You can perform a calibration perfectly and still lose the argument about it. The scan tool said pass, the target was square, the tech knows the job got done — and none of that is in the file three months later when an adjuster short-pays the line or an attorney asks what you actually verified. The work and the proof of the work are two different deliverables, and most shops only get paid reliably for one of them.

That gap is what a Collision Industry Conference panel spent its time on this summer, in a session Autobody News wrote up on Sept. 10. The topic was artificial intelligence, but the useful part wasn't the estimating-side AI everyone has heard about. It was two operators describing something narrower: pointing an AI at the OEM procedures a shop already pulls, and using it to force the repair plan to be followed and photographed.

The repair planner still does the thinking

Nader Akbari of Central Florida Paint and Body told the panel his shop uses Carmic to help technicians pull what they need out of OEM procedures — and he was specific that the AI is not the one deciding what matters.

The repair planner, who understands the automaker service information, compiles the documents for that specific job first. Only then does the technician get to query it. "The AI is only looking at the documentation that the repair planner has put in there, and the repair planner has taken the time to put in the accurate information for that specific repair," Akbari said. From there it becomes a lookup problem: one-time-use parts, torque specs, and, in his words, a visual answer to a question like how to remove a particular bumper.

That distinction matters more than it sounds. A general-purpose model asked what a vehicle needs for a front radar calibration will produce something plausible and unsourced — the last thing you want in a liability file. A model constrained to the procedures your planner already pulled for that VIN is doing retrieval, not invention.

Sergio Palma, founder of Carmic, framed the same idea from the vendor side: shops are sitting on material they already paid for. Loaded OEM procedures contain safety considerations, materials, adhesives, and — the line worth underlining — required photos. He described prompting an AI to return everything pertinent to a single repair line item, including any required photos needed to maximize reimbursement documentation. Palma sells the tool, so read the pitch accordingly. But the underlying claim is checkable against any procedure set you own: the OEM already told you what to document, and most shops never systematically extract that list.

The checklist is the actual news

The part of the panel that translates directly into money is what Akbari described next.

Once the repair plan is written and ingested, and labor is assigned, the system generates a checklist for each technician. The tech checks items off and attaches a photograph of each one being done. Akbari's reasoning was a shop owner's reasoning, not a technologist's: "I've always had that concern that we take hours to write a great repair plan, but then is all of this stuff actually happening?"

That is a verification loop, and the industry has spent 2026 arguing about who owes one. A separate CIC panel this cycle weighed the shop-insurer rift over safety inspections. Industry leaders remain publicly split over how calibration work gets identified, performed and documented. A per-technician checklist with timestamped photos settles none of that. It just means you arrive at the argument holding evidence.

Where the documentation gap is costing real money

The reimbursement data says the gap is specific, not general. In the January 2026 Who Pays for What? results, 90% of shops said they were paid for calibrations always or most of the time by the largest eight insurers, up 0.9% from the prior read. Post-repair scans came in at 93%, pre-repair scans at 92%.

So calibration itself is largely a solved billing question. The unpaid work is everything around it:

• Vehicle safety inspections: 32% paid always or most of the time — down 3.3% from the previous survey.
• In-process scans: 28%.
• Seat calibrations: 70%, well below the 90% for calibration generally.
• DTC research: about 40% of shops have never itemized it at all, though that's improved from 50% in 2020. Of those who do bill it, only 21% are paid always or most of the time; 46% get paid some of the time.

Collision Advice owner Mike Anderson, who runs the survey with CRASH Network, made the case that diagnostic labor deserves to be separated from scan labor entirely. "Isn't the time it takes to research one diagnostic trouble code different from researching five?" he said. "And it's not just looking up the code. You also have to navigate the OEM repair flowchart to determine what the most probable cause is."

Every one of those low-paid lines is a research-and-documentation line. They get denied because they look like overhead and the shop has nothing granular to show. That is precisely the category a procedure-grounded checklist produces a record for.

What It Means for Your Shop

You don't need to buy anything this week to act on this. Three things are worth doing regardless of tooling:

Extract the photo requirements from your OEM procedures, once, per common platform. You already pull the procedures. Nobody is reading them for the documentation list. Build that list for your top five vehicle lines and it becomes a standing checklist.

Separate diagnostic research from scan time on the estimate. Anderson's point is that they're different operations with different variability. If 40% of shops never itemize DTC research, an adjuster has no baseline to compare you to — and that cuts both ways. Itemizing is how the baseline moves.

Attach the photo to the operation, not to the job. A folder of 200 loose photos is not evidence. A photo tied to a named line item, with a tech's name and a timestamp, is. That is the whole functional difference the CIC panelists were describing, and it works with a shared drive and discipline if that's what you have.

One caution: none of this improves the underlying procedures. If your planner pulls an outdated bulletin, an AI will answer confidently from the wrong document. Grounding a model in your own files removes hallucination risk about the content; it does nothing about whether you pulled the right content. The research step still belongs to a human who knows the automaker's information.

The File Is Half the Job Now

The trade has spent two years arguing about who is allowed to see OEM data. That argument mostly resolved. What replaced it is a harder problem: proving, later, to someone skeptical, that a specific procedure was performed on a specific vehicle by a specific person. Calibration gets paid at 90%. The work that surrounds and substantiates calibration gets paid at 28 to 32%. The shops closing that spread aren't doing different repairs — they're producing a different record of them.