Introduction
The best way to understand the Claude Certified Associate – Foundations (CCAO-F) exam is to see its questions. This post gives you free sample questions with full explanations, drawn across the seven domains. They are written in the exam's judgement-led style, where the task is usually to pick the most responsible or effective option rather than to recall a fact.
Work through each one before reading the answer. Notice how often the tempting, fast option is the wrong one.
These are original practice questions written by CCA Prep. They are not real exam questions. CCA Prep is an independent practice platform and is not affiliated with, authorised by, or endorsed by Anthropic. Claude and Anthropic are trademarks of Anthropic PBC.
Domain 1: Prompting and Task Execution
Question 1. You need Claude to rewrite a 40-page policy document into a one-page summary for senior managers. Which prompt is most likely to produce a usable first draft?
- A. "Summarise this."
- B. "Summarise this document."
- C. "You are writing for busy senior managers. Summarise the attached policy in under 250 words, as five bullet points covering scope, key changes, risks, cost, and the decision required."
- D. "Make this shorter and more professional."
Answer: C. The strong prompt gives a role (who it is for), a task, a format (five bullets under 250 words), and the specific content each bullet should carry. A, B, and D leave almost every decision to the model, so the output is a lottery. Specifying audience, length, and structure is the core skill this domain tests. Practise more prompting questions for free.
Domain 2: Output Evaluation and Validation
Question 2. Claude produces a market summary that includes three confident statistics with no sources. You need the summary for a client report due tomorrow. What should you do first?
- A. Use the statistics as written, since Claude is usually accurate.
- B. Delete the statistics entirely to be safe.
- C. Verify each statistic against a primary source before including it, and cite the source.
- D. Ask Claude whether the statistics are correct and trust its answer.
Answer: C. Confident, unsourced figures are exactly where hallucination risk lives. You verify against a primary source and cite it. A over-trusts the output. D is worse, because asking a model to confirm its own claim does not make the claim true. B throws away potentially useful content instead of validating it. This verify-then-use pattern is the most heavily tested idea on the exam.
Question 3. Which of the following are good reasons to keep a human in the loop before an AI-drafted output is sent externally? (Select all that apply.)
- A. The output makes factual claims that carry reputational or legal risk.
- B. The output will be attributed to a named person or the organisation.
- C. The task is high-volume and low-stakes, such as internal tagging.
- D. The output touches regulated advice or confidential data.
Answer: A, B, and D. Human review matters most where errors are costly: factual or legal risk, external attribution, and regulated or confidential content. C describes the opposite case, high-volume low-stakes work, where heavy human review adds little value. Multiple-response questions like this are usually scored all-or-nothing, so identify every correct option. Practise output evaluation.
Domain 3: Product and Model Selection
Question 4. A colleague wants to process a 300-page contract in a single request and is frustrated that earlier pages seem to be ignored. What is the most likely explanation to check first?
- A. The model is broken and should be reported.
- B. The document may exceed or strain the available context window, so relevant sections should be provided or the task split up.
- C. Contracts cannot be processed by AI.
- D. A faster model tier would fix it.
Answer: B. Context windows are finite, and very long inputs can crowd out earlier material. The practical fix is to supply the relevant sections or break the task into parts, not to assume the tool is broken or to switch tiers for speed. Understanding context limits is a core selection skill. Practise product and model selection.
Domain 4: Workflow Integration and Solution Design
Question 5. Your team wants to automate responses to inbound support emails. Which design is the most responsible starting point?
- A. Auto-send AI-drafted replies to all customers immediately.
- B. Draft replies with AI, route them to an agent for review and approval, and measure quality before increasing automation.
- C. Ban AI from customer communication entirely.
- D. Let AI reply to complaints but not to simple questions.
Answer: B. Good workflow design introduces AI with a human checkpoint and a way to measure impact, then scales as confidence grows. A removes the checkpoint on customer-facing risk. C forgoes the benefit without cause. D inverts the risk, putting AI on the most sensitive messages. Designing hand-offs and escalation is the point of this domain. Practise workflow integration.
Domain 5: Configuration and Knowledge Management
Question 6. Several team members get inconsistent tone and formatting when they use Claude for the same recurring report. What is the best fix?
- A. Ask everyone to write better prompts individually.
- B. Set up a shared Project with custom instructions and a saved prompt or template so everyone starts from the same configured context.
- C. Have one person do all the reports.
- D. Accept the inconsistency as unavoidable.
Answer: B. Consistency at team scale comes from shared configuration: a Project, custom instructions, and reusable templates, so everyone works from the same context. Relying on individual prompting (A) reproduces the inconsistency. C does not scale, and D gives up. Practise configuration and knowledge management.
Domain 6: Governance, Risk, and Responsible Use
Question 7. A colleague wants to paste a spreadsheet of named customers and their contact details into a public AI tool to draft outreach emails. What is the most appropriate response?
- A. Go ahead, it saves time.
- B. Check your organisation's data-handling policy first, and avoid entering personal or confidential data into tools not approved for it, using anonymised or synthetic data where possible.
- C. Paste the data but delete the chat afterwards.
- D. Only paste half the records.
Answer: B. Governance questions reward the privacy-aware, policy-first answer. Personal data should not go into unapproved tools; check the policy and use anonymised or synthetic data instead. Deleting the chat (C) or reducing the volume (D) does not address the underlying confidentiality problem. When a governance scenario is close, favour the option that protects data and follows policy. Practise governance and responsible use.
Domain 7: Troubleshooting and Optimization
Question 8. Claude keeps returning answers that are close but consistently miss a specific requirement you care about. What is the most effective next step?
- A. Regenerate repeatedly until one output happens to be right.
- B. Make the missing requirement explicit in the prompt, add an example of a correct output, and state it as a must-have.
- C. Switch to a different tool.
- D. Accept the output and fix it by hand every time.
Answer: B. When outputs are close but repeatedly miss the same point, the prompt is under-specifying that requirement. Naming it explicitly and showing an example fixes the failure mode at its source. Regenerating (A) is a lottery, switching tools (C) is premature, and manual fixes (D) do not solve the recurring cause. Practise troubleshooting and optimization.
How to Use These Questions
Notice the recurring pattern across domains: the correct answer usually verifies, discloses, protects data, or keeps a human in the loop, while the trap is the fast, unchecked option. Learning to recognise that pattern reliably is most of what passing the CCAO-F requires.
These eight are a small sample. Our full question bank covers all seven domains with a detailed explanation on every question. Start with the free Prompting domain, no card required, then continue from the CCAO-F practice hub. For a full study plan, see How to Pass the Claude Certified Associate (CCAO-F).