AI, and Network Operations and Management · Topic 2 of 6

Prompts for Generative AI in Network Operations

Learn the parts of a good generative AI prompt for network work: persona, instructions, context, data classification and output format, and how to choose.

  • 6 min
In a nutshell
  • A prompt includes persona, instructions, context and the requested output format.
  • Persona sets the role and depth of the answer.
  • Instructions state the task, using clear verbs and limits.

What it is

A prompt is the text you give a generative AI system to tell it what to do. The quality of the answer depends heavily on the quality of the prompt.

A vague prompt such as "fix my network" gives a vague or wrong answer. A good prompt states who the AI should act as, what task it must do, what it needs to know, what it must not receive or reveal, and what shape the answer should take.

The exam asks you to select the best prompt from several choices. The winning prompt is the one that has the right components, asks for something specific and does not leak sensitive data.

How it works

A strong prompt is built from a few clear components. Learn the four the objective names, plus the context that ties them together.

Personawho the AI acts asInstructionsthe task and the rulesContext and datawhat it needs, cleaned firstOutput formatthe shape of the answer

The components

ComponentWhat it doesNetwork example
PersonaSets the role and level of the answer"Act as a CCNA level network engineer"
InstructionsStates the task, the steps and the limits"Explain the likely cause and list read only checks first"
ContextGives the facts the AI needs"Two routers, OSPF area 0, neighbor stuck in EXSTART"
Data classificationControls what sensitive data is included"Passwords, keys and customer data removed"
Output formatDefines how the answer is laid out"A table with columns Check, Command, Expected result"

Persona. The persona gives the answer a point of view and a depth. "Act as a network engineer who explains to a junior colleague" gets a different answer than "Act as a security auditor".

Instructions. Say exactly what to do. Use clear verbs: list, compare, explain, summarize, draft. Add limits, such as "do not suggest configuration changes, only show commands".

Context. Give the symptoms, the device type, what you already tried and what changed. Better context gives a more relevant answer.

Output format. Ask for the shape you can use: a numbered troubleshooting list, a table, a short summary for a ticket, or a block of CLI commands to review. A format request makes the answer easier to check.

Data classification

Data classification means knowing how sensitive your data is before you share it, and sending only what the tool is approved to receive.

A public AI tool is run by a third party outside your organization. Treat anything pasted into it as leaving your control. An approved internal tool, running under your organization's policy, may be allowed to receive more.

Do not paste confidential data such as passwords, enable secrets, pre shared keys, SNMP community strings, API keys or customer data into a public AI tool. Replace sensitive values with placeholders before sending.

DataSafe to paste into a public tool?What to do
Device passwords, keys, tokensNoRemove or replace with a placeholder
Customer names, addresses, account dataNoRemove or anonymize
Full running configurationUsually noSend only the relevant lines, cleaned
A generic question about how OSPF worksYesNo change needed
Sanitized show output with placeholdersDepends on policyFollow the organization's rules

Comparing prompts

Here are two prompts for the same problem.

Weak prompt:
My OSPF is broken, here is my whole config, fix it.
(whole configuration pasted, including enable secret and SNMP strings)

Strong prompt:
Act as a network engineer helping a junior colleague. Two Cisco routers are
directly connected on GigabitEthernet0/1 in OSPF area 0. The neighbor stays in
EXSTART. I have removed all passwords and keys from the data below. Using only
the output provided, list the three most likely causes in order, and for each
one give the read only show command that confirms it. Answer as a table with
the columns Cause, Command and What to look for. Do not suggest changes yet.

The weak prompt has no persona, no clear task and no format, and it exposes secrets. The strong prompt has all five components and asks for read only checks first.

Choosing the best prompt

Work through the options in this order.

Optionread the promptSafe datano secrets or customer data?Clear taskspecific instructions?Contextenough facts?Formatrequested output shape?
  1. Reject any option that includes passwords, keys or customer data.
  2. Prefer the option with a clear, specific task over a vague one.
  3. Prefer the option that supplies enough context to work from.
  4. Prefer the option that requests a usable output format.
  5. Prefer the option that asks for verification steps, because the output must be checked against real devices.
Exam tip

When two prompts look similar, the one that removes sensitive data and asks for a defined format usually wins.

Common mistakes

Pasting secrets. Passwords, keys and customer data do not belong in a public tool.

Being vague. "Why is the network slow" gives a generic answer. Name the device, the symptom and the scope.

Skipping the output format. Without one, the answer may be long and hard to use.

Dropping the persona. The role sets the depth of the answer, for example junior friendly or expert.

Trusting the answer without checking. Verify against device output before acting.

Watch out

Anything pasted into a public AI tool may leave your control. Remove confidential data first, and when in doubt, leave it out.

What to remember for the exam

  • A prompt includes persona, instructions, context and the requested output format.
  • Persona sets the role and depth of the answer.
  • Instructions state the task, using clear verbs and limits.
  • Data classification means protecting sensitive data before sending it.
  • Never paste passwords, keys or customer data into a public AI tool.
  • Name an output format such as a table, steps or a ticket summary.
  • The best prompt is specific, has enough context, removes sensitive data and asks for a usable format.
  • Check AI answers against real device output.

Check yourself

  • Which four prompt components does the objective name?

    Persona, instructions, data classification and output format, supported by context about the problem.

  • Why should you remove passwords and customer data before using a public AI tool?

    Anything pasted into a public tool may leave your organization's control, so confidential data must be removed or replaced with placeholders.

  • A prompt says "act as a network engineer, here is the sanitized output, list likely causes in a table and give a read only show command for each". What makes it strong?

    It has a persona, a specific task, supplied context, sanitized data and a defined output format, and it asks for read only checks.

Topic 2 of 6 in AI, and Network Operations and Management

Test yourself on AI, and Network Operations and Management

The full Cisco CCNA 200-301 v2.0 question bank has 1,242 questions, with timed practice exams and an explanation for every answer.

This study guide is independent and is not affiliated with, endorsed by, or sponsored by Cisco. Vendor names, certification names and exam codes are trademarks of their respective owners and are used for identification only.