AI Troubleshooting Prompts That Work

Who this is for

Anyone using AI Pilot to diagnose infrastructure problems and wanting better, faster answers from the AI.

What you will complete

Learn the prompt patterns that produce useful AI Pilot responses, and understand why vague prompts produce vague answers.


The core principle: specificity gets results

AI Pilot has access to your server metrics, alert history, deployment logs, and service statuses. But it cannot help you if your prompt does not tell it where to look. The more specific your prompt, the more focused and useful the AI's response.


Prompt patterns that work

Pattern 1: Name the target

Always include the server name, site name, or service name.

Instead of: "My site is slow"

Use: "My site named my-wordpress-site on server london-01 has been slow for the past hour. What is causing it?"

Instead of: "Something is wrong with my server"

Use: "Server db-prod-02 is showing high memory usage. What is consuming memory and should I be concerned?"


Pattern 2: Give a time range

Metrics and logs span days. Give the AI a time frame to focus on.

Instead of: "Why did my site go down?"

Use: "My site my-app went down at approximately 2 PM UTC today. What happened around that time and what caused it?"

Instead of: "CPU has been high lately"

Use: "CPU on server-name has been above 85% for the past 3 hours. What is consuming it?"


Pattern 3: State what you already tried

This prevents the AI from recommending steps you have already taken.

Instead of: "My backup failed"

Use: "My backup for server-name failed twice today. I already checked disk space — it has 40 GB free. What else could cause the failure?"


Pattern 4: Tell the AI what outcome you want

Do you want analysis only, or do you want the AI to fix it?

For analysis only: "Diagnose why nginx on server-name restarted 3 times in the past hour. Do not change anything."

For a fix: "Nginx on server-name is down. Please diagnose the cause and restart it."


Pattern 5: Ask for a comparison

Useful for identifying regressions.

"Compare server-name's current CPU and memory usage to its average over the past 7 days. Is this an anomaly?"


Pattern 6: Chain tasks explicitly

For multi-step operations, list the steps in order.

"First create a backup of server-name. Then, once the backup completes, restart the mysql service."


Prompts for common troubleshooting scenarios

Site returning 502 or 503 errors: "My site my-site is returning 502 errors. Please check the web server status, PHP-FPM status, and recent error logs on server-name."

Deployment is stuck: "Deployment for my-site has been in 'Running' state for 20 minutes. What is blocking it and how do we fix it?"

Database is slow: "My WordPress site my-site has been very slow since 10 AM UTC today. The site is on server-name. Is the database healthy? Is disk I/O a factor?"

Out of disk space warning: "Server-name is showing a disk space warning at 85% usage. What is consuming disk space and what can be safely removed?"

Service keeps crashing: "The worker service on server-name has crashed 4 times in the past 2 hours and keeps restarting. What is causing the crashes?"

SSL certificate expired: "SSL on my-site expired this morning. Please investigate and attempt renewal."

Server unreachable after a change: "Server-name became unreachable at 3 PM after I changed a firewall rule. The SSH connection is timing out. What should I check?"


Prompts to avoid

Too vague (AI cannot help):

  • "Something is wrong"
  • "Fix my server"
  • "Help"
  • "Why is it slow"

Overly broad (AI will need to ask clarifying questions):

  • "Check all my servers"
  • "Make everything faster"
  • "Look at my logs"

When the AI asks a clarifying question

If AI Pilot responds with a question before acting, answer it directly and in the same chat. For example:

AI: "Which server should I check? I see three servers in your organization." You: "Please check server-name."

The AI will then proceed with the correct target.


Related articles