Practice exam 3 (full length) · CCAO-F

Practice exam 3 is a second full-length rehearsal with all-new questions. Take it a few days before your exam, timed at 120 minutes. Answers follow question 60.

1. Claude summarizes a supplier's financial statements and reports "net profit margin of 18%." You will use this in a risk assessment. What should you do?

A. Accept it, since financial statements were attached
B. Recalculate the margin from the statements, using code or a spreadsheet, before relying on it
C. Ask Claude whether 18% is reasonable for the industry
D. Round it to 20% for simplicity

2. An events coordinator wants to upload attendee lists, including dietary and accessibility needs, to plan catering. Policy treats health-related information as sensitive. What is best?

A. Upload the full list; catering is a legitimate purpose
B. Upload it to an incognito chat
C. Share only aggregated counts of dietary and accessibility needs, without names
D. Ask Claude to keep the list confidential

3. You need Claude to convert rough bullet notes into a polished client update. Which prompt is strongest?

A. "Polish these notes."
B. "Make this sound professional."
C. "Rewrite."
D. "Turn these notes into a client update for [client]'s operations director. Three short paragraphs: progress, issues and next steps. Confident, plain tone. Under 180 words. Keep all dates exactly."

4. A team spends hours each week answering the same questions about expense rules. Which solution fits CCAO-F best?

A. A Project grounded in the expense policy, used by the finance team to draft answers they review
B. A custom chatbot integrated into the HR system, built by the finance team
C. Asking Claude to answer from general knowledge
D. Removing the expense policy to simplify things

5. Which situation most justifies the most capable model tier?

A. Translating short product labels
B. Sorting emails into folders
C. Synthesizing three conflicting expert reports into a recommendation with trade-offs
D. Correcting spelling in a memo

6. Claude's draft press release quotes your CEO saying something she never said. What is the correct response?

A. Keep it if it sounds like her
B. Ask Claude to rephrase the quote
C. Remove the word "said"
D. Remove the quote and use only quotes the CEO has approved

7. A Project's instructions say "keep answers short," but answers are still long. What is the best fix?

A. Add "really short" to the instructions
B. Replace it with a measurable rule, such as "under 100 words or five bullets," plus an example
C. Delete the instructions
D. Use a different Project

8. After switching a summarization task to a lighter model tier, quality on complex technical reports dropped noticeably. What is the best next step?

A. Use the lighter tier for simple reports and a more capable tier for complex ones
B. Accept the lower quality
C. Stop summarizing reports
D. Use extended thinking on every task

9. Claude lists "key regulations" for your industry, including one you have never heard of. What is the most likely issue and the best response?

A. Your knowledge is outdated; add it to policy
B. Ask Claude to explain the regulation in more detail
C. The regulation may not exist or may not apply; verify it on an official government source
D. Ignore the whole list

10. A manager wants to show leadership the impact of a Claude pilot. Which evidence is most convincing?

A. A collection of the best outputs
B. Before-and-after measures of time, quality and errors, including review time
C. Number of prompts sent
D. Staff enthusiasm alone

11. Which is the strongest example of responsible disclosure?

A. Stating that a public report was drafted with AI assistance and reviewed by named experts, as your policy requires
B. Never mentioning AI
C. Saying a human wrote everything
D. Mentioning AI only if errors are found

12. You want Claude to evaluate a business idea critically, not just praise it. What instruction helps most?

A. "Be nice."
B. "Tell me it is good."
C. "Act as a skeptical investor. List the five biggest risks and the evidence that would change your mind."
D. "Be balanced."

13. Which TWO practices most improve the reliability of a summary of a long policy document? (Choose two.)

A. Ask Claude to quote the section behind each key point
B. Ask Claude to be confident
C. Compare key points against the policy text yourself
D. Ask for a longer summary
E. Change the font

14. A busy HR manager wants consistent formatting for interview notes across many chats. What is the best feature?

A. Research mode
B. Web search
C. A new chat with no context each time
D. A Project with a note template in the instructions

15. A connector to a shared drive gives Claude access to folders you rarely use, including confidential board papers. What should you do?

A. Limit use to the content needed, and ask your admin about narrowing access if possible
B. Ask Claude to summarize all the board papers
C. Share the Project with everyone
D. Nothing; access is harmless

16. Which is the best example of redesigning, rather than augmenting, a workflow?

A. Using Claude to correct grammar in existing reports
B. Asking Claude to shorten one email
C. Using Claude to translate one document
D. Restructuring client onboarding so Claude drafts a tailored welcome pack from the intake form, and staff review only exceptions

17. Claude's market analysis assumes customers will pay the same price in every region. You know prices vary by 30%. What kind of problem is this?

A. A formatting issue
B. A flawed assumption that must be corrected before using the analysis
C. Hallucinated citations
D. A tone problem

18. A team member wants to paste a customer's credit card number into Claude to draft a payment dispute letter. What should happen?

A. Proceed in incognito
B. Proceed if the letter is internal
C. Proceed and tell Claude to forget it
D. Exclude the card number; the letter can be drafted without it

19. You need a prompt that will be reused by colleagues for weekly client updates. What should it include?

A. Placeholders for variable details, a fixed structure, audience, tone and length
B. Only the client name
C. A request to "do the usual"
D. A very short instruction to keep it flexible

20. Claude's output in a Project seems to ignore a newly uploaded guideline. What should you check first?

A. Whether the model is down
B. Whether your account expired
C. Whether the guideline conflicts with existing instructions or older files, and whether the prompt refers to it
D. Whether the file is too small

21. A communications team receives a Claude draft for a sensitive announcement about layoffs. What review is appropriate?

A. A quick spell-check
B. No review if the draft reads well
C. Claude's self-review
D. Review by HR, legal and leadership before release

22. A school administrator asks whether Claude could mark student essays and assign final grades automatically. What is the best response?

A. Yes, with no teacher involvement
B. No, AI cannot help with education
C. Claude can suggest feedback against a rubric, but teachers should review and assign final grades
D. Yes, if the essays are short

23. A support team needs fast, low-cost drafts for routine password-reset questions. Which tier fits best?

A. The fastest, lowest-cost tier
B. The most capable tier with extended thinking
C. Whichever tier was used last
D. A different tier for each request

24. Which statement about a personal consumer account used for work is most accurate?

A. It is always acceptable
B. It may not meet your organization's policy and data terms; use approved tools for work data
C. It is safer than business plans
D. It is acceptable for confidential data if chats are deleted

25. Which knowledge-file set is best for a sales-enablement Project?

A. Every file from the shared drive
B. Old and new pricing sheets, for comparison
C. Current product facts, current pricing, approved case studies and objection-handling notes
D. Personal notes from all sales reps

26. Claude says a competitor "launched a new product last month." You asked in a plain chat with no web search. What should you assume?

A. The claim may be outdated or invented; verify with a current source
B. The claim is current, since Claude stated it
C. Claude has live access to news in every chat
D. The competitor told Claude

27. A first draft has the right content but the wrong tone for a grieving customer. What is the best next prompt?

A. "Try again."
B. "Rewrite with a warm, compassionate tone. Acknowledge the loss in the first sentence, avoid sales language, and keep the practical next steps."
C. "Be better."
D. "Make it longer."

28. Before expanding a Claude workflow to the whole department, what is most important?

A. Documented setup, standard prompts, training, review steps and an owner
B. Removing all human review
C. Switching every task to the top tier
D. Announcing it before testing

29. What is the best description of prompt injection?

A. A long prompt
B. A prompt with examples
C. Instructions hidden in content Claude processes, trying to override your intent
D. A prompt written in another language

30. You ask Claude to compare two insurance policies. It produces a confident table, but one cell says "covers flood damage" for a policy whose text excludes floods. What is the main lesson?

A. Tables are unreliable
B. Even structured outputs must be checked against the source on the points that matter
C. Insurance is too complex for Claude
D. Use a longer prompt

31. You need Claude to produce a polished, reusable onboarding checklist your team will keep editing. Which output surface is best?

A. A short inline answer
B. An Artifact or a document file
C. A Research report
D. Memory

32. A colleague's prompts produce inconsistent formats every week. Which fix is most efficient?

A. Ask them to type more carefully
B. Switch them to the top tier
C. Have them start a new account
D. Give them a standard prompt template, or put the format in a shared Project

33. A Research report states that a regulation takes effect on 1 January, citing a news article. The official government page says 1 July. What should you do?

A. Use 1 January, since it was cited
B. Average the dates
C. Use the official source, 1 July, and note the discrepancy
D. Ask Claude to choose

34. Which is the best way to keep a shared Project's knowledge trustworthy over time?

A. Let anyone upload anything
B. Never update it
C. Upload every draft for completeness
D. Assign an owner, keep one current version of each source and review on a schedule

35. For a sensitive negotiation email, you want Claude to consider the other party's perspective before drafting. What should you ask?

A. "First list the other party's likely priorities and concerns, then draft an email that addresses them while protecting our position on price."
B. "Write a tough email."
C. "Write an email."
D. "Be persuasive."

36. A team plans to measure a Claude pilot only by how many documents Claude produces. What is wrong with this?

A. Nothing
B. Volume ignores quality, review time and errors
C. It overstates costs
D. It is too detailed

37. A colleague suggests using Claude to write anonymous negative reviews of a competitor. What is the correct response?

A. Acceptable if the reviews are accurate
B. Acceptable if posted from a personal account
C. Acceptable for internal use
D. Decline; it is deceptive and may be illegal

38. Claude's summary of a customer contract says the notice period is 30 days. What is the best way to confirm?

A. Find and read the notice clause in the contract
B. Ask Claude if it is sure
C. Assume 30 days is standard
D. Ask Claude to reformat the summary

39. Claude produces strong first drafts, but your team spends a long time fixing the same factual errors about product specifications. What is the best optimization?

A. Hire more reviewers
B. Ground drafts in a current product facts file and instruct Claude to use only that file for specifications
C. Stop reviewing
D. Shorten the drafts

40. You need to check whether a long chat's earlier decisions are still being followed. Claude has started contradicting them. Which action fits best?

A. Ask Claude to try harder
B. Switch to Research mode
C. Get a summary of agreed decisions, verify it, and continue in a new chat or Project
D. Continue and correct each contradiction

41. Which statement best explains Claude's value to a skeptical finance director?

A. "It replaces the finance team."
B. "It is never wrong with numbers."
C. "It is a fun tool to try."
D. "It can cut drafting time for reports and commentary; we verify every figure against the ledger, and approvals stay with finance."

42. Where should a rule that applies only to one client's deliverables go?

A. That client's Project instructions
B. Global memory
C. Every colleague's personal preferences
D. A Skill used by the whole company

43. You want Claude to produce a weekly risk log from status reports. Which instruction best ensures consistent output?

A. "Find risks."
B. "Be thorough."
C. "Return a table with columns: risk, likelihood (High, Medium, Low), impact (High, Medium, Low), owner, mitigation. Use 'Unassigned' if no owner is named."
D. "List anything worrying."

44. A Claude-generated training quiz contains one question with two correct answers. What does this show?

A. Quizzes should not use AI
B. The topic is too hard
C. Learners should guess
D. Generated assessments need expert review for accuracy and clarity before use

45. Policy forbids uploading unreleased product designs to AI tools. A designer wants feedback on a new concept. What should they do?

A. Follow the policy, for example by asking for feedback on general design principles without uploading the design, or by seeking an approved exception
B. Upload it in incognito mode
C. Upload it and delete the chat
D. Upload a screenshot instead of the file

46. A department wants to start using Claude. Which first project is the best choice?

A. Automated decisions on staff promotions
B. Publishing customer communications without review
C. Drafting internal meeting summaries that attendees review
D. Integrating Claude into the payroll system

47. You notice your team's prompts for a recurring task have grown very long and inconsistent. What should you do?

A. Move the stable parts into Project instructions and keep a short, standard prompt for the variable parts
B. Make them even longer
C. Let everyone keep their own version
D. Stop doing the task

48. Which task should use Research mode rather than a plain chat?

A. Rewording a paragraph
B. Drafting a thank-you note
C. Summarizing an uploaded memo
D. Building a cited overview of recent industry funding announcements

49. Which instruction best handles uncertainty in a Project?

A. "Always give an answer."
B. "If information is missing from the sources, say what is missing and ask, rather than guessing."
C. "Never ask questions."
D. "Guess if needed."

50. You asked Claude to summarize a survey's open-text responses. The summary says "most respondents were positive," but you notice many negative comments. What should you do?

A. Trust the summary
B. Ask Claude to count positive, negative and neutral responses with code, show examples, and check a sample yourself
C. Delete the negative comments
D. Ask for a more positive summary

51. You need Claude to write ten FAQ answers in your company's voice. Which approach works best?

A. Ask for "our voice"
B. Ask for formal language
C. Write all ten yourself first
D. Provide two or three approved FAQ answers as labeled examples, plus voice rules

52. A manager wants to use Claude to analyze employee engagement survey comments. What is the best governance step?

A. Include employee names for context
B. Remove identifying details, use the approved tool, and report themes rather than individual comments where identities could be inferred
C. Share raw comments with all managers
D. Ask Claude to identify who wrote each comment

53. An analyst wants to know which reporting steps to hand to Claude first. Which steps should they prioritize?

A. Frequent, text-heavy steps whose output is easy to check
B. Rare, high-risk judgment calls
C. Steps that need system integration
D. Steps with no clear definition of good output

54. You suspect a Project instruction is causing poor outputs, but you also changed the model tier and a knowledge file this week. What is the best approach?

A. Revert everything at once
B. Change something else
C. Test changes one at a time to isolate the cause
D. Start a new Project immediately

55. A user wants Claude to remember their job role and preferred summary format across all chats. What fits best?

A. Memory or personal preferences
B. A connector
C. Research mode
D. A new Project for each chat

56. A Project is used by both the sales and legal teams, and their needs conflict. What is the best configuration?

A. One Project with contradictory instructions
B. No instructions at all
C. Separate Projects with instructions and knowledge suited to each team
D. Let Claude decide which team is asking

57. Which TWO signals suggest an output needs closer checking? (Choose two.)

A. It uses headings
B. It contains precise statistics with no source
C. It is written in plain English
D. Its totals do not match its parts
E. It is under 200 words

58. You want three different marketing angles for a product, each developed enough to compare. Which prompt pattern fits best?

A. Extract to table
B. Options then decide: three distinct angles with pros, cons and audiences, then a recommendation
C. Grounded answer
D. Summarize

59. Which statement about accountability is correct?

A. If Claude makes an error, Anthropic is responsible for your decision
B. Nobody is responsible for AI errors
C. The people and organization using Claude remain responsible for decisions and published content
D. Claude is responsible

60. A team's Claude workflow is popular, but managers worry about quality slipping over time. What is the most effective safeguard?

A. Ban the workflow
B. Remove review to save time
C. Use the top tier for everything
D. Keep a regular sampling review, track errors, and update instructions and knowledge when issues appear

Practice exam 3: answer key
Q Answer Domain Why
1 B D2 Figures used for decisions must be recomputed or checked
2 C D6 Aggregate and minimize sensitive data
3 D D1 Audience, structure, tone, length and fidelity to facts

4 A D4 A grounded Project with review is in scope; building integrations is not 5 C D3 Synthesis of conflicting expert input is complex reasoning 6 D D2 Never publish invented quotes

7 B D5 Measurable rules and examples work
8 A D7 Match the tier to task complexity
9 C D2 Verify unfamiliar regulations at official sources

10 B D4 Measured before-and-after evidence including review time 11 A D6 Disclosure as policy requires, with named human review

12 C D1 A critical role with specific asks counters agreeable output 13 A, C D2 Quoted support plus your own comparison

14 D D3 Recurring format belongs in a Project

15 A D5 Use only what you need; narrow over-broad access with your admin 16 D D4 The process itself changes

17 B D2 Flawed assumptions invalidate analysis until corrected

18 D D6 Payment card data is not needed; exclude it

19 A D1 Reusable templates need placeholders and fixed structure 20 C D7 New files can conflict with existing instructions or files Q Answer Domain Why

21 D D2 Sensitive, high-impact communications need expert review 22 C D4 Claude supports feedback; teachers keep grading decisions 23 A D3 Routine, high-volume drafting suits the fastest tier

24 B D6 Use approved tools and data terms for work

25 C D5 Current, approved, relevant knowledge only

26 A D2 Without search, recent claims may be outdated or invented 27 B D1 Specific tone guidance

28 A D4 Standardize before scaling
29 C D6 Definition of prompt injection
30 B D2 Structured outputs still need source checks

31 B D3 Reusable, editable output belongs in an Artifact or file 32 D D7 Standard templates or shared Projects fix inconsistency

33 C D2 Official sources win over secondary reporting
34 D D5 Ownership, single versions and scheduled review
35 A D1 Ask for perspective-taking before drafting
36 B D4 Volume alone ignores quality and review cost
37 D D6 Fake reviews are deceptive
38 A D2 Check the primary source
39 B D7 Grounding reduces recurring factual errors
40 C D3 Summarize, verify and restart or persist
41 D D4 Value, controls and human approval
42 A D5 Client-specific rules go in that client's Project
43 C D1 A defined table structure gives consistency
44 D D2 Generated assessments need expert review
45 A D6 Follow policy; seek approved alternatives
46 C D4 Start with low-risk, easily checked work
47 A D7 Move stable context into configuration
48 D D3 Recent, multi-source, cited information
49 B D5 Explicit uncertainty behavior prevents guessing
50 B D2 Quantify with code and check a sample
51 D D1 Examples plus voice rules
52 B D6 Anonymize and report themes to protect identities
53 A D4 Prioritize frequent, text-heavy, checkable steps
54 C D7 Isolate changes
55 A D3 Cross-chat preferences belong in memory or preferences
56 C D5 Separate configurations for conflicting needs
Q Answer Domain Why
57 B, D D2 Unsourced precise figures and internal inconsistency
58 B D1 Options-then-decide pattern
59 C D6 Humans and organizations stay accountable
60 D D4 Ongoing sampling and maintenance sustain quality

Score it with the same domain table and thresholds as practice exam 2. Question numbers by domain: D1: 3, 12, 19, 27, 35, 43, 51, 58. D2: 1, 6, 9, 13, 17, 21, 26, 30, 33, 38, 44, 50, 57. D3: 5, 14, 23, 31, 40, 48, 55. D4: 4, 10, 16, 22, 28, 36, 41, 46, 53, 60. D5: 7, 15, 25, 34, 42, 49, 56. D6: 2, 11, 18, 24, 29, 37, 45, 52, 59. D7: 8, 20, 32, 39, 47, 54.