The Estimate-and-Authorization Problem on Your Auto Shop’s Website

A driver messages your repair shop’s site late at night. The car grinds when they brake, and they want to know what is wrong, what it will cost, and whether it can wait until Friday. The website assistant answers all three: worn brake pads, a price, and a green light to keep driving for a few days. Nobody put the car on a lift. Nobody heard the noise. The customer shows up two days later already certain about the diagnosis, the bill, and the safety call, none of which came from a person at the shop.

Auto repair is more regulated than it looks, and a general-purpose chatbot does not know it. It was added to the site to book appointments. What it actually does is diagnose, quote, and clear a vehicle for the road, three things that carry real consequences when they turn out wrong.

A Quote Sight-Unseen Collides With the Estimate Rules

Most states regulate this trade tightly. The recurring rule is simple: the shop has to put an estimate in writing and get the customer to authorize the work before touching the car, and it cannot blow past that estimate without permission. The point is to keep people from getting ambushed by the bill. Now picture a chatbot saying "brakes usually run about $350" before anyone has seen the vehicle. That number sticks. When the inspected estimate comes back higher, because the rotors are scored or a caliper is seizing, the customer is not hearing a revised quote. They are hearing a bait and switch, and the shop is the one left explaining the difference.

"Safe to Drive" Is a Safety Representation

The reassurance is the most dangerous part. "It’s fine to drive until Friday" is a safety judgment about a vehicle no one examined, on a complaint, grinding brakes, that can indicate a genuine hazard. If the customer drives on it and something fails, the shop made a documented safety representation with no inspection behind it. A general-purpose bot offers that reassurance because keeping the customer comfortable is its default, not because the assessment is sound.

Repair Promises Are Warranty Statements

Then there is the warranty layer. A bot that says "this fix is guaranteed" or "that repair is covered" is making a warranty representation, and a written promise about the work can bring the Magnuson-Moss Warranty Act into play, on terms the shop never set. Misstating what a repair includes or guarantees also sits within the FTC’s authority over unfair and deceptive practices. The customer leaves believing they have coverage the shop never agreed to provide.

"Will Not" Is a Suggestion. "Cannot" Is an Architecture.

Here is what trips shops up. A driver who wants answers does not ask once and accept "bring it in for an inspection." They push. "Just roughly, what’s a brake job run, and can I make it through the week?" A chatbot steered only by a prompt eventually answers the version of the question it was not specifically warned about, because being helpful is its default and a prompt is only a polite request to suppress that default.

That is the gap between a tool that is told not to diagnose or quote and one that is built so it cannot. A disclaimer bolted onto a system whose whole job is to give confident answers does not change what it does when a motivated customer leans on it. It just puts the diagnosis, the price, and the safety clearance in writing.

Who Owns the Answer

Strip away the software and this is an old problem. Picture a service writer who diagnoses a car over the phone, names a firm price, and tells the customer to keep driving on bad brakes, all without seeing the vehicle. No shop would defend that. Put the same behavior in a website widget and the risk does not go away. It lands on the shop. Now there is a written record, no technician was in the loop, and the estimate and safety rules assume a human did the work.

The shops that get burned are not the ones that modernized their websites. They are the ones that dropped in a generic chatbot, assumed helpful and responsible were the same thing, and found out otherwise when a customer waved a transcript at the bill. The fix is not to take AI off the site. It is to run an assistant that books the inspection and shares only general information, then leaves the diagnosis, the estimate, and the safety call to a technician.

Frequently asked questions