How would you prompt an LLM to provide reliable technical support?
Instruction: Explain your approach, give a practical example, and describe its limits.
Updated
Example Answer
I would supply the relevant product version and verified troubleshooting material, then ask the assistant to identify the problem, ask for missing details, and suggest the smallest supported next step. It should cite the source for a technical instruction and say when the evidence is insufficient.
I would not rely on the prompt to prevent unsafe commands or unauthorized changes. The application must restrict tool access and check permission before an action runs. For a destructive or disruptive step, I would explain the effect and require the appropriate approval. I would test the workflow with outdated documentation, ambiguous symptoms, and instructions embedded in a support ticket.
Concrete example
If a ticket only says “the service is slow,” ask about the affected operation, timing, and version before suggesting a configuration change. Never ask the customer to paste credentials into the chat.
Follow-up to practice
What would you do if a retrieved support article conflicts with the installed product version?