Free PSPO-AI Full-Length Practice Exam: 20 Questions

Try 20 free PSPO-AI questions across the exam domains, with answers and explanations, then continue in PM Mastery.

This free full-length PSPO-AI practice exam includes 20 original PM Mastery questions across the exam domains.

The questions are original PM Mastery practice questions aligned to the exam outline. They are not official exam questions and are not copied from any exam sponsor.

Official-count note: Scrum.org currently lists PSPO-AI Essentials as 20 multiple-choice questions in 30 minutes with an 85% passing score. Use Scrum.org for final eligibility and exam-day rules; use this page as an original full-length PM Mastery diagnostic.

Open the matching PM Mastery practice page for timed mocks, topic drills, progress tracking, explanations, and full practice.

For concept review before or after this set, use the PSPO-AI guide on PMExams.com.

How to run this diagnostic

Set a 30-minute timer and answer all 20 questions without pausing. For each miss, classify the issue as AI theory, security and ethics, or AI product ownership.

Suggested timing checkpoints:

Question rangeTarget elapsed time
1-710 minutes
8-1420 minutes
15-2030 minutes

Exam snapshot

ItemDetail
IssuerScrum.org
Exam routePSPO-AI
Official exam nameScrum.org Professional Scrum Product Owner - AI Essentials (PSPO-AI)
Full-length set on this page20 questions
Exam time30 minutes
Topic areas represented3

Full-length exam mix

TopicApproximate official weightQuestions used
AI Theory and Primer33%7
AI Security and Ethics33%6
AI Product Ownership34%7

Practice questions

Questions 1-20

Question 1

Topic: AI Security and Ethics

A Product Owner is considering releasing an AI feature that ranks job applicants for recruiters. The Developers share the following evaluation excerpt.

Exhibit: Model evaluation (excerpt)

Model: ResumeRank v0.9 (trained on 2018–2022 hiring decisions)
Top features: years_experience, school_tier, gap_months, zip_code
Observed "recommended" rate:
- Group A: 62%
- Group B: 38%
Note: zip_code strongly correlates with ethnicity in this region

What is the best next action supported by the exhibit?

  • A. Focus only on improving overall ranking accuracy, since higher accuracy will automatically reduce group differences
  • B. Release as planned because recruiters remain accountable for final decisions, so the model’s recommendations are not an ethics concern
  • C. Release as planned, but add a disclaimer that the model was trained on historical hiring decisions
  • D. Pause release and address potential unfair bias by reviewing features like zip_code, testing for disparate impact, and adjusting the solution before wider use

Best answer: D

What this tests: AI Security and Ethics

Explanation: The exhibit indicates a fairness concern: materially different recommendation rates for two groups and use of zip_code, a known proxy for ethnicity in that region. Responsible AI requires investigating and mitigating bias before scaling use, rather than relying on disclaimers or assuming humans-in-the-loop removes the risk. The Product Owner remains accountable for outcomes and appropriate safeguards.

Fairness in responsible AI is about avoiding unjustified, systematic differences in outcomes across groups, especially when features may act as proxies for protected characteristics. Here, the recommendation-rate gap (62% vs 38%) combined with the note that zip_code correlates with ethnicity is a clear signal to stop and validate the approach.

Appropriate next steps typically include:

  • Confirm whether the disparity is explainable by job-related factors
  • Evaluate bias/disparate impact with relevant stakeholders
  • Remove/limit proxy features, adjust the model, and re-test
  • Define acceptance criteria and monitoring before release

Transparency (disclosures) and human review are helpful, but they do not by themselves mitigate an identified fairness risk.

The exhibit shows a likely proxy for a protected attribute and a large outcome disparity, indicating a fairness risk that should be mitigated before release.


Question 2

Topic: AI Security and Ethics

A Product Owner observes that stakeholders are pasting generative AI outputs (for example, customer summaries) directly into the CRM, treating them as authoritative without checking for accuracy or accidental inclusion of personal data. Which term best describes this risky behavior?

  • A. Prompt injection
  • B. Automation bias
  • C. Hallucination
  • D. Retrieval-augmented generation (RAG)

Best answer: B

What this tests: AI Security and Ethics

Explanation: This situation describes people deferring to AI output as if it were correct and complete, then storing it as an official record. That over-reliance increases the chance that incorrect statements or unintended personal data end up in a system of record. The term for this tendency is automation bias.

Automation bias is a human tendency to place undue trust in automated or AI-generated outputs and to reduce critical thinking and verification. In product work, that shows up when people copy AI drafts into systems of record (like a CRM or knowledge base) without reviewing accuracy, sources, and whether sensitive information (such as PII) was included. Because AI can be wrong or include inappropriate content, the accountability remains with humans to validate and edit before information becomes “official.” A key mitigation is to treat AI output as a draft and require a review step before it is stored as a system-of-record entry.

It is the tendency to over-trust AI output and skip needed human verification before recording it as truth.


Question 3

Topic: AI Security and Ethics

A Product Owner wants to paste the following support ticket into a generative AI prompt to create a summary.

Ticket: “I can’t log in since yesterday.”
Name: Priya Shah
Email: priyashah@example.com
Device: Android 14
Error shown: “Invalid session”

Which field is an example of PII that should be removed or masked before prompting?

  • A. The customer’s device operating system version
  • B. The ticket’s short problem description
  • C. The error message text
  • D. The customer’s email address

Best answer: D

What this tests: AI Security and Ethics

Explanation: PII is information that can identify a person, either directly or indirectly. In the ticket, the email address is a direct identifier and should be redacted or masked before sending the text to an AI system, especially outside your organization. Keeping prompts free of unnecessary identifiers reduces privacy risk while still enabling useful summarization.

When using AI for summarization or analysis, the Product Owner remains accountable for privacy and confidentiality. PII includes direct identifiers (like names, email addresses, phone numbers, government IDs) and indirect identifiers when combined with other data. Safe prompt handling typically means minimizing data, removing or masking identifiers, and only sharing what is necessary for the task.

A practical approach is:

  • Copy only the context needed to achieve the outcome.
  • Redact/mask direct identifiers (and any credentials or secrets).
  • Prefer anonymized or synthetic examples when possible.

Device/OS details and generic error text can be helpful for understanding an issue, but they are not direct identifiers in this excerpt.

An email address can directly identify an individual and is PII, so it should be redacted or masked before prompting.


Question 4

Topic: AI Security and Ethics

A Product Owner for a recruiting platform is considering using AI to automatically reject some job applicants to reduce recruiter workload. This is a high-impact decision for individuals and could introduce unfair outcomes.

Which evidence/validation step best supports trusting the AI-informed decision before enabling auto-rejection?

  • A. Require the AI to generate a detailed rationale for each rejection
  • B. Backtest on labeled historical cases and compare errors and bias to baseline
  • C. Demo a few AI rejections to stakeholders and collect their opinions
  • D. Rely on the model’s confidence scores to approve auto-rejection

Best answer: B

What this tests: AI Security and Ethics

Explanation: Because auto-rejection impacts people’s opportunities, it requires stronger controls than subjective review or model self-reported signals. A backtest on representative, labeled cases provides objective evidence of real-world performance. Comparing results to the current baseline helps decide whether the change is acceptable and what additional safeguards are needed.

A risk-based approach means the higher the potential harm, the stronger the evidence and review you need before changing product behavior. For high-impact decisions, validate with objective, auditable evidence rather than relying on persuasive outputs.

A strong step is to evaluate the AI on representative, labeled historical cases (ground truth) and compare to the current process baseline (e.g., error rates and indicators of disparate impact across relevant groups). This produces measurable evidence you can review with appropriate SMEs and use to decide on safeguards (such as human-in-the-loop, limited rollout, or stopping criteria). Confidence scores, demos, and generated rationales can be helpful, but they do not validate real-world correctness or fairness.

High-impact decisions warrant stronger validation using representative ground-truth data and explicit comparison to current performance and fairness.


Question 5

Topic: AI Product Ownership

A Product Owner asks an AI assistant: “Act as a struggling low-income customer and list the top problems with personal finance apps.” Stakeholders want to use the response to shape the Product Backlog, but the PO worries the prompt may steer the AI into stereotypes and biased assumptions.

Which principle or practice best fits what the PO should do next?

  • A. Rely on the Product Owner to make the final decision
  • B. Disclose to stakeholders that AI helped generate the insights
  • C. Ensure no personally identifiable information is included in prompts
  • D. Validate the output with real user evidence before using it

Best answer: D

What this tests: AI Product Ownership

Explanation: Role/persona prompting can be useful for structure or tone, but demographic or emotionally loaded personas can amplify bias and stereotypes. The safest next step is to treat the AI output as a starting hypothesis and validate it against real user data and other trusted sources before it influences backlog decisions. This keeps learning evidence-based while still benefiting from AI-assisted ideation.

Role prompting is helpful when it sets a functional perspective (for example, “act as an experienced UX researcher” or “act as a critical reviewer”) to improve structure, completeness, or critique. Persona prompting becomes risky when it encodes demographic or value-laden assumptions that can steer outputs toward stereotypes.

In this scenario, the prompt is likely to bias the model’s framing of “top problems,” so the PO should use validation to mitigate harm:

  • Reframe prompts to avoid demographic stereotypes and ask for uncertainty/alternatives
  • Triangulate with interviews, analytics, support tickets, and domain experts
  • Only then translate insights into Product Backlog items or experiments

Transparency and accountability still matter, but they do not correct biased content; validation does.

Persona prompting can inject assumptions, so the AI’s claims should be treated as hypotheses and verified with unbiased evidence.


Question 6

Topic: AI Theory and Primer

A Product Owner uses a generative AI tool to draft user stories from stakeholder notes. The team notices that moving the instruction “Do not include any customer names or identifiers” from the end of the prompt to the beginning significantly reduces accidental leakage and produces more consistent results.

Which change best optimizes for quality and risk reduction while staying privacy-safe, and is most aligned with why models are sensitive to phrasing and information order in prompts?

  • A. Move constraints to the end and add “be creative” to avoid over-constraining the model’s first response
  • B. Front-load non-negotiable constraints and output structure, then provide delimited context using placeholders for any identifiers
  • C. Include the full notes with real names to improve accuracy, then instruct the model to remove identifiers at the end
  • D. Keep the prompt unchanged, but generate multiple drafts and manually choose the safest one

Best answer: B

What this tests: AI Theory and Primer

Explanation: Generative models produce output based on the entire prompt as context, and small wording or ordering changes can shift what the model attends to when predicting the next tokens. Putting hard constraints and formatting requirements first reduces ambiguity and makes them more likely to govern generation. Using placeholders preserves privacy while still giving the model the structure it needs to respond consistently.

Generative AI models generate text by predicting the next token from the prompt’s context, so phrasing and information order can change the probability distribution of what comes next. Instructions placed earlier can “set the frame” for the rest of generation, while constraints buried later may be underweighted, partially forgotten in long prompts, or conflicted by earlier cues. Ambiguous wording also creates multiple plausible interpretations, increasing variability.

To reduce this sensitivity in product work, prefer a repeatable prompt template that:

  • states non-negotiable constraints and priorities first
  • uses clear section delimiters (instructions vs. context)
  • specifies the required output format
  • replaces identifiers with placeholders to stay privacy-safe

This improves consistency without outsourcing accountability or increasing exposure of sensitive data.

Earlier, clearly structured instructions shape the model’s token-by-token generation more strongly and reduce the chance that later constraints are ignored or diluted.


Question 7

Topic: AI Security and Ethics

A Product Owner wants to use a general-purpose AI chat tool to summarize the last 1,000 customer support conversations and recommend changes to the onboarding flow. The conversations include names, emails, and account IDs.

The tool’s terms state that prompts and chat history may be retained in logs for service improvement unless you have an approved enterprise agreement.

What is the best evidence/validation step to support a trustworthy decision while addressing retention risk?

  • A. Share the AI summary with stakeholders for quick alignment
  • B. Paste full transcripts so the AI has maximum context for accuracy
  • C. Ask the AI to provide confidence scores for each recommendation
  • D. Use a de-identified export and confirm retention controls before analysis

Best answer: D

What this tests: AI Security and Ethics

Explanation: Retention risk is created when sensitive support data is entered into an AI system that may store prompts and chat history in logs. The strongest validation step is to verify how the tool retains data and to reduce exposure by de-identifying inputs. This supports a decision you can responsibly trust and explain.

When an AI tool may retain prompts/chat history, pasting raw conversations with PII creates a confidentiality risk that is independent of model quality. A safe usage pattern is to minimize sensitive data exposure (de-identify/redact or aggregate) and validate the data-handling controls (approved environment, retention disabled/contracted, and ability to delete or avoid storing chat history/logs). This is evidence you can rely on because it checks the underlying privacy assumptions that make the analysis permissible, before you evaluate the AI’s recommendations. Confidence scores or stakeholder alignment can be helpful later, but they do not validate whether sensitive data was inappropriately retained.

It validates both the input data (de-identification) and the tool’s retention behavior so sensitive content is not stored in chat history/logs.


Question 8

Topic: AI Product Ownership

A Product Owner is proposing a new product strategy to reduce onboarding drop-off for a B2B product. They want to use an AI assistant to help draft the key hypotheses and assumptions and decide what to validate first. Constraints: they have only one Sprint to get evidence, and they must not share customer PII outside approved systems. Which approach best optimizes learning and risk reduction while staying transparent and accountable?

  • A. Use AI to draft hypotheses and then finalize the roadmap
  • B. Build the AI-suggested feature first to validate in production
  • C. Give AI an anonymized evidence summary; draft hypotheses, assumptions, and tests
  • D. Paste raw support transcripts into an AI to draft hypotheses

Best answer: C

What this tests: AI Product Ownership

Explanation: The best option uses AI to rapidly draft clear hypotheses and underlying assumptions, and to propose lightweight validation experiments that can run within one Sprint. It also respects the privacy constraint by using anonymized inputs and keeps accountability with the Product Owner through review and transparent communication of AI’s role and uncertainty.

When using AI to support product strategy, the Product Owner can use it to accelerate discovery work (drafting hypotheses/assumptions and suggesting tests) but must keep human accountability and avoid privacy leakage. In this scenario, optimizing for learning within one Sprint means turning strategy statements into testable hypotheses, making assumptions explicit, and selecting the smallest experiments that reduce uncertainty fastest.

A good approach is:

  • Provide an anonymized, consented summary of relevant evidence and constraints.
  • Ask AI to draft hypotheses, assumptions, and measurable signals.
  • Ask AI to suggest Sprint-sized validation experiments (e.g., concierge test, prototype usability test, message test, small cohort A/B).
  • Review, adjust, and share the plan transparently with stakeholders, then create Product Backlog items for the tests and learning.

This reduces risk quickly without treating AI output as a decision or exposing PII.

It uses AI to structure hypotheses and validation experiments quickly while protecting privacy and keeping humans accountable for decisions.


Question 9

Topic: AI Theory and Primer

A Product Owner asks an AI assistant to “summarize our biggest renewal risks” and pastes full customer contracts plus an unreleased pricing roadmap into the chat. The summary looks useful, but InfoSec stops further use and stakeholders now refuse AI help because they believe confidential information may have been exposed.

What is the most likely underlying cause?

  • A. Unclear success criteria for what the summary should include
  • B. Automation bias toward accepting the AI summary as “truth”
  • C. Prompt injection caused the AI to leak internal documents
  • D. Failure to recognize and handle private/proprietary data requests

Best answer: D

What this tests: AI Theory and Primer

Explanation: The core issue is not the quality of the summary but the unsafe handling of sensitive information. Pasting customer contracts and an unreleased roadmap indicates the request involved private/proprietary data, requiring the Product Owner to constrain what is shared and use safer alternatives (e.g., redacted inputs or approved systems). The resulting mistrust stems from the perceived data exposure risk.

A Product Owner remains accountable for responsible AI use, including protecting confidential customer and company information. In this scenario, the decisive clue is that full contracts and an unreleased pricing roadmap were pasted into an AI chat, triggering InfoSec intervention and stakeholder mistrust. That points to a failure to recognize the request as involving private/proprietary data and to respond appropriately by minimizing data, removing identifiers, using synthetic examples, or using only approved, governed environments.

The key takeaway is that even “good” AI output is unacceptable if it was produced by sharing sensitive data inappropriately.

Sensitive, non-public information was shared without applying guardrails like redaction, minimization, or approved environments.


Question 10

Topic: AI Theory and Primer

A Product Owner uses an AI assistant to help prepare for Product Backlog refinement. The AI was given 200 anonymized support tickets from the last 30 days.

Exhibit: AI output (excerpt)

Themes: billing confusion (42%), slow search (31%), export failures (12%)
Suggested top items (ordered):
1) Redesign billing page
2) New search index
3) Fix exports
Confidence: High; no further validation needed

What is the best next action based on this exhibit?

  • A. Ask the AI to create the Sprint Goal and select items for Sprint Planning
  • B. Use the themes to draft candidate PBIs, then validate and order them yourself
  • C. Accept the suggested ordering and update the Product Backlog accordingly
  • D. Discard the output entirely and switch back to only manual ticket review

Best answer: B

What this tests: AI Theory and Primer

Explanation: The exhibit shows useful pattern-finding (themes and rough sizing) but also an unsafe claim that no validation is needed. In Scrum, the Product Owner is accountable for Product Backlog ordering and must make value decisions using evidence and stakeholder input. Use AI to augment synthesis and drafting, then validate and decide as a human-led activity.

AI is well-suited to augmenting knowledge work like clustering feedback, summarizing themes, and drafting candidate Product Backlog Items or questions for refinement. However, value judgments and accountability—such as ordering the Product Backlog, deciding what to pursue next, and communicating trade-offs—remain human-led responsibilities of the Product Owner in Scrum.

Because the AI explicitly claims “no further validation needed,” the safe interpretation is to treat the output as a hypothesis. Use it to speed up discovery and refinement preparation, then verify with evidence (e.g., ticket sampling, usage data, stakeholder/customer input) and collaborate with the Developers on feasibility and slicing before making ordering decisions.

AI can accelerate synthesis and drafting, but the Product Owner remains accountable for validation and Product Backlog ordering decisions.


Question 11

Topic: AI Product Ownership

A Product Owner uses a generative AI tool to draft a customer-facing “Quarterly Reliability Update” by pasting raw incident notes and support ticket excerpts into the prompt. The AI output reads well, so the Product Owner emails it to external partners. Later, Security and Legal point out it contains customer identifiers and internal system details, and stakeholders become reluctant to use AI going forward.

What is the most likely underlying cause?

  • A. The team skipped a security/privacy review to detect and remove sensitive data
  • B. The AI was compromised by prompt injection from external attackers
  • C. The prompt lacked enough product context, lowering the output quality
  • D. The Product Owner trusted the AI too much and shared it without human judgment

Best answer: A

What this tests: AI Product Ownership

Explanation: Because the content was shared externally and was derived from raw internal incident and ticket data, it required a security/privacy check (and redaction) before distribution. The resulting leak of customer identifiers and internal details is a review-and-handling failure, not a general quality problem. In responsible AI use, humans remain accountable for preventing disclosure of sensitive information.

AI-generated content should be treated like any other draft: it may unintentionally include sensitive information, especially when the prompt includes raw operational artifacts (incident notes, logs, support tickets). When the intended audience is external, the Product Owner should ensure the draft goes through the organization’s security/privacy review path (and appropriate redaction/approval) before sharing.

In this scenario, the harm is specific: customer identifiers and internal system details were exposed, triggering Security/Legal escalation and stakeholder mistrust. That pattern points to a weak review gate for confidentiality and privacy, rather than a problem of creativity, prioritization, or model capability. The key takeaway is to route externally shared AI-assisted outputs through the same (or stricter) data-handling checks as human-written content.

Externally sharing AI-generated text built from raw internal notes without review allows PII and confidential details to leak.


Question 12

Topic: AI Theory and Primer

A Product Owner shares anonymized product-usage event data with an AI system and asks: “Find 5 distinct customer types based on behavior patterns and describe each type.” The AI returns five segments with short descriptions and suggested onboarding focus areas. Stakeholders want to immediately redesign onboarding around these segments.

What is the best next step?

  • A. Create labeled data and measure prediction accuracy for each type
  • B. Validate the segments as a clustering output with domain experts
  • C. Ask the AI to output a churn probability for each user
  • D. Generate final onboarding copy for each segment and ship it

Best answer: B

What this tests: AI Theory and Primer

Explanation: The AI is grouping users into types without pre-defined labels, which is a clustering task. Before making Product Backlog decisions, the responsible next step is to validate that the segments are stable, representative, and useful with knowledgeable people, rather than treating the output as ready for implementation.

Classification assigns items to predefined labels, regression predicts a numeric value, clustering discovers groups in unlabeled data, and generation creates new content. Here, the request “find 5 distinct customer types based on behavior patterns” is clustering because the system is discovering segments rather than predicting a known label or number.

A responsible next step after receiving the clusters is to validate them before acting:

  • Sanity-check inputs/assumptions (sampling, missing data, bias)
  • Verify segments are interpretable and actionable with domain experts
  • Check stability/representativeness and whether decisions could harm users

Moving straight to implementation or switching to a different task type skips the needed validation and can lead to low-value or harmful product decisions.

The AI performed clustering (grouping unlabeled data), so the next step is to validate whether the segments are meaningful and safe to act on.


Question 13

Topic: AI Product Ownership

You are the Product Owner for a product that helps HR teams screen job applicants. In 48 hours, you need an interview guide for discovery interviews with recruiters to understand pain points and validate assumptions about fairness.

Constraints:

  • You must avoid leading questions and reduce bias in wording.
  • You cannot include any real candidate or recruiter data in AI prompts.
  • You remain accountable for what questions are asked.

What is the BEST next action?

  • A. Use AI to conduct the recruiter interviews directly to save time
  • B. Give AI de-identified context and ask for neutral, open-ended questions plus bias/leading checks, then review and refine
  • C. Prompt AI with your hypotheses and ask it to write questions that confirm them
  • D. Paste prior interview notes verbatim into AI to tailor the best recruiter questions

Best answer: B

What this tests: AI Product Ownership

Explanation: The Product Owner can use AI to accelerate drafting an interview guide, but must explicitly request neutral, open-ended questions and have the AI flag potentially leading or biased phrasing. Keeping prompts de-identified protects confidentiality. Final responsibility stays with the Product Owner through review and refinement before using the questions with recruiters.

When using AI to generate discovery prompts, the key is to prevent the model from amplifying your assumptions into leading questions. Provide only de-identified context and instruct the AI to produce neutral questions that explore multiple possibilities, including disconfirming evidence.

Practical approach:

  • Ask for open-ended questions (avoid yes/no and prescriptive wording).
  • Request probes that test alternative explanations and edge cases.
  • Ask the AI to label questions that sound leading, loaded, or biased and suggest neutral rewrites.
  • Review and edit the guide yourself (and optionally with stakeholders) before interviews.

This meets the timebox while protecting confidentiality and maintaining human accountability for the discovery outcome.

It uses AI to draft and critique non-leading questions without sharing sensitive data, while keeping the Product Owner accountable through human review.


Question 14

Topic: AI Product Ownership

A Product Owner used an AI assistant to synthesize interview notes and support tickets into “top 5 discovery insights” to share with the Scrum Team and key stakeholders. The output includes confident-sounding statements (for example, percentages and “users want X”), but it does not show which sources support each claim or what was inferred.

What should the Product Owner verify or ask for FIRST to communicate the insights and uncertainty transparently?

  • A. For each insight, request the supporting sources and any confidence/limitations
  • B. Ask marketing to draft messaging for the most exciting insight
  • C. Ask stakeholders which solution they prefer to fund this quarter
  • D. Ask the Developers to estimate the effort for the top insights

Best answer: A

What this tests: AI Product Ownership

Explanation: Transparent communication of discovery insights requires separating evidence from interpretation and clearly stating limitations. By first obtaining traceability to the underlying sources and an explicit view of confidence and gaps, the Product Owner can explain what is known, what is uncertain, and what follow-up discovery is needed before decisions are made.

When AI is used to synthesize discovery, the Product Owner remains accountable for the message shared with the Scrum Team and stakeholders. Before presenting “insights,” first establish what the AI output is based on and where it may be unreliable. That enables you to communicate uncertainty explicitly (sample coverage, missing segments, inferred vs. observed claims, and any weakly supported quantitative statements) and to propose next discovery steps.

A practical first check is:

  • What sources support each insight (traceability)
  • What was inferred vs. directly observed
  • The confidence level and known limitations (coverage, recency, bias)

Without that, confident wording can create false certainty and drive premature backlog decisions.

You cannot be transparent about uncertainty without knowing what evidence exists, what is inferred, and the key limitations behind each insight.


Question 15

Topic: AI Security and Ethics

You are the Product Owner for a new customer help center. A stakeholder proposes using generative AI to “rephrase” sections of online articles so you can publish them as your own help content, keeping the same meaning but changing the wording. You have not reviewed the articles yet.

What is the most important question to ask/verify FIRST?

  • A. Which AI approach will minimize text-similarity scores versus the originals?
  • B. What are the sources’ licenses/permissions and required attribution for reuse or derivative works?
  • C. What success metric will we use to measure improved readability?
  • D. What tone and reading level should the rephrased content target?

Best answer: B

What this tests: AI Security and Ethics

Explanation: Rephrasing does not eliminate copyright risk if the underlying content is reused without permission. Before using AI to paraphrase, the Product Owner should clarify whether the source content can legally be reused, whether derivative works are allowed, and what attribution is required. This keeps accountability with the team and avoids treating “different wording” as a compliance strategy.

The core issue is that AI “paraphrasing” can still produce an unauthorized derivative work or plagiarism-like outcome if you publish content based on copyrighted material without the appropriate rights. The first clarification should therefore be about the source material’s permission model: where the text comes from, what license or terms apply, whether commercial reuse/derivatives are allowed, and any attribution requirements. Only after those constraints are clear should you define quality goals (tone, audience, readability) and decide on a workflow to validate outputs (e.g., human review, citations, and records of sources). Optimizing for low similarity scores is not a reliable way to avoid infringement because meaning and protected expression may still be substantially similar.

Key takeaway: confirm reuse/derivative rights and attribution obligations before generating rephrased content.

You must confirm you have the right to create and publish derivative content before attempting AI-based rephrasing.


Question 16

Topic: AI Product Ownership

A Product Owner encourages the Scrum Team to use a shared “prompt library” in the company wiki to speed up discovery and Product Backlog refinement. The library has no named owner, prompts are not reviewed, and there is no deprecation/expiry process.

In the last Sprint, the team reused an old prompt that referenced a retired customer segment and outdated pricing, and used the AI output to draft and order Product Backlog items.

What is the most likely near-term impact?

  • A. Increased vendor lock-in because prompts cannot be migrated
  • B. Inability to meet the Product Goal because AI replaces accountability
  • C. Regulatory penalties because any prompt library is non-compliant
  • D. Lower product decision quality and rework due to stale prompts

Best answer: D

What this tests: AI Product Ownership

Explanation: A prompt library without clear ownership, review, and deprecation is likely to accumulate outdated context. Reusing stale prompts can quickly produce misleading summaries and recommendations, which reduces product decision quality. In the near term this shows up as misordered Product Backlog items, loss of trust in the workflow, and avoidable rework.

Maintaining a prompt library or product knowledge base is a product asset that needs governance: an owner, review cadence, versioning/traceability, and a clear deprecation/expiry mechanism. Without these, prompts and embedded assumptions (segments, metrics, policy, pricing, terminology) drift out of date while still looking “approved,” so teams reuse them and get confidently wrong outputs. In the scenario, the immediate consequence is poor transparency about prompt freshness and reduced decision quality, which quickly turns into rework when the team discovers the backlog items were based on outdated context. The key takeaway is that lightweight ownership and deprecation protect value delivery by keeping reusable prompts current and trustworthy.

Without ownership, review, and deprecation, outdated prompts drive incorrect outputs that quickly lead to misordered backlog items and rework.


Question 17

Topic: AI Product Ownership

A Product Owner wants to use a generative AI tool to summarize themes from the last 3 months of support tickets and app-store reviews to improve Product Backlog ordering. The raw tickets include customer names, emails, and occasional payment references. No analysis has been started yet.

What is the best next step?

  • A. Ask AI to propose backlog items and estimate value
  • B. Upload raw tickets and ask for top themes
  • C. Run the AI summary first, then validate with Developers
  • D. Classify data and create a de-identified, minimal dataset

Best answer: D

What this tests: AI Product Ownership

Explanation: Before analyzing support tickets with AI, the Product Owner should clarify the objective and protect customer privacy by minimizing data and removing or masking PII. This enables useful theme analysis while reducing exposure of sensitive information and keeping accountability with the Scrum Team. Only after the dataset is safe should prompting and validation of outputs occur.

Responsible AI use in product discovery starts with defining the purpose and ensuring the data you provide is appropriate for that purpose. With support tickets, raw text often contains PII and sensitive details, so the next step is to classify the data and prepare a de-identified, minimal dataset (or aggregated excerpts) that still supports theme discovery. Then you can draft prompts against that sanitized input and validate AI outputs through spot checks, triangulation with other evidence (e.g., tagged ticket categories), and review with the Developers/stakeholders as needed. The Product Owner remains accountable for decisions; AI can assist analysis but should not be given unnecessary sensitive data or allowed to “decide” backlog ordering without human evaluation.

You must first minimize and de-identify sensitive customer data before using AI for analysis.


Question 18

Topic: AI Theory and Primer

A stakeholder asks the Product Owner to paste an exported customer list (names, emails, contract values) and a draft, unreleased pricing proposal into a free public AI chat tool to “quickly generate a segmentation and launch messaging.” The organization’s guidance says only publicly shareable information may be entered into external AI tools.

What is the best response?

  • A. Decline; use only de-identified, approved-share data instead
  • B. Paste only the customer list; AI needs real records
  • C. Paste everything; remove names and emails afterward
  • D. Paste only the pricing proposal; it has no personal data

Best answer: A

What this tests: AI Theory and Primer

Explanation: The request asks to share private customer data and proprietary pricing strategy in a public external AI tool, which violates the stated guidance. The Product Owner should refuse to provide that data and propose a safer alternative, such as using de-identified information and only data approved for external sharing.

A key part of using AI responsibly is recognizing when a request involves private or proprietary information and preventing inappropriate disclosure. In this scenario, the customer export contains personal data (and likely confidential commercial terms), and the draft pricing proposal is proprietary business information. Because the guidance limits external AI inputs to publicly shareable information, the appropriate response is to not paste the data into the public tool and to instead use approved, minimized inputs (for example, aggregated or de-identified data, synthetic examples, or an organization-approved workflow) while keeping human accountability for the decision.

The request includes private and proprietary data, so the PO should not enter it into an external AI tool and should switch to de-identified/approved inputs.


Question 19

Topic: AI Theory and Primer

A stakeholder tells you: “Add AI to our support portal so it can recommend the best next step for each customer.” The request is intentionally high level, and you need to decide whether this is an AI need or simply automation/rules.

What is the best question to ask FIRST?

  • A. Which recommendations require learning from data versus fixed business rules?
  • B. Which competitor’s AI product should we copy?
  • C. Can Developers start a model prototype this Sprint to see what’s possible?
  • D. Can we assume all customer conversations can be used for training data?

Best answer: A

What this tests: AI Theory and Primer

Explanation: Start by clarifying the nature of the capability requested: is it deterministic and expressible as explicit rules, or does it require pattern recognition and adaptation from data. AI is appropriate when the solution must handle variability and uncertainty rather than follow predefined if/then logic. This question directly distinguishes AI from automation and rules-based systems before discussing implementation details.

AI refers to systems that perform tasks requiring human-like judgment, such as recognizing patterns, making predictions, or generating recommendations, often under uncertainty. In contrast, automation and rules-based systems execute predefined logic (e.g., fixed decision trees or if/then rules) and behave deterministically for a given input.

In this scenario, the fastest way to determine whether “add AI” is warranted is to ask for concrete examples of the desired recommendations and assess whether:

  • Stable business rules can produce acceptable results, or
  • The portal must infer the “best” action from data because the situation varies and cannot be fully specified upfront.

Only after that distinction is clear should you move into questions about data, metrics, and constraints.

This clarifies whether the problem needs adaptive, probabilistic behavior (AI) or deterministic rule-based automation.


Question 20

Topic: AI Theory and Primer

A Product Owner is exploring improvements to a product’s onboarding experience. They have interview notes, support tickets, and funnel analytics, and want to use generative AI to speed up discovery.

Which action should the Product Owner AVOID?

  • A. Ask AI to draft experiment hypotheses and success measures, then refine and approve them
  • B. Ask AI to brainstorm onboarding ideas, then select options using human judgment
  • C. Ask AI to choose the best solution and commit it to the Product Backlog as a decision
  • D. Ask AI to cluster feedback into themes, then validate themes with stakeholders

Best answer: C

What this tests: AI Theory and Primer

Explanation: Using AI for ideation, summarization, and drafting can accelerate discovery when the Product Owner remains accountable for decisions. The boundary is that AI may propose options, but it should not be treated as the decision-maker for what to build. Decisions require human judgment, evidence, and transparent rationale.

A practical boundary is: use AI to generate or organize possibilities (ideas, themes, hypotheses, draft text), and use humans to make product decisions (ordering the Product Backlog, selecting outcomes, committing to a plan). In the scenario, clustering feedback and drafting experiments are ideation/support tasks that still require validation and refinement. Treating an AI recommendation as “the decision” (and committing it directly) weakens product ownership because it bypasses critical evaluation of evidence, trade-offs, and stakeholder alignment, while also obscuring who is accountable.

Key takeaway: AI can accelerate thinking, but the Product Owner must own and be able to explain decisions.

This outsources product decision-making and accountability to AI instead of using AI as an ideation aid with human validation.

How to interpret your result

  • 90% or higher: you are likely handling AI product-ownership trade-offs well if the questions were unseen.
  • 85-89%: review every miss because the official threshold leaves little room for weak ethics, evidence, or backlog decisions.
  • 75-84%: you likely need more practice turning AI output into accountable product decisions.
  • Below 75%: rebuild the AI theory, security, and product-owner use-case topics before another timed set.

Repeated strong unseen scores are a sign to take the real assessment, not to keep memorizing the same practice items. The skill you want is product judgment under new AI scenarios.

What PM Mastery adds after this diagnostic

This page gives one complete public PSPO-AI diagnostic. PM Mastery adds the larger PSPO-AI bank, topic drills, timed mocks, progress tracking, and explanations for evidence, backlog, ethics, privacy, and AI-assisted product decisions.

Retake protocol

Before retaking, rewrite each miss as a product-owner rule: protect sensitive data, validate AI claims, keep the Product Owner accountable, make assumptions visible, and use AI to support evidence rather than replace it.

Continue with full practice

Use the PSPO-AI Practice Test page for the full PM Mastery route, mixed-topic practice, timed mock exams, explanations, and web/mobile app access.

Open the matching PM Mastery practice page for timed mocks, topic drills, progress tracking, explanations, and full practice.

Focused topic pages

Free review resource

Read the PSPO-AI guide on PMExams.com for concept review, then return here for PM Mastery practice.

Revised on Thursday, May 14, 2026