Exam Objective 5.2: Select a prompt to send to a generative AI system to support network operations considering prompt components such as data classification, output format, persona, and instructions.
Building on objective 5.1's conceptual overview of agentic AI, this objective gets practical: how do you actually write an effective prompt to get useful, accurate help from a generative AI system for a real network operations task? A poorly constructed prompt produces a vague, generic, or even unsafe response; a well-constructed prompt — one that clearly defines the persona, provides the right data, gives specific instructions, and specifies the desired output format — produces a focused, actionable, and appropriately scoped answer. This objective also folds in a security-conscious element: data classification, meaning you must think carefully about what information is actually safe to include in a prompt sent to an AI system in the first place.
The exam objective explicitly calls out four components to consider when constructing a prompt: data classification, output format, persona, and instructions. Together, these four elements transform a vague request into a precise, well-scoped prompt likely to produce a genuinely useful response.
A representative example prompt illustrating all four components:
"Act as a senior network engineer (persona). Analyze this OSPF debug output (data). List any neighbor adjacency issues (instructions) in a numbered list (output format)."
Persona means instructing the AI system to respond from a specific role or perspective, shaping the tone, depth, and assumed expertise level of its response.
Why persona matters: Asking an AI system to "act as a senior network engineer" versus "act as someone explaining networking to a complete beginner" produces meaningfully different responses — different vocabulary, different assumed baseline knowledge, different level of technical depth. Specifying a persona helps ensure the response is pitched at the right level and from the right professional viewpoint for the actual audience or use case.
Examples of useful personas in a network operations context:
"Act as a senior network engineer reviewing this configuration for errors"
"Act as a CCNA-level network technician explaining this concept to a junior colleague"
"Act as a security-focused network architect evaluating this ACL for potential gaps"
"Act as a technical writer creating end-user-facing documentation from this troubleshooting session"
Exam Alert: Persona is about who the AI should respond as, which shapes tone, assumed expertise, and terminology — recognize a persona instruction in a sample prompt as specifically the piece that establishes the responding "voice" or "role."
Instructions define the specific task the AI system should actually perform — the core action you want it to take with the provided data.
Why instructions must be specific: A vague instruction like "look at this and tell me what you think" produces an unfocused response. A specific instruction like "identify any neighbor adjacency issues" or "list the three most likely root causes for this connectivity failure, in order of probability" gives the AI system a clear, narrow task, producing a far more directly useful response.
Examples of clear, specific instructions in a network operations context:
"Identify any interface error counters that indicate a physical-layer problem"
"Summarize this syslog output, grouping entries by severity level"
"Compare these two routing table outputs and identify any differences"
"Explain why this OSPF neighbor is stuck in the 2-Way state rather than reaching Full"
Exam Alert: Instructions are the action/task itself — the verb-driven core of the request. A prompt with a well-defined persona and good data, but a vague or missing instruction, still fails to produce a genuinely useful response, since the AI system doesn't actually know what specific task to perform with the information provided.
Output format specifies exactly how the AI system's response should be structured, making the result immediately usable rather than requiring the requester to manually reformat it afterward.
Why output format matters for network operations specifically: Network engineers frequently need AI-assisted output in a very particular structure — a numbered list of issues found, a table comparing before/after configuration states, a specific CLI command block, a structured incident summary matching an internal template. Specifying the desired format upfront saves rework and ensures the response fits directly into whatever downstream process (a ticket, a report, a runbook) it's meant to support.
Examples of output format instructions:
"Respond in a numbered list"
"Present the answer as a table comparing the before and after configuration"
"Provide only the corrected configuration commands, with no additional explanation"
"Summarize in three sentences or fewer"
"Format the response as a structured incident report with sections for Summary, Root Cause, and Remediation Steps"
Exam Alert: Output format is about the structure/presentation of the response, distinct from instructions (which define the task itself) — a prompt can have perfectly clear instructions but still produce an unusable result if the output format doesn't match what the requester actually needs to do with the answer next.
Data classification, in the context of this objective, refers to the practice of evaluating the sensitivity of information before including it in a prompt sent to a generative AI system — recognizing that not all network data is safe or appropriate to paste into an external AI tool.
Why this matters specifically for network operations: Network configuration files, log output, and troubleshooting data can easily contain sensitive information — internal IP addressing schemes, hostnames revealing internal naming conventions, SNMP community strings, pre-shared keys, passwords (even hashed ones), customer-identifying information, or details about security posture (which ports/services are open, what security tooling is in place) that could be valuable to an attacker if exposed. Before pasting real device output into a prompt, an engineer needs to consider whether that specific data is appropriate to send to the AI system being used, particularly for external, cloud-hosted, third-party AI tools where the organization has limited control over how submitted data is stored, logged, or potentially used to further train the underlying model.
Practical data classification considerations before constructing a prompt:
Is this data public, internal-only, confidential, or highly sensitive/regulated, according to the organization's own data classification policy?
Does the AI tool being used have an appropriate data handling agreement in place (an enterprise AI deployment with contractual data protections, versus a free public consumer AI tool with no such guarantees)?
Can the prompt be constructed using sanitized or redacted data — replacing real IP addresses, hostnames, and credentials with generic placeholders — while still preserving enough structure for the AI system to provide a useful, accurate answer?
Is there an internal or enterprise-approved AI tool that should be used instead of a general public-facing consumer AI service for this specific type of data?
Exam Alert: Data classification is the security-conscious component of prompt construction — recognize that this component is fundamentally about deciding what information is safe/appropriate to actually include in the prompt, not about formatting or instructing the AI at all. Expect a scenario question where a prompt example includes real, sensitive data (actual passwords, real public IP addressing, actual customer information) and you're asked to identify that this represents a data classification problem, distinct from a persona, instruction, or output-format issue.
Combining all four components into a single, effective prompt for a network operations task:
Example — troubleshooting scenario:
"Act as a senior network engineer (persona). Here is sanitized output from show ip ospf neighbor showing a neighbor stuck in the 2-Way state, with hostnames and IP addresses replaced with generic placeholders (data, with data classification already applied). Identify the most likely reasons this neighbor has not reached the Full state (instructions). Present your answer as a numbered list of possible causes, ordered from most to least likely (output format)."
Example — documentation scenario:
"Act as a technical writer creating internal documentation (persona). Here is a summary of the troubleshooting steps taken to resolve last night's WAN outage, with all customer-identifying information removed (data, with data classification already applied). Write a clear, chronological incident summary suitable for a post-incident review (instructions). Format the response with headings for Summary, Timeline, Root Cause, and Remediation (output format)."
Exam Alert: Expect to be shown a complete sample prompt and asked to identify which part of the prompt corresponds to each of the four named components (persona, data/data classification, instructions, output format) — this direct-mapping skill is very likely to be tested.
Vague or missing instructions — asking the AI system to "look at this" without specifying what specific task to perform, producing an unfocused, generic response
No specified output format — receiving a response in an unpredictable structure that then requires manual reformatting before it can actually be used
No persona specified for a task where the appropriate expertise level/voice genuinely matters — resulting in a response pitched at an inconsistent or inappropriate level of technical depth
Including sensitive or improperly classified data directly in the prompt without first considering whether that data should be sanitized, redacted, or whether an alternative, appropriately vetted AI tool should be used instead
Overloading a single prompt with too many unrelated instructions at once, producing a response that only partially addresses each individual request rather than thoroughly addressing one clear task
Exam Alert: A scenario prompt missing one of the four components, or improperly including sensitive data, is a strong candidate for a "what's wrong with this prompt" style question — practice recognizing which specific component is absent or mishandled.
Determine the actual goal of the request — what specific outcome do you need (a diagnosis, a summary, a draft document, a comparison)?
Decide on an appropriate persona if the expertise level or professional voice of the response matters for how it will be used.
Gather the relevant data, and before including it in the prompt, apply data classification thinking — confirm the data is appropriate to share with the specific AI tool being used, sanitizing or redacting sensitive details (real IP addressing, credentials, customer information) where necessary.
Write clear, specific instructions defining exactly what task the AI system should perform with that data.
Specify the desired output format so the response arrives in a structure that's immediately usable for its intended purpose.
After receiving the response, apply the evaluation skill covered in objective 5.1 — critically review the output against the actual evidence before relying on or acting on it.
Q1. In the prompt "Act as a network security analyst. Review this firewall rule set for overly permissive entries. Respond with a table listing each risky rule and its recommended fix," which part represents the "instructions" component?
A. "Act as a network security analyst"
B. "Review this firewall rule set for overly permissive entries"
C. "Respond with a table listing each risky rule and its recommended fix"
D. The firewall rule set itself
Answer: B. Instructions define the specific task the AI system should perform. "Review this firewall rule set for overly permissive entries" is the clear, specific action being requested — distinct from the persona ("Act as a network security analyst"), the output format ("respond with a table..."), and the data (the firewall rule set itself).
Q2. A network engineer is about to paste raw configuration output — including actual pre-shared keys and real public IP addresses — into a prompt for a free, public, consumer-facing AI chatbot to help troubleshoot a VPN issue. What consideration from this objective is most directly relevant here?
A. Persona
B. Output format
C. Data classification
D. Instructions
Answer: C. Data classification is specifically about evaluating the sensitivity of information before including it in a prompt sent to an AI system. Pasting real pre-shared keys and public IP addressing into a public, consumer-facing AI tool without first considering data sensitivity and appropriate handling is exactly the kind of scenario this component addresses — the engineer should consider sanitizing/redacting the data or using an appropriately vetted enterprise AI tool instead.
Q3. Which prompt component specifically controls whether an AI system's response is delivered as a numbered list, a table, a short paragraph, or another specific structure?
A. Persona
B. Data classification
C. Output format
D. Instructions
Answer: C. Output format specifies exactly how the response should be structured and presented — a numbered list, a table, a specific template, a word/sentence limit — ensuring the response is immediately usable for its intended purpose without requiring manual reformatting afterward.
Q4. Why does specifying a persona (such as "act as a senior network engineer") typically improve the usefulness of an AI-generated response in a network operations context?
A. It guarantees the response will be factually accurate
B. It shapes the tone, assumed expertise level, and terminology of the response to better match the intended audience or use case
C. It replaces the need for providing any actual data
D. It has no meaningful effect on the response
Answer: B. Specifying a persona establishes the role or viewpoint the AI system should respond from, which shapes the tone, the assumed baseline technical knowledge, and the vocabulary used in the response — helping ensure the answer is pitched appropriately for the intended audience or use case, though it does not by itself guarantee factual accuracy.
Q5. A network engineer sends the following prompt: "Here's some log output. What do you think?" What is the most significant weakness in this prompt, based on the components covered in this objective?
A. It includes too much data
B. It specifies an inappropriate output format
C. It lacks a clear persona, specific instructions, and a defined output format, making the request vague and likely to produce an unfocused response
D. It violates data classification best practices
Answer: C. This prompt fails to specify a persona, gives only a vague instruction ("what do you think?" rather than a specific task), and does not define any output format — all three missing components make it highly likely to produce a generic, unfocused response rather than a precise, actionable answer.
Q6. Before including real customer network data in a prompt sent to an AI system, which of the following questions is most directly related to the data classification component of prompt construction?
A. Should the response be formatted as a table or a list?
B. Should the AI system respond as a beginner-level assistant or a senior engineer?
C. Is this specific data appropriate to share with this particular AI tool, given its sensitivity and the tool's data handling practices?
D. What specific task should the AI system perform with this data?
Answer: C. Data classification is specifically about assessing the sensitivity of information and whether it's appropriate to include in a prompt sent to a given AI system, considering that system's data handling and storage practices — a distinct concern from output format (A), persona (B), or instructions (D), each of which addresses a different aspect of prompt construction.