AI Prompt Design Practice Questions

Answers and explanations appear when you work through the set above.

1. An engineer wants help analysing a router configuration using a public generative AI service. Which action best respects data classification?

   • Paste the full running-config, including type 7 passwords and SNMP community strings
   • Remove or mask secrets, public IP addresses and customer identifiers, and share only the lines needed for the question
   • Paste the config and ask the AI to keep it confidential
   • Send the config as an attachment so it counts as private

2. Which part of a prompt is the persona?

   • 'Return the answer as a table with columns Device, Interface, and Issue.'
   • 'You are a senior network engineer who explains findings to a help desk audience.'
   • 'Do not include any passwords in your response.'
   • 'Here is the output of show interfaces GigabitEthernet0/1.'

3. A team needs AI output that a script can parse automatically. Which prompt element helps most?

   • A persona of 'friendly assistant'
   • An output format instruction such as 'Respond only with valid JSON using the keys device, severity, and action'
   • A longer description of the company history
   • A request to be as creative as possible

4. Which prompt is best for asking a generative AI system to help diagnose an interface problem?

   • 'Fix my network.'
   • 'Act as a CCNA-level engineer. Using the sanitised show interfaces output below, list the three most likely causes of the input errors and the show commands that would confirm each. Answer as a numbered list.'
   • 'Why is it slow? Be brief.'
   • 'Write everything you know about Ethernet.'

5. Which element of a prompt tells the model what task to perform and what constraints apply?

   • Instructions
   • Persona
   • Data classification label
   • Temperature setting

6. A generative AI assistant returns a configuration snippet for an ACL. What should the engineer do before applying it to a production router?

   • Apply it immediately, because the prompt included a persona
   • Review and test it, since generative AI output can contain wrong syntax or unintended matches
   • Ask the AI to confirm it is correct and apply it if it says yes
   • Apply it only to the interface with the most traffic to test it under load

7. Which prompt part is most directly responsible for the model producing a bulleted list rather than a paragraph?

   • Output format
   • Persona
   • Data classification
   • Context data

8. Company policy classifies full device configurations as confidential and allows them only in an approved internal AI tool. An engineer needs help with an OSPF problem using a public AI service. Which prompt respects the policy?

   • The full running-config, prefixed with the label 'CONFIDENTIAL' so that the model treats it carefully
   • Only the router ospf section and show ip ospf neighbor output, with addresses and hostnames replaced by placeholders
   • The full running-config, compressed into a zip archive so that the text is not directly readable
   • A screenshot of the running-config instead of text, so that no configuration data is transmitted

9. Which item should be removed before pasting show running-config output into any external generative AI system, even a sanitised prompt?

   • The interface descriptions that name each uplink
   • The router ospf network statements
   • The enable secret hash, SNMP community strings and pre-shared keys
   • The IOS version line at the top of the output