CCAO-F sample questions with answers
10 questions from the Claude Certified Associate – Foundations practice bank, spread across its domains. Pick your answer, then open the explanation to see why each option is right or wrong.
- Question 1Output Evaluation and Validation
Before a budget meeting, a marketing coordinator asks Claude to explain the difference between gross margin and contribution margin and to give a one-line example of each. Which format should she expect, and why?
- A
A short inline answer, because it is a quick explanation she will read once and not reuse or share.
- B
An artifact, because financial definitions are reference material that should be saved as standalone documents she can return to.
- C
A CSV table with columns for term, definition and example, so she can import it into the budget spreadsheet.
- D
A slide, because the meeting is a presentation context and the definitions may be useful to show.
Show the answer and explanation
Answer: A
Match the format to how the output will be used. Quick, one-time explanations belong inline; artifacts are for content that is significant and will be reused, edited or shared; structured data is for tool consumption.
Why the other options are wrong
B. Artifacts are for substantial content that will be edited or reused; a two-definition answer does not meet that bar and clutters the workspace.
C. Structured data suits downstream tools; nothing here is going into a tool, and a CSV is harder for a person to read than a sentence.
D. The purpose is her own understanding before the meeting, not presenting; the format should follow the use, not the setting.
- A
- Question 2Workflow Integration and Solution Design
A project manager used Claude to turn workshop notes into a requirements document for a clinic scheduling tool. The document is polished, but the team notices acceptance criteria such as "the system should be fast" and "reminders should be sent promptly", and a section about "patient portal single sign-on" that nobody in the workshop mentioned. Regulatory sign-off requires testable criteria, and stakeholders review the document next week. What is the best next step?
- A
Ask Claude to make the document more detailed and comprehensive before the review.
- B
Send the document to stakeholders as it stands with a covering note asking them to read carefully and report anything that looks wrong, vague or unfamiliar before the regulatory review begins.
- C
Supply the workshop notes and real targets, ask Claude to rewrite vague criteria as measurable ones and to flag any requirement it cannot trace to the notes for stakeholder confirmation.
- D
Regenerate the document with a more capable model, since the current one hallucinated a requirement.
Show the answer and explanation
Answer: C
Two classic failures of AI-drafted requirements are unmeasurable criteria and content with no source. The remedy is to supply the real targets, ask for measurable rewrites, and require every requirement to trace back to an input, flagging the rest for humans to confirm.
Why the other options are wrong
A. More detail from the same inputs produces more invented specifics. The problem is unverifiable and unsourced content, not a lack of length.
B. This shifts detection to busy reviewers who will skim, and the vague criteria would still fail the regulatory requirement for testability.
D. Any model will fill gaps when the notes are silent. Traceability to sources is a process control, not something a model tier provides.
- A
- Question 3Governance, Risk, and Responsible Use
A communications team at a nutrition-supplements company is planning how to use Claude for public content. Which two of the proposed uses are inappropriate under Anthropic's Usage Policy and responsible-AI practice? (Choose two.)
Choose 2.
- A
Generate 20 customer testimonials with realistic names and cities and post them on the product page as genuine reviews.
- B
Draft first versions of press releases that the communications manager rewrites and the legal team approves before release.
- C
Publish a blog post containing Claude's statement that the product lowers cholesterol by 15%, without a qualified reviewer checking the claim.
- D
Sort 1,200 free-text survey comments into themes and spot-check a sample of the assignments by hand.
- E
Brainstorm 30 names for a new product line for the team to shortlist.
Show the answer and explanation
Answer: A and C
Drafting, brainstorming and analysis with verification are appropriate. Creating content that deceives people about who produced it or whether it is genuine, and publishing unverified health claims as fact, are inappropriate regardless of the business motive.
Why the other options are wrong
B. Drafting with human editing and approval is an ordinary, appropriate use.
D. Analysis with human verification of a sample is an appropriate use of Claude.
E. Brainstorming options for humans to select from is low risk and appropriate.
- A
- Question 4Prompting and Task Execution
An HR business partner pastes a job description for a customer-success manager into Claude and asks "Give me interview questions." She gets 30 questions of mixed quality, including several about technical skills the role does not need. She wants a structured set she can hand to a panel of four interviewers. Which rewritten prompt best fits?
- A
"Give me better interview questions than last time, and leave out the technical ones."
- B
"Act as an experienced HR interviewer. Using the job description below, write 8 behavioural questions, each mapped to one of the five competencies it lists. Return a table with columns: competency, question, what a strong answer includes."
- C
Remove the job description from the prompt so Claude is not distracted by the technical requirements, and ask for general management questions.
- D
Paste the job description again and add "Only ask about things in this document."
Show the answer and explanation
Answer: B
A strong business prompt combines a role, the source material, precise scope (how many, what kind, mapped to what) and the output format the downstream user needs. Restricting or repeating the source is not a substitute for saying what you actually want.
Why the other options are wrong
A. "Better" is not a specification. Claude still does not know how many questions, which competencies matter, or what format the panel needs.
C. The job description is the most valuable context in the prompt; removing it produces generic questions unrelated to this role.
D. Restricting Claude to the document helps a little, but the prompt still says nothing about quantity, structure, question type or how the panel will use the output.
- A
- Question 5Product and Model Selection
A consulting partner wants a simple interactive scenario calculator for a client, built from the assumptions in a 20-page engagement brief. Two consultants will refine it over three weeks in separate sessions, and every version must respect the brief's assumptions. Which combination of features fits best?
- A
A plain chat each time, pasting the brief and the previous version of the calculator at the start of every session.
- B
A Project holding the brief as knowledge with the key assumptions in its instructions, with the calculator built and iterated as an artifact in chats inside that Project.
- C
Research mode, so Claude can look up scenario-calculator best practices and generate the tool from them.
- D
Turning on memory so Claude remembers the brief, then building the calculator in ordinary chats.
Show the answer and explanation
Answer: B
Recurring, shared context belongs in a Project; a deliverable that is edited and versioned belongs in an artifact. Combining them gives both consultants the same assumptions and the same evolving tool across sessions. Research and memory solve different problems.
Why the other options are wrong
A. Re-pasting a 20-page brief in every session is error-prone, and each new chat has no link to the previous version unless it is pasted too.
C. Research gathers external information; the calculator's inputs are the client's own brief, which does not need web research and would not be persisted by it.
D. Memory captures preferences and working context, not a 20-page document with specific figures; it is not a reliable store for the brief's assumptions.
- A
- Question 6Configuration and Knowledge Management
A consultant is configuring a Project for a new client engagement with a logistics company. Which two configuration choices are correct? (Choose two.)
Choose 2.
- A
Upload the client's raw customer database export as knowledge so Claude has full context.
- B
Add the discovery notes, statement of work and org chart as knowledge files with descriptive names such as
sow-2026-03.pdf. - C
Write the instructions as "Be helpful, accurate and professional."
- D
Write instructions that define Claude's role as engagement analyst, the format for deliverables, and what to do when information is missing (ask rather than assume).
- E
Skip instructions entirely, since the knowledge files provide all the context Claude needs.
Show the answer and explanation
Answer: B and D
Configure a Project with the relevant reference material as clearly named knowledge files and concise instructions covering role, output format and how to handle gaps. Keep unnecessary personal data out and avoid vague adjectives.
Why the other options are wrong
A. A customer database contains personal data the engagement does not need in Claude; it breaches data-minimisation and likely the client's policy.
C. Generic adjectives give Claude nothing specific to follow; effective instructions name the role, outputs and constraints.
E. Knowledge tells Claude what the facts are, not how to behave; without instructions, tone, format and handling of gaps vary from chat to chat.
- A
- Question 7Troubleshooting and Optimization
A finance team uses Claude to draft a weekly performance report that goes to 30 clients. The workflow is: paste the data, generate the report, send. Twice this quarter a wrong percentage reached clients. The team wants to keep the time savings but stop the errors. Which change best fits?
- A
Add 'double-check every figure before answering' to the prompt.
- B
Go back to writing the reports by hand to eliminate the risk.
- C
Generate the report twice with slightly different prompts and send whichever version looks more polished and internally consistent.
- D
Add a review checkpoint before sending: a team member checks each figure against the source data using a short checklist.
Show the answer and explanation
Answer: D
Optimising a workflow is not only about speed. Placing a lightweight human review checkpoint at the point of highest consequence, with a checklist that focuses on the things that have gone wrong before, keeps the efficiency while making the output trustworthy.
Why the other options are wrong
A. A self-check instruction is a weak safeguard; the model can restate a wrong figure confidently, and nothing outside the model catches it.
B. This discards the workflow gains; a review checkpoint achieves the reliability at far lower cost.
C. Polish is not accuracy, and neither version has been checked against the source data.
- A
- Question 8Output Evaluation and Validation
A finance analyst uploads a spreadsheet of five years of regional sales and asks Claude to calculate the compound annual growth rate for each region. Claude reports 14.2% for the largest region, which will appear in a board deck. The analyst is fluent in spreadsheets and has 20 minutes. Which validation approach best fits?
- A
Ask Claude to explain its reasoning step by step and accept the figure if the logic sounds right.
- B
Re-run the request in a new chat and accept the figure if the two results match.
- C
Trust the figure because Claude used its data analysis tool, which executes real calculations on the uploaded file rather than estimating from the text.
- D
Ask Claude which cells and years it used, then recompute the growth rate in the spreadsheet with a formula and compare.
Show the answer and explanation
Answer: D
When the person can independently recompute a figure from the primary data, that is the strongest validation: it checks the arithmetic and the inputs. Re-running, reasoning walkthroughs and tool-based execution each leave a class of error unchecked.
Why the other options are wrong
A. A sound-looking explanation can accompany a wrong number; reasoning review helps find errors but is not an independent computation.
B. Agreement between runs can still reflect the same wrong input cells or the same period assumption; it is consistency, not correctness.
C. Code execution removes arithmetic error but not choice-of-input error (wrong column, wrong start year, currency mix); the calculation still needs an independent check.
- A
- Question 9Workflow Integration and Solution Design
A consultant scoping an engagement for a regional retailer that wants "AI for customer service" has uploaded two discovery-call transcripts and asked Claude to draft a requirements document. Which two practices make the resulting analysis reliable enough to take back to the client? (Choose two.)
Choose 2.
- A
Ask Claude to separate requirements stakeholders stated explicitly from ones it inferred, and to list open questions for the next client meeting.
- B
Treat everything in the document as confirmed, because it was generated from the client's own transcripts.
- C
Ask Claude to add standard retail customer-service requirements the client did not mention and present them as client needs.
- D
Ask Claude to stress-test the draft for requirements that conflict, cannot be measured, or depend on information nobody has yet supplied.
- E
Have Claude produce the final sign-off version so the client can approve it without another review cycle.
Show the answer and explanation
Answer: A and D
When Claude drafts requirements from raw inputs, make its reasoning visible: separate stated from inferred, gather open questions, and run a critique pass for conflicts and unmeasurable statements. Confirmation stays with the stakeholders.
Why the other options are wrong
B. A transcript is a source, not a validation. Claude fills gaps with plausible inferences, and only the client can confirm them.
C. Presenting invented requirements as the client's own misrepresents what was said. Suggestions can be offered, but they must be labelled as the consultant's proposals.
E. Requirements are a decision the client owns. Skipping their review to save a cycle is how misread needs become scope disputes.
- A
- Question 10Governance, Risk, and Responsible Use
A finance analyst at a bank uses Claude through the company's Enterprise workspace. She wants to analyse a transaction file classified Restricted. The bank's policy says Restricted data may be processed in the Enterprise workspace only with the data owner's approval and in minimised form. The analyst argues that because Enterprise data is not used to train models, she can upload the full file now. Which response is correct?
- A
She is right: the no-training default removes the risk, so the policy's approval step is unnecessary.
- B
The training default does not replace the policy control: she still needs the data owner's approval and should minimise the file first.
- C
She should use her personal Pro account with model-improvement turned off, which gives the same protection as the Enterprise workspace without the approval step.
- D
Restricted data can never be analysed with Claude, so she must do the analysis by hand.
Show the answer and explanation
Answer: B
Vendor commitments such as not training on commercial data reduce one risk but do not satisfy an organisation's own controls. Approval by the data owner, minimisation and use of the approved plan are all still required.
Why the other options are wrong
A. Training is only one risk; access, retention, exposure to colleagues and regulatory obligations remain, which is why the policy requires approval and minimisation.
C. A personal account is outside the bank's contract, admin controls and audit trail, and it is exactly what the policy forbids for Restricted data.
D. The policy explicitly allows Restricted data in the Enterprise workspace under conditions; the task is to meet them, not to abandon the tool.
- A
Practise all 153 CCAO-F questions
Start with the free 15-question diagnostic. It shows where to focus, and your results carry over if you sign up.