As of 13 August 2026, AI can only partly investigate the root cause of product defects.

What to actually do

1. Hand it to a person

   The route this page recommends

   A person who owns the outcome does this end to end, worth it when the failure is dear.

2. Use a tool built for this

   Second choice

3. Do it yourself

   The distant third

   A chat interface, chat-fluent skill, and roughly 2 hours until you can act on the result.

   How to actually do it

   1. Open the defect log, non-conformance reports, batch records, inspection results, work instructions and maintenance records for the affected product.
   2. Remove customer names and other unnecessary personal data, then combine the relevant records into a dated table with defect type, batch, product variant, measurement, machine, material, supplier, operator, shift and process step.
   3. Paste the process description, acceptance criteria and evidence into the prompt, clearly labelling anything that is missing or uncertain.
   4. Ask the model to produce the problem statement, timeline, defect pattern analysis, possible causes and ranked hypotheses without treating correlation as proof.
   5. Take the proposed checks to the responsible quality colleague and agree which inspections, samples or controlled tests are safe and authorised.
   6. Run the agreed checks against unaffected and affected examples where appropriate, record the actual results, and paste those results back into the conversation for an updated cause assessment.
   7. Compare the final proposed containment and corrective actions with your quality procedure, approve them through your normal responsibility chain, and record the evidence supporting the selected root cause.

   Prompt

   Act as a quality investigation assistant, not the final decision-maker. Investigate the possible root causes of the product defect described below using only the evidence supplied. Do not invent facts, probabilities, test results, standards or causes. Separate confirmed facts, observations, assumptions and unknowns. Produce: 1) a concise problem statement; 2) a timeline of relevant events; 3) a defect pattern analysis by product, batch, date, machine, material, supplier, operator, shift and process step where the data supports it; 4) a fishbone-style list of possible causes; 5) a ranked hypothesis table with the evidence for, evidence against, missing evidence and confidence stated as low, medium or high rather than as a percentage; 6) the smallest safe checks or controlled tests that could distinguish the leading hypotheses, including what result would support or reject each one; 7) immediate containment actions that do not assume the root cause; 8) proposed corrective and preventive actions; and 9) a short report suitable for a quality manager. Flag any action that could affect safety, compliance, customer shipments or product release for approval by the responsible quality professional. Ask focused questions for missing information before drawing conclusions.
   
   Product and defect: [DESCRIPTION]
   Business process and relevant work instructions: [PROCESS DETAILS]
   Defect records and measurements: [RECORDS]
   Affected batches, dates and quantities: [BATCH DETAILS]
   Machine, material, supplier, operator and shift information: [AVAILABLE DETAILS]
   Recent changes, incidents or maintenance: [CHANGES]
   Checks or tests already completed: [EXISTING EVIDENCE]
   Applicable internal acceptance criteria or standards: [CRITERIA]
   Constraints on inspections and testing: [CONSTRAINTS]

   Open it prefilled in ChatGPT or Claude, or copy it into Gemini, which takes no prefill link.