CertifyAIxAce every AI certification.

CCAO-F sample questions with answers

10 free practice questions for the Claude Certified Associate: Foundations exam. Try each one, then open the answer to see why the right option wins and every other option loses.

Question 1Prompting and Task Execution

A team is adding worked examples to a prompt so that outputs follow the house style. (Select TWO.)

Choose 2.

  1. A.

    Use a single example, so that there is no risk of the examples contradicting one another in any respect.

  2. B.

    Vary the examples so they cover edge cases as well as ordinary ones.

  3. C.

    Make the examples mirror the real cases the prompt will be used on.

  4. D.

    Choose examples that are deliberately unlike the real work, so that the general principle is what gets learned.

  5. E.

    Leave the examples unmarked in the body of the prompt so they read continuously with the surrounding instructions.

Show answer

Answer: B, C

Anthropic asks examples to be relevant and diverse; both properties are needed for them to steer reliably.

  • A. One example makes everything incidental in it look intentional; three to five is advised.
  • B. Diversity stops an incidental feature of one case being copied as a rule.
  • C. Relevance is a documented requirement: examples should mirror the actual use case.
  • D. Unlike examples steer the output towards the wrong target entirely.
  • E. Unmarked examples can be read as further instructions, which tagging prevents.
Question 2Prompting and Task Execution

A user tells Claude that a draft "is not quite right" and asks for another attempt, and the second draft has the same qualities as the first.

  1. A.

    Start a new conversation and give the original prompt again, so the unwanted draft does not influence the next attempt.

  2. B.

    Ask for a third attempt, since repeated attempts explore different approaches and one of them will eventually be right.

  3. C.

    Say specifically what is wrong and what the output should do instead, rather than that it is not right, so the next attempt is a correction rather than another sample.

  4. D.

    Ask Claude to explain why it wrote the draft as it did.

Show answer

Answer: C

"Not quite right" carries no information; name the defect and the target.

  • A. A fresh conversation with the same prompt reproduces the original conditions.
  • B. More attempts without new information keep sampling the same space.
  • C. Naming the defect and the target makes the next attempt a correction rather than a repeat.
  • D. An explanation of the draft produces rationalisation rather than improvement.
Question 3Prompting and Task Execution

A user has spent an hour refining a prompt for a one-off task, and the current output would take about ten minutes to finish editing by hand into exactly what is needed.

  1. A.

    Refine the prompt for a further hour and then decide whether hand editing is still necessary at that point.

  2. B.

    Continue refining the prompt until it produces the output without editing, so the task is fully automated.

  3. C.

    Edit the current output by hand and stop refining, since the task is one-off and the remaining work is small.

  4. D.

    Abandon the output produced so far and write the piece by hand, since the refinement has not yet succeeded.

Show answer

Answer: C

Refinement pays off on repeated tasks; for a one-off, finish it by hand.

  • A. Another hour before deciding commits the cost before making the decision.
  • B. Full automation of a task that runs once spends more to save less.
  • C. A one-off task should be finished, not automated; ten minutes beats further refinement.
  • D. Discarding a nearly finished output throws away the work already done.
Question 4Prompting and Task Execution

An open research question is handed over with no indication of what a good answer would contain, and each attempt explores a different aspect of the subject.

  1. A.

    Ask for a list of the aspects that could be explored and then choose one of them once that list has arrived and been reviewed.

  2. B.

    Ask the same question repeatedly and combine the aspects that the various attempts happened to explore.

  3. C.

    Narrow the subject until it is small enough that only one aspect of it remains to be explored.

  4. D.

    State the success criteria for the research up front: what the answer must settle, and what would count as having settled it.

Show answer

Answer: D

Anthropic's first rule for research prompts is to provide clear success criteria for the question.

  • A. Choosing from a menu is itself defining success, so this only postpones the step.
  • B. Pooling runs collects partial explorations aimed at an undefined target.
  • C. Narrowing the subject discards the question rather than defining success for it.
  • D. Defining what a successful answer contains is the documented first step for research.
Question 5Prompting and Task Execution

A prompt gives a good answer to one wording of a question and a noticeably different answer when the same question is phrased another way.

  1. A.

    Accept the variation, since any two phrasings of a request are genuinely two different requests.

  2. B.

    Instruct users to submit the request using only the exact wording that the prompt was originally developed and tested with.

  3. C.

    Treat consistency as a criterion in its own right and test the prompt on paraphrases of the same request.

  4. D.

    Run each request several times and publish whichever of the answers appears most frequently.

Show answer

Answer: C

Consistency is a named success criterion, tested by comparing answers to paraphrased inputs.

  • A. Abandoning the criterion is not the same as meeting it.
  • B. Mandating one exact wording pushes the fragility onto users and will not hold.
  • C. Consistency is a documented criterion measured across paraphrased inputs.
  • D. A majority vote masks the instability and multiplies the work per request.
Question 6Prompting and Task Execution

A user pastes a long customer email into a prompt together with instructions about what to do with it, and Claude sometimes follows sentences from inside the email as though they were instructions from the user.

  1. A.

    Separate the instructions from the pasted material with clear delimiters, and say which part is the document and which is the instruction, so the boundary does not depend on the wording of either.

  2. B.

    Shorten the customer email before pasting it, so that fewer sentences within it can be mistaken for instructions to follow.

  3. C.

    Place the instructions after the email rather than before it, so that the most recent text in the prompt is the instruction.

  4. D.

    Ask Claude to confirm what it intends to do before it does it, so that any misreading is caught in the reply.

Show answer

Answer: A

Mark the boundary between instruction and material explicitly with delimiters.

  • A. Delimiters plus a statement of which part is the document removes the ambiguity structurally.
  • B. Shortening the email reduces the opportunity without addressing the cause.
  • C. Reordering helps a little while leaving the boundary itself unmarked.
  • D. A confirmation step catches some errors after the fact rather than preventing them.
Question 7Prompting and Task Execution

A team writes down "the tone should be good" as the standard a prompt will be judged against.

  1. A.

    It is acceptable, provided the person who wrote it is the only person who ever applies it.

  2. B.

    It is too ambitious, because tone is a quality no prompt should be expected to influence at all.

  3. C.

    It is too vague to act on: a criterion should be specific and measurable, such as a stated scale applied consistently.

  4. D.

    It is acceptable, because tone is an inherently subjective quality and therefore cannot usefully be expressed any more precisely than this in writing.

Show answer

Answer: C

Success criteria must be specific and measurable; subjective qualities can still be given a scale.

  • A. A criterion only one person can apply is not a shared standard at all.
  • B. Tone and style are listed as an ordinary success criterion, not an unreachable one.
  • C. Specific and measurable are documented properties; a consistent scale supplies both.
  • D. Subjective qualities are scored on defined scales in the documented guidance.
Question 8Prompting and Task Execution

A prompt contains three worked examples and the author suspects the set is too narrow but is unsure how to widen it.

  1. A.

    Ask Claude to assess the examples for relevance and diversity and to produce further ones based on the existing set.

  2. B.

    Delete one of the three examples and see whether the output gets any worse as a result.

  3. C.

    Duplicate the three existing examples so that the prompt contains six in total.

  4. D.

    Leave the examples alone and add a sentence saying that they are illustrative rather than exhaustive.

Show answer

Answer: A

Anthropic suggests asking Claude to evaluate your examples for relevance and diversity, or to generate more.

  • A. Documented suggestion: have Claude judge example relevance and diversity, or extend the set.
  • B. Removing one example tests quantity when the concern is coverage.
  • C. Duplicates raise the count while reinforcing the same narrow pattern.
  • D. A label aimed at human readers does not change the pattern the examples carry.
Question 9Prompting and Task Execution

A long policy document is attached and the answer must stay tied to what the document actually says.

  1. A.

    Ask for the shortest possible answer, so there is less room in it for anything unsupported to appear.

  2. B.

    Ask Claude to find and set out the relevant passages first, then answer on the basis of those passages.

  3. C.

    Ask for the answer first and then ask separately whether the document supports each part of it.

  4. D.

    Ask for the answer to be given with a confidence rating attached to each of the statements it contains.

Show answer

Answer: B

Anthropic advises grounding long-document answers by asking for the relevant quotes before the task.

  • A. Fewer statements are not better-supported statements; brevity is not grounding.
  • B. Quoting first is the documented way to ground an answer in a long document.
  • C. Checking afterwards inspects an answer that was produced without the constraint.
  • D. A confidence rating reports how sure the answer sounds, not what text supports it.
Question 10Prompting and Task Execution

Each stage of a staged calculation feeds the next, and the team wants errors caught at the stage that produced them.

  1. A.

    Ask for every stage to show its arithmetic so that a reviewer can repeat all of it at the end of the run.

  2. B.

    End each stage by asking for its own output to be checked against the conditions that stage was given.

  3. C.

    Ask for each stage to be produced twice and compare the two versions before moving on.

  4. D.

    Check only the final figure, since an error at any stage will change it and so will be visible there.

Show answer

Answer: B

Asking Claude to verify its own output against stated criteria catches errors at the stage that made them.

  • A. Deferring all review to the end is the delay the team set out to remove.
  • B. A per-stage check against its own criteria catches the error where it was made.
  • C. Two matching runs demonstrate stability, not correctness.
  • D. A wrong final figure shows that something failed without saying which stage.

Keep going with 560 more CCAO-F questions

Free papers every day, in the real exam formats, with progress by exam domain. Unlock every paper and timed mock exam when you are ready.

Practise freeSee all CCAO-F papers
CCAO-F sample questions with answers (10 free) · CertifyAIx