on the subject is a useful read for the whole team: Post-Repair Calibration Essential for Safety.

Photo evidence: a fixed set, and a missing-shot check

Photos settle most arguments about what the car looked like when it arrived and when it left, but only if the set is the same every time. Fix the list once. For a job with two repaired panels the set comes to 15 shots: VIN plate; odometer; dashboard with ignition on (warning lights visible); four corners; each repaired panel straight on; two gap close-ups per repaired panel; the scan or calibration report; and the driver's footwell and seat. Intake gets the first seven plus any existing damage; handover repeats the set after the work.

Name files so a person and a script can read them: job number, stage, shot, for example J1043_final_rear-quarter-gap-left.jpg. Then use the AI for the one job it does well here: comparing a list with a list. You are sharing file names, not the photos, so nothing about the customer or the car's plate leaves the shop.

Required handover shots, numbered:
1 VIN plate; 2 odometer; 3 dashboard on, no warning lights; 4-7 four corners; 8 each repaired panel straight on (rear quarter left, rear bumper); 9 two gap close-ups per repaired panel; 10 scan report; 11 driver footwell and seat.

These are the file names uploaded for job J1043. Tell me which required shots are missing, using the numbers above. Do not guess from vague names: list those as "unclear".

J1043_final_dash-on.jpg
J1043_final_odometer.jpg
J1043_final_corner-fl.jpg
J1043_final_corner-fr.jpg
J1043_final_corner-rl.jpg
J1043_final_corner-rr.jpg
J1043_final_rear-quarter-left.jpg
J1043_final_rear-bumper.jpg
J1043_final_rear-quarter-gap-left-1.jpg
IMG_4471.jpg

A sensible reply (illustrative): missing shot 1 (VIN plate), shot 9 for the rear bumper (two close-ups) and the second close-up of the rear quarter, shot 10 (scan report), shot 11 (footwell); "unclear": IMG_4471.jpg. Five minutes spent retaking those at the car beats a week of chasing an insurer or a customer for evidence that does not exist.

Sign-off rules that keep the lists honest

A list only works if the signing rules are tight. These five fit on one line each and survive contact with a busy Friday:

1. Each stage has an owner who signs it with initials and the time. A tick without both counts as not done.
2. A failed item blocks the next stage. It is fixed, re-checked and re-signed, and the fail stays on the sheet; do not rub it out.
3. "N/A" needs initials as well. A blank box is not allowed to mean "not applicable".
4. Paint and final inspection are signed by two different people. Nobody is the only judge of their own work.
5. Only the workshop manager can release a car with an open item, and the reason goes in writing on the sheet.

The customer walk-round is the last signature. Take the customer to the car with the sheet, point at the repaired areas and hand over the aftercare note. If a customer spots something the list should have caught, that goes straight into next month's comeback notes.

One rear-quarter job, from first check to handover

This illustrative job puts the stages together: a hatchback with rear parking sensors, a replaced rear bumper and a repaired and resprayed rear quarter. Five defects and one gap in the photo set turned up before the keys left the building.

Total: 37 minutes of fixes, none of them a return visit. The proud seam and the dead sensor were the two that would have ended up as comebacks. Notice that four of the five defects were caught at the stage that produced them, which is the point of signing stage by stage.

Job-specific lists from the repair order

A master list of 60 items is too long to use on every car. Let the AI cut it down for each job by feeding it the operation lines only (no customer details).

Operation lines from one repair order:
- Replace front bumper cover, refinish
- Replace right headlamp
- Replace windscreen
- Refinish bonnet edge, blend into left wing

Using the master checklist below, write a stage-by-stage yes/no list that applies only to these operations. Keep the master wording. Add a line starting "LOOK UP:" wherever the manufacturer's repair information may require a calibration, scan or special procedure. Do not say what the procedure is, and never say that none is required.

MASTER CHECKLIST:
[paste master list]

The output (illustrative) is a list of roughly 25 lines: gaps on bumper, headlamp and bonnet seams; paint and blend checks for the bumper cover and bonnet edge; refit checks for undertrays and clips; headlamp function and aim; rain sensor and wipers; and three "LOOK UP" lines (front camera or radar after bumper and windscreen work; headlamp aim after replacement; scan). What to fix: delete anything not on the order (an assistant will sometimes add a wheel-alignment check because bumper work "often" involves one), keep every LOOK UP line, and print the result on one page for the job folder.

A weekly rework log the AI can summarise

Comebacks are the signal, but a better signal is where each defect was found. Keep a simple log with the columns date, job number, stage that missed it, where it was caught (at the stage, at final check, after handover), minutes lost and a one-line cause. On Friday, paste the table into an assistant and ask for the counts, the stage with the most misses and one check to add or tighten.

This is an illustrative week. By type: five dirt nibs, three trim clips, two sensor faults, two gap issues, one polish mark, one cleaning miss. The pattern worth acting on is that a defect found at the stage costs about seven minutes, one found at the final check costs about 20, and the single one found after handover cost 150. The action it points to is a nib check under the inspection lamp at the booth exit, because two of the five nibs slipped through to the final walk-round.

A realistic mistake from this step: a first summary said 15 defects because it counted one job twice. Counting the table gave 14. For anything that feeds a decision, calculate the totals in the spreadsheet and give the AI the totals to describe, or ask it to list the row numbers behind each count so you can spot-check them.

Invented tolerances and other faults in AI-drafted checklists

• Invented specifications. Tolerances, torque values and "manufacturer requires" claims that look authoritative. Every figure needs a source in the repair information or the supplier's sheet.
• Confident negatives. "No calibration needed" or "no scan required". Only the maker's document can say that.
• Length. Assistants love thoroughness. A list technicians skip is worse than a short one they complete, so cut items that have never failed.
• Generic wording. "Check quality of work" belongs nowhere. Each line should name the thing and the way to check it.
• Data in the wrong place. Keep plates, VINs and customer names out of consumer AI tools. Job numbers and file names are enough.

The habits for checking any AI output before relying on it are covered in how to check AI is doing good work, not just fast work, and they apply directly to the first draft of your list.

Two weeks to a working list

1. Day 1 (about an hour): gather the last 30 comeback or stopped-at-final notes and clean out personal details.
2. Day 2 (an hour): run the prompt, split mixed rows, delete unsourced figures.
3. Day 3 (two hours): sit with the lead painter and lead technician, cut the list to 8 to 12 items per stage and settle who signs what.
4. Days 4 to 5: print it and use it on five jobs. Time how long the final walk-round takes.
5. Week 2: start the rework log. Review it on Friday, change at most two checks, and tell the team why.

After six weeks, compare the monthly comeback count with the one you started from, and look at where defects are being caught. If the share found at the stage is rising and returns are falling, the list is doing its job. If returns are flat, look at which stage keeps missing the same item and rewrite that one line. Customer messages when a car fails a check and is delayed are covered in how garages use AI to send repair updates, and the insurer side of photos and job notes is in AI for body shops: insurer paperwork, supplements and job notes. The same photo-and-list method works for other trades, as shown in photo checklists and quality control for cleaning teams.

Questions body shop owners ask about post-repair checks

Should a painter sign off their own paint work?

Yes for their own stage, because they know what they did, but never as the last word. A second person, usually the final inspector, should look at the paint independently and sign separately. Missed nibs and colour mismatches slip through most often when one person both checks and approves their own work.

How long should a final inspection checklist be?

Aim for roughly 20 to 25 yes/no items on the final page, stage lists of 8 to 12 items each, and a final walk-round that takes under ten minutes. If it takes longer, technicians start ticking without looking. Move any item that never fails to the stage list or drop it.

Can AI check a paint match from a photo?

No. Colour in a photo depends on the camera's white balance, the light and the screen it is viewed on, so a photo records condition but cannot approve a match. Keep the spray-out card with the job, look at the car in daylight and under the inspection lamp, and have two people agree.

Do we need special software for these checklists?

No. A shared spreadsheet with pass/fail dropdowns plus columns for initials and time works, as does a printed sheet in the job folder, provided every tick has a name and a time beside it. Choose whatever your technicians will actually complete at the car. An AI assistant can reformat the same list for either.

Further reads

• Should a Small Auto Repair Shop Use AI? A Decision Checklist — Decide which shop jobs are worth handing to AI at all.
• Auto Repair Intake Form: Let AI Capture the Complaint Clearly — Get the complaint and existing damage recorded properly at intake.
• Best AI for Mechanic Diagnosis: Fault Codes, Manuals and Limits — Where AI helps with fault codes and where it stops.
• Opening and Closing Checklists: Build Them With AI in an Hour — Build the shop's daily start and finish lists the same way.
• Can ChatGPT Read PDFs, Spreadsheets and Photos? What Breaks — What an assistant can and cannot read in your files and photos.
• ChatGPT for Car Mechanics: Fault-Finding Help and Where It Stops — Fault-finding help, and the limits to respect.
• AI Tools and AI Development: The Complete 2026 Guide — the AI hub, including every tutorial in the AI-for-business series.

Sources: I-CAR Repairer Driven News (post-repair calibration and typical calibration requirements, plus its index of manufacturer scanning position statements); FenderBender, 'Blending' (blendable colour matching and test cards).