Turning a customer's symptom into a diagnostic question checklist
A vague complaint — "it shudders sometimes on the highway" — can send a diagnosis down the wrong path fast. Experienced techs structure the hunt: the right intake questions, then the cheapest, most-likely checks first. AI is genuinely useful for building that question-and-test checklist so nothing obvious gets skipped. What it must never do is diagnose the car or hand you a procedure — that comes from testing and OEM service data, and the model will fabricate both with total confidence.
You are an experienced shop foreman helping me STRUCTURE a diagnostic approach — not diagnose the vehicle. I will do the actual testing against OEM service data. Vehicle: {{vehicle}} Customer's complaint in their words: {{symptom}} Any trouble codes pulled: {{codes}} When it happens: {{conditions}} Produce three sections: 1. Questions to ask the customer — the intake questions that would narrow this down (when it started, conditions, recent work, frequency). 2. Checks to perform, ordered — a logical sequence from quickest and most-likely to rule out first, toward more involved. Describe WHAT to check, not how. 3. Candidate systems — the systems worth investigating for this symptom, each labeled as a candidate to confirm or rule out. Hard rules: - This is a starting checklist, NOT a diagnosis. Do NOT tell me what is wrong with this vehicle. - Do NOT provide torque specs, procedures, wiring details, component locations, fluid types, or capacities — I get those from OEM service information (ALLDATA / Mitchell ProDemand / factory manual). If you don't have a real value, you must not invent one. - Do NOT cite TSB numbers, recall numbers, or "known issues" as fact — flag them as "check service bulletins yourself" instead. - Label every possible cause "candidate — verify by testing." Never state a cause as certain.
Fill in your details and the prompt updates live — then copy.
QUESTIONS TO ASK THE CUSTOMER - When did it start, and did any service (plugs, fuel, tires) happen just before? - Does it track with speed, engine RPM, or load (uphill, AC on)? - Steady or does it come and go within a single drive? CHECKS TO PERFORM, ORDERED (what, not how) - Recheck for stored and pending codes and freeze-frame data. - Inspect and evaluate ignition components and fuel trims. - Evaluate for a driveline/vibration source separate from a misfire. CANDIDATE SYSTEMS (verify by testing) - Ignition/misfire — candidate, given P0300. - Fuel delivery — candidate. - Driveline/tire balance — candidate for a speed-linked shudder. Confirm everything against service data and your own tests.
The full workflow
- Enter the complaint, codes, and conditions — vehicle only, no owner info.
- Use the question list to interview the customer at intake.
- Work the checks in order, confirming each finding by actual test.
- Pull real specs, procedures, and bulletins from OEM service data — never from the AI.
- Record what you ruled out so the diagnosis is defensible.
Watch out for
This is a checklist to organize your thinking, not a diagnosis — the model does not know what is wrong with the car, and treating a 'candidate' as the answer skips the testing you are paid to do.
AI is a substitute for neither OEM service data nor real diagnosis: it invents torque specs, procedures, TSB numbers, and 'known issues' and states them confidently. Documented failures include telling an owner to change the fuel filter on an EV and claiming lifters could be replaced without pulling a cylinder head. Verify every spec and step against the factory manual.
Never enter customer identity or full VIN; the vehicle description is enough to build the checklist, and identifying data is regulated and unnecessary here.
Where this comes from
Every use case on this site is grounded in real reports from working auto mechanics — not invented by us.