In a repair shop AI helps in two places: in the office (quotes, customer letters, reviews) and increasingly in diagnostics. The risks sit exactly where safety and money are at stake: invented technical values from AI summaries — and pricing or customer data in open tools.
By Patrick de Kathen, Founder of KlarComply · Reviewed on
Language models answer technical questions with impressive precision — including when they invent the answer. A torque “per the standard” that does not exist, a fill quantity from a hallucinated data sheet: in a workshop such values are not a cosmetic flaw but a safety risk with liability attached.
Safety-relevant values come exclusively from manufacturer documentation and repair databases — never from a language model. AI may summarise and translate; every value is checked against the original source before it goes into a work instruction or onto a vehicle.
Modern diagnostic systems ship with AI features that interpret fault patterns and suggest repair paths. Nothing wrong with that — as long as one thing stays clear: the suggestion is a tool, the repair decision is the master technician's responsibility. New diagnostic or workshop tools enter the business through the approval process (data protection, contract, classification), not through a mechanic's free download — that is how shadow AI is avoided.
A damage photo with a readable number plate or a person in frame is personal data; the job management system holds names, addresses and vehicle data. The same line as everywhere applies: not into open AI tools. The complaint reply or the service-due reminder can be drafted anonymised — name and plate go back in afterwards.
Hourly rates, purchasing terms and margins are trade secrets — and their legal protection under the EU Trade Secrets Directive requires that you take reasonable steps to keep them secret. The AI may supply structure and wording for a quote; you insert the numbers yourself. Copying the full calculation into a public tool “just to check it” risks more than a data leak.
AI-drafted replies to reviews are permitted and sensible — friendly, factual, without customer data. For images the labelling line applies: a photorealistic AI image of the “workshop at night” that could be taken for a photo is disclosed as an AI image (Article 50 of the EU AI Act, since 2 August 2026); recognisable illustrations stay free.
Whoever works for fleets, leasing companies or insurers increasingly sees AI questions in framework contracts and supplier questionnaires: tool list, responsible person, training records. A maintained AI inventory plus a record per person answers that in two minutes.
Free 2-minute check; training (35–45 minutes per person, with a publicly verifiable certificate), inventory, policy and audit file in the subscription from €49/month — cancel monthly.
No. Safety-relevant values such as torque figures or fill quantities come exclusively from manufacturer documentation and repair databases. Language models invent such values convincingly — AI may shorten the path to the source, it does not replace the source.
A readable plate makes the vehicle attributable to a person — which makes the photo personal data. It does not belong in open AI tools; in approved systems with a contract, for workshop documentation, it is unproblematic.
The free quick check walks through ten points — AI inventory, training status, transparency duties, responsibilities. Two minutes, no sign-up, an honest result and your concrete gaps.
Start the quick check — free