Free Anthropic Claude Certified Associate – Foundations (CCAO-F) Sample Questions

Correct answer: A — Provide the instructions as a numbered list of steps in the exact order they must run

• A. Correct. Providing instructions as sequential steps using numbered lists or bullet points is recommended specifically when the order or completeness of the steps matters, removing ambiguity about sequence.
• B. A single comma-separated sentence obscures step boundaries and order even more than a plain paragraph would, making the sequence less clear rather than more.
• C. Describing steps out of order and relying on inference is the opposite of making the required sequence unambiguous.
• D. Relying only on paragraph breaks without numbering removes the explicit ordering cue that numbered lists provide for steps where sequence matters.

Correct answer: C — Guide Claude to create a marketing email: first, write an attention-grabbing subject line; then, open with a hook; next, present three key benefits with persuasive language; and finally, include a clear call to action linked to the product page.

• A. Incorrect. This approach provides no guidance on audience, tone, or key benefits, leaving the output too open-ended. Without clear direction, the email may lack persuasive elements and fail to meet marketing goals.
• B. Incorrect. This instruction results in a formal internal report, not a marketing email. It emphasizes technical specifications rather than attention-grabbing copy, benefits, and a call to action.
• C. Correct. This prompt gives step-by-step guidance tailored to marketing email creation, ensuring inclusion of a subject line, hook, benefits, and call to action. It's a structured approach that yields a persuasive, complete email.
• D. Incorrect. While producing multiple drafts can be useful for exploration, it is less efficient for a single drafting task. Option C provides a more direct and structured method to get a well-crafted email in one go.

Correct answer: D — With no record of what the document must satisfy, each round can undo an earlier fix and nobody notices.

• A. Time spent is a real cost, but the team would accept it if the rounds were converging. The problem is that they are not.
• B. An accumulating request is an untidy side effect, and it does not by itself cause earlier fixes to be lost.
• C. Prose quality drifting under repeated revision is worth watching, but it is a milder issue than requirements silently reversing.
• D. Correct. Without a written list of everything the document must satisfy, each round is checked only against the latest complaint. Requirements settled earlier are not re-tested, so the work circles instead of converging.

Correct answers: A, E — It bundles several distinct deliverables that are meant for different audiences and uses.; Later parts are only worth producing if the earlier factual work is correct.

• A. Correct. Multiple deliverables with different readers pull the output in different directions, and one response has to compromise. Splitting lets each part be specified properly.
• B. Unfamiliarity affects how carefully the analyst must review the result, but it says nothing about whether the request itself is structurally overloaded.
• C. A long prompt is not the same as an overloaded one. Pasting rich source material is good practice and often makes a single-step request work better.
• D. Output format is a presentation choice. A table can be produced in one step just as easily as prose, so it is not a reason to break the work up.
• E. Correct. Dependency is a separate signal from bundling: when step two builds on step one, a single pass propagates any early error into everything downstream unchecked.

Correct answer: A — Rewrite the article with shorter sentences and everyday words, keeping every step accurate

• A. Correct. Simplifying sentence structure and vocabulary while preserving the accurate steps directly targets the readability problem without losing correctness.
• B. Incorrect. Removing steps changes the article's completeness and could omit cases some customers need, which does not fix a reading-level problem.
• C. Incorrect. Adding more technical detail moves the article further from, not closer to, the audience's reading level.
• D. Incorrect. A formal specification format is written for engineers and would make the article harder for typical customers to follow.

Correct answer: A — As accurate but off target, because it explains the topic instead of addressing the rise she asked about.

• A. Correct. Everything present may be correct while the question remains unanswered. Accuracy and relevance are separate tests, and this output passes the first and fails the second.
• B. Accepting background as progress leaves the actual question open, and the material she now has does not move her any closer to a cause.
• C. A general answer indicates that the request was read broadly, not that the specific question is unanswerable given her own data.
• D. Shortening reduces the volume of material that does not address her question, leaving a briefer answer to the wrong question.

Correct answer: A — Open the report and find that exact sentence, since a close paraphrase may not appear in it.

• A. Correct. A quotation asserts that specific words appear in a specific document, so only locating those words in the report validates it.
• B. Incorrect. A page number can be generated as readily as the quotation, so it adds no independent confirmation.
• C. Incorrect. Similar wording elsewhere does not establish that this report contains this sentence.
• D. Incorrect. The report existing says nothing about whether the quoted sentence is in it.

Correct answer: B — Agree which suppliers and which criteria belong in the comparison before asking for any output.

• A. Incorrect. Combining unreconciled answers produces a long document that still contains whatever gaps and conflicts the originals had.
• B. Correct. A comparison is only as good as its frame. Fixing the rows and columns first is what stops the result being shaped by whatever happened to come up in earlier chats.
• C. Incorrect. Format is a downstream decision; choosing it before the content is defined locks in a shape that may not fit.
• D. Incorrect. A ranking produced before the criteria are agreed reflects unstated assumptions she cannot inspect or defend.

Correct answer: B — Open the cited sources and confirm that each figure in the draft does appear in the page it points to.

• A. A disclosure line may well be required, but it does nothing about accuracy. Adding it and publishing would put unverified figures in front of an external audience.
• B. Correct. Where web search was used, the help centre's recommendation is to review Claude's cited sources, because reading the original content reveals anything that was misread or taken out of context. With citations present, checking them is faster and stronger than any other available check.
• C. A self-reported confidence rating is not evidence. It can sound reassuring about a figure that the cited page does not actually support.
• D. Two runs can repeat the same error, so agreement between them is not confirmation. The sources are already cited and can be checked directly.

Correct answer: D — Use the thumbs-down control on the response that contained the incorrect statement, in addition to fixing her own draft.

• A. A note in the chat helps her own record of what happened, but it stays inside her conversation. It does nothing to report the response as problematic beyond her own session.
• B. A fresh conversation may or may not produce a better answer, and starting over is a workaround for her immediate task rather than a way of recording that the response was wrong.
• C. Emphatic rewording is a weak remedy and addresses only her future prompts. It neither reports the faulty response nor makes the problem visible to anyone else.
• D. Correct. The thumbs-down feedback control on the response is the mechanism provided for flagging an incorrect or misleading answer, and it complements, rather than replaces, correcting the draft she is working on.

Correct answer: A — Let the conversation continue and condense itself, since nothing in it needs to outlive this session.

• A. Correct. When the output is disposable, allowing automatic context management to make room is the lowest-effort choice; there is no durable knowledge worth persisting.
• B. Incorrect. A project is for context that recurs. Setting one up for work that will never be revisited is unnecessary overhead.
• C. Incorrect. Restarting discards the ideas generated so far and forces her to regenerate them at further cost.
• D. Incorrect. Fragmenting the session means no chat holds the full set of ideas, which undermines the brainstorm itself.

Correct answers: C, D — Use Research mode, so the briefing is built by multi-step searching and returns citations.; Ask for the briefing as an artifact, so the team can edit and reuse it outside the chat.

• A. Incorrect. Re-pasting an inline answer for every edit is precisely the friction artifacts exist to remove, and versions drift apart quickly.
• B. Incorrect. An internal mailbox is not the source of public competitor information, and connecting one does not answer the sourcing need here.
• C. Correct. Research mode runs an agentic, multi-step search and returns citations, which is exactly what a sourced competitor briefing needs.
• D. Correct. An artifact is the durable, editable deliverable format, so weeks of later editing happen on one object rather than in chat replies.
• E. Incorrect. Splitting by competitor fragments the briefing and does nothing about sourcing or about the editing the team has ahead of it.

Correct answers: B, C — Use the deepest-reasoning tier for the contract dispute, accepting the larger quota it uses.; Use the fastest, cheapest tier for the high-volume classification, where throughput matters.

• A. Incorrect. Tier choice changes depth, speed and quota use, so treating it as incidental is how both of these jobs go wrong.
• B. Correct. A tangled one-off analysis is the case where deeper reasoning is worth the meaningfully larger quota it consumes.
• C. Correct. High-volume, low-complexity classification is exactly where the fastest and cheapest tier earns its place.
• D. Incorrect. The tiers differ in reasoning depth as well as speed, so the lightest tier is a poor fit for a tangled dispute.
• E. Incorrect. Using the deepest tier for routine classification spends quota on work that does not need the extra reasoning depth.

Correct answers: A, D — The deepest tier suits genuinely hard problems, and it consumes meaningfully more allowance.; The mid tier is the fast, capable default, and it is the right pick for most everyday work.

• A. Correct. The deepest tier buys reasoning depth for harder problems and costs meaningfully more of the usage allowance.
• B. Incorrect. The difference is reasoning depth, speed and cost, not writing style, and the tiers are not interchangeable.
• C. Incorrect. The lightest tier is the fastest of the three, and it trades some reasoning depth for that speed and low cost.
• D. Correct. The middle tier is positioned as the fast, capable default that handles most knowledge work well.
• E. Incorrect. A vague answer usually signals an under-specified prompt, and a deeper tier cannot supply context that was never given.

Correct answer: D — Write down what good output means, then run a real sample on the default tier, then compare that sample against the definition, then change tier only where it falls short.

• A. Choosing on instinct and defining quality a month later means there is no standard to judge the intervening results against, and a month of work has already been committed to an unexamined choice.
• B. Committing several tiers to the task before any standard exists means the trial is designed without knowing what it is testing for, and it duplicates the work and the allowance draw on every run.
• C. Comparing results before the task has been run is out of order, and benchmarking against manual work is not the same as agreeing what this task's output has to achieve.
• D. Correct. Define the quality bar, test the default tier on real work, check that sample against the bar, and change tier only where the evidence shows the default falls short or is more than the task needs.

Correct answer: C — How much a downstream human check would catch, since the unchecked workload has to carry its own quality with no safety net.

• A. Shared subject matter makes the two workloads look alike, but it is not what determines the risk. Consistency for its own sake ignores the fact that only one of them has anyone checking the result.
• B. Equal volume does mean similar throughput demands, and that matters. It is still secondary to the fact that one workload reaches the outside world without any person having read it.
• C. Correct. The strength of the downstream check changes how much the output has to get right on its own, so the unchecked workload justifies more reasoning depth while the edited one can lean towards speed and allowance.
• D. Ownership determines who answers for a problem, not how likely a problem is. Seniority is an accountability question and does not tell the team where errors would actually escape unnoticed.

Correct answer: D — Ask Claude to restate each entry as the outcome sought, then flag entries where that outcome is unclear.

• A. Designing what was asked for skips the analysis entirely. If the submitter needed a weekly exception alert rather than a dashboard, building the dashboard confirms only that both parties misread the need together.
• B. Pushing the work back onto submitters is slow and often futile, since people describe needs in terms of solutions naturally. It also discards information the analyst already has in the entries.
• C. Counting entries by department produces a useful management view but leaves every entry still expressed as a solution. The backlog would be tidier and no better understood.
• D. Correct. The analytical move is to convert stated solutions back into the outcomes behind them, which is language work Claude does well at volume. Flagging the entries where no outcome can be recovered tells the analyst exactly which submitters to go back to.

Correct answer: B — Ask the three testers for specific examples, with what the output said and what it should have said.

• A. Rewriting on a hunch changes the arrangement without knowing what is wrong. If the guess misses, the pilot now has two versions and still no diagnosis, and the testers' original complaint remains unexplained.
• B. Correct. "Not quite right" is not actionable, and the cheapest way to make it actionable is to ask for cases. A concrete pair of what was produced and what was wanted turns vague dissatisfaction into a specific gap the instructions can address.
• C. Learning from the satisfied testers may eventually help, but it assumes the difference lies in how the tool was used rather than in the briefings themselves. It also leaves the three unresolved complaints uninvestigated.
• D. Direct observation is thorough but slow, and pausing the pilot stops the flow of evidence just when it is needed. Asking for examples costs an email and usually locates the problem immediately.

Correct answer: B — What changes, what stays with each person, and where their review is still required.

• A. A projected annual saving is the business case rather than the operating instruction. It tells the team why the change is happening but leaves them unsure what they are now responsible for.
• B. Correct. People need to know exactly where the boundary now sits: which part Claude drafts, which judgments remain theirs, and that their review is still a required step. Without that, reports are either double-handled or sent out with nobody checking them.
• C. Naming other departments builds confidence in the decision but says nothing about how this team should work. It is reassurance, not direction.
• D. Describing the testing is good practice and supports trust in the change. It still leaves the team without a clear statement of their own remaining responsibilities.

Correct answers: B, E — Use Research mode and check the citations it returns back against the original sources.; Ask Claude to draft the research plan and its open questions, then have the team refine it.

• A. Incorrect. Confident phrasing is not evidence, and market figures used for a real decision need checking against sources.
• B. Correct. Research mode returns citations, and checking them against the originals is how a claim is actually verified.
• C. Incorrect. Collapsing a study into one prompt and publishing the result removes every point where a human could catch an error.
• D. Incorrect. Stating the decision the study supports is what makes the research useful; withholding it produces unfocused output.
• E. Correct. Drafting the plan and the open questions is planning work Claude does well, and the team refining it keeps ownership where it belongs.

Correct answers: B, D — Where a human still reviews the output, and why that review step has been kept in place.; A specific, measured example of what the workflow improved during the pilot it ran on.

• A. Incorrect. Benchmark scores are not what a business audience needs, and they say nothing about this workflow's value.
• B. Correct. Naming where human review remains, and why, is honest limitation-setting and is what makes the value claim credible.
• C. Incorrect. Good instructions reduce error but do not eliminate it, so this claim is false and will damage trust later.
• D. Correct. A measured pilot result is concrete evidence of value rather than an assertion stakeholders are asked to take on trust.
• E. Incorrect. Promising that checking is no longer needed oversells the tool and sets the expectation that will break first.

Correct answer: D — Upload the reference to project knowledge and put the rule about unapproved abbreviations in the instructions.

• A. Two pages of reference text in the instructions field buries the rule that is meant to stand out.
• B. A behaviour rule stored as background material is far weaker than one stated as project direction.
• C. This puts the reference out of reach and turns a standing rule back into manual repetition.
• D. Correct. Reference material belongs in knowledge; the standing behaviour rule belongs in the instructions.

Correct answer: A — Tell the members who use the project what changed, so they know earlier answers may no longer hold.

• A. Correct. Maintaining shared configuration includes informing the people whose past work relied on the old source.
• B. Copies multiply the maintenance problem and guarantee that some members keep using the old matrix.
• C. Old conversations do not override project knowledge, and deleting them destroys a useful record.
• D. A title is not a change log, and members are not prompted to read it before chatting.

Correct answers: A, E — Connect the Drive folder, so the weekly handbook changes are picked up from the source.; Upload the fixed annual report into Project knowledge, since its content will not change.

• A. Correct. A connector reaches the live source, which is what a document that changes every week needs.
• B. Incorrect. Deleting the original to leave a single copy destroys the source of truth for no benefit.
• C. Incorrect. A one-time copy of a weekly-changing handbook goes stale fast and produces confidently outdated answers.
• D. Incorrect. Re-pasting a whole handbook per chat is the manual work Project knowledge and connectors exist to remove.
• E. Correct. A stable document is well suited to a one-time upload into Project knowledge, where it stays available to every chat.

Correct answer: A — Anyone who opens the shared project can draw the salary details out through ordinary questions.

• A. Correct. Project knowledge is background for every chat in the project, so sharing the project shares everything uploaded to it with everyone who can open it.
• B. A larger knowledge base is a performance and tidiness concern, not the serious exposure here.
• C. Documents can be removed from project knowledge at any time; the exposure is the real problem.
• D. Claude does not refuse to work in a project because a sensitive file has been uploaded to it.

Correct answer: C — It removes human oversight from hard-to-reverse actions, risking harm before anyone can intervene

• A. Incorrect. The oversight concern applies regardless of whether the integration uses the API or consumer product; both are subject to the same responsible-use principles.
• B. Incorrect. The choice of framework does not change the underlying risk of removing human oversight from consequential actions.
• C. Correct. Granting an agent unsupervised authority over irreversible, high-impact actions removes the human oversight needed to catch and correct model errors before they cause real harm.
• D. Incorrect. Agentic use must still comply with the full Usage Policy, and unsupervised high-stakes actions are not exempted from oversight expectations.

Correct answer: C — Owners can switch off the chat rating feature for the whole organisation in its settings.

• A. Ratings are feedback to Anthropic; they are not retained locally instead.
• B. No automatic redaction step is applied to submitted feedback.
• C. Correct. Team and Enterprise owners can disable rating so that conversations are not submitted with feedback.
• D. There is no approval workflow layered on top of ratings.

Correct answer: B — Incognito chats are never used to improve Claude, but the policy's rules for regulated data still apply.

• A. A personal plan in incognito mode is not the managed workspace and carries none of its contractual or administrative controls.
• B. Correct. Incognito chats are not used to improve Claude, but the policy question is whether regulated data may be processed in that environment at all, and a privacy toggle does not answer it.
• C. Overstating what incognito does is the trap here; the policy governs where regulated data may go, not only whether it trains a model.
• D. Pseudonymising fields does not make regulated records unregulated, and the approval question remains open.

Correct answer: D — Establish what the policy requires for a deliverable of this kind, before the report goes out.

• A. The client's comfort does not replace the firm's own requirements. Asking them cannot tell her what her firm obliges her to do, and their agreement would not excuse skipping it.
• B. This guesses at one requirement and treats it as the whole of the policy. Disclosure may be needed, but so may review, sign-off or restrictions on the client data involved.
• C. A sign-off may well be one of the requirements, but choosing it because it shifts responsibility is not following the policy. She still does not know what the policy asks for.
• D. Correct. The firm has a written standard covering exactly this situation, so the first step is to read it and identify which requirements apply to this deliverable. Everything else she might do is a guess at what the policy says.

Correct answer: C — Separate deciding who gets a place from writing the letters, and keep the decision with staff.

• A. Confirming that Claude can repeat the criteria says nothing about whether it should be the thing applying them. A correct restatement would make the team more comfortable handing over a decision it should not hand over.
• B. Using one school as the trial means real families receive decisions from an untested arrangement, and complaints are a late and unreliable signal of whether the allocation was right.
• C. Correct. A school place materially affects a child's life and the decision is hard to reverse once families have been told, so it belongs with accountable staff. Drafting the letters afterwards is a text task where an error is recoverable, and splitting the two makes the appropriate part safe to delegate.
• D. A firmer instruction addresses how faithfully the criteria are applied, not whether an allocation affecting children's schooling should be made this way at all.

Correct answer: C — Check what the participant consent and company policy allow; remove the names and employers that the summary does not need; run the summary and review it before it is circulated; note what was removed so the result can be interpreted properly.

• A. Data is edited before anyone has established what participants agreed to, so the amount of personal detail removed is guesswork and may be either too little or more than the analysis can afford.
• B. The personal data is processed and the output reviewed before either the consent position or the minimisation step, which is the sequence that causes the exposure rather than preventing it.
• C. Correct. Consent and policy set the boundary for what may be used at all; that boundary determines which identifiers can be stripped before anything is uploaded; the summary is then produced and reviewed within those limits; and recording what was removed lets colleagues read the findings knowing what the data no longer contains.
• D. The consent check happens first, but the full identifiable transcripts are still put into the tool before the names and employers are removed, so the minimisation step comes too late to protect anyone.

Correct answer: C — Rewrite the prompt to list the exact categories, the desired output format, and the criteria for choosing between them.

• A. Incorrect. Temperature affects sampling randomness, not whether the model understands what categories exist or how to choose between them; it does not address an underspecified prompt.
• B. Incorrect. Swapping models does not resolve ambiguity in the instructions themselves; an underspecified prompt will still under perform on a larger model.
• C. Correct. Claude responds well to clear, explicit instructions; a vague one-line prompt lacking the category list, output format, and selection criteria is the classic underperformance pattern, and the documented fix is to be specific about desired output and constraints.
• D. Incorrect. Manual relabeling is a workaround applied after generation, not a fix to the prompt causing the inconsistency.

Correct answer: C — Decide what she needs before starting; put the request and the relevant material in the opening message; work through follow-ups that refer back to it; open a new chat for the next unrelated task.

• A. Everything here is done before she has decided what she needs, so the opening message is assembled blind.
• B. Opening a chat for the next task before the current one has been started puts the last step first.
• C. Correct. Planning before starting, front-loading the material in one well-specified message, then building on it, is what keeps the exchange short; unrelated work starts fresh rather than dragging old context along.
• D. Following up before the request has even been made, and deciding what is needed last, inverts the whole habit.

Correct answer: C — Turn off the connectors and tools this task does not need, and switch them back on later.

• A. An instruction in every message is a per-message reminder rather than a change to the setup.
• B. An empty project addresses knowledge documents, not the tools and connections that stay enabled.
• C. Correct. Turning off tools and connectors that a task does not need is a recognised way to keep routine work efficient, and they can be re-enabled when the next task needs them.
• D. Removing them permanently costs her the reconnection work every time the other tasks come round.

Correct answers: A, C — Ask for two digests where an open risk was missed, with the missing risk actually named.; Compare those digests against the shift logs to see exactly what was left out of them.

• A. Correct. Vague feedback cannot be acted on; specific failing examples turn an impression into something diagnosable.
• B. Changing shared instructions before the cause is known risks fixing the wrong thing for everyone at once.
• C. Correct on separate grounds: only the source shows whether the risk was absent from the log, buried, or genuinely dropped.
• D. A score captures sentiment but still never says which risks went missing, or why they went missing.
• E. A format change may help presentation, but it cannot add a risk the digest never captured.