Free CCAO-F mock exam, timed, with a reason for every option

Answer as you read

The CCAO-F paper: every question

60 items, each tagged with its official domain. Work through the paper one question at a time: pick an option to check it, or ask for the key, and read why each option holds or fails. Nothing is saved. For a scored sitting against the clock, use the start modes above.

1. Question 1Domain 1: Prompting and Task ExecutionSingle answer

   Halvard Marine Insurance uses Claude to draft letters explaining why a claim was declined. The first draft, written from a two-line request with nothing attached, is fluent, but it cites a policy clause number absent from the customer's policy wording, and a readability check puts it at postgraduate level. Letters must quote only clauses from the customer's own wording and be readable by customers without insurance training. The claims handler has time for one more revision before the letters are due. Which prompt revision best addresses both problems in that single pass?

   Select the best answer.

   AAsk Claude to redraft in a warmer, more personal tone and add a closing paragraph inviting the customer to call the claims team, since a friendlier letter is easier to accept than a technical one.

   BAsk Claude to review its own first draft, confirm that every clause number it cited is genuine, and then rewrite the letter for a general audience using the same information it already has.

   CAttach the customer's policy wording, instruct Claude to quote clause numbers only from that text and to flag anything it cannot find there, name the audience as customers without insurance training, and set a plain-language target.

   DSplit the work into two conversations, one to identify the correct clause from the policy wording and a second to write the letter, and paste the identified clause across by hand for each customer's letter before it is sent.
   Show the answer
2. Question 2Domain 1: Prompting and Task ExecutionSingle answer

   Corrigan Freight asked Claude, in one request, to read forty carrier contracts, extract each penalty clause, compare it with the company's new standard terms, and rank the contracts for renegotiation. The output mixed clauses between contracts, covered thirty-four of the forty, and gave a ranking that could not be traced to any specific clause. The ranking goes to the board, which has asked that every recommendation be traceable to contract text. Which restructuring of the request best meets that requirement?

   Select the best answer.

   ARun an extraction pass per contract into a fixed table of contract, clause text, penalty amount and trigger, then a separate comparison pass against the standard terms, then a ranking pass that cites table rows.

   BRepeat the original request with a stronger instruction to be thorough, to confirm that all forty contracts have been covered, and to state the clause behind each ranking position before presenting the final list.

   CAsk Claude to produce the ranking first, since that is what the board needs, and then generate the supporting clause extracts only for the ten contracts that appear at the top of the recommended list.

   DUpload all forty contracts to a Project so Claude has them as knowledge, then ask the original question again in one pass and rely on retrieval to keep each contract's clauses separate from every other contract's clauses.
   Show the answer
3. Question 3Domain 1: Prompting and Task ExecutionSingle answer

   Brightwater Council's communications officer has iterated a prompt for roadworks notices four times, each time asking Claude to make the notice clearer. The notices are now shorter, yet the call centre still logs about thirty calls per notice from residents asking which dates the road is closed. In the current draft the closure dates appear once, in the third paragraph. Every new version of a notice must go through a legal review that takes two working days. What is the most effective next iteration?

   Select the best answer.

   AAsk Claude once more to make the notice clearer, but this time to explain what it changed and why, so the officer can judge whether the successive edits are converging on something residents understand.

   BCompare the last two drafts against the call-log questions, name the failure as closure dates buried in the third paragraph, and change the prompt to require dates and affected roads in the first two lines.

   CStart a fresh conversation and ask Claude to write the notice from a blank page, since four rounds of edits have accumulated wording that no longer serves the resident and a clean draft avoids that accumulated debt.

   DAsk Claude to produce five alternative versions of the notice and send all five to the legal team together, so that they can choose the version they believe residents will understand most easily on first reading.
   Show the answer
4. Question 4Domain 1: Prompting and Task ExecutionSingle answer

   Sable & Rooke LLP's knowledge manager uses a single prompt template for every request: "Give me a thorough answer about the following." It works for internal memos. For naming a new client seminar series it produced three near-identical ideas, and for analysing a two-hundred-row matter list it produced confident totals that did not add up. The manager wants one adjustment that fixes both without a separate template for each fee earner. Which adjustment reflects adapting the prompting strategy to the task type?

   Select the best answer.

   AKeep the single template, since consistency across the firm matters more than any one output, and add a line asking Claude to try harder on tasks where the first answer seems thin or uncertain in its figures.

   BFor both tasks, instruct Claude to adopt the persona of a senior partner, on the basis that the quality gap comes from the model not understanding the standard the firm expects of its written work.

   CFor the seminar names, ask for one carefully argued recommendation; for the matter list, ask for a narrative summary rather than figures, so that no total can be reported incorrectly to a fee earner.

   DFor the seminar names, ask for many varied candidates and defer evaluation until the list is complete; for the matter list, supply the rows as a table, state the calculations, and ask for the working.
   Show the answer
5. Question 5Domain 1: Prompting and Task ExecutionSelect 2

   Kestrel Grocery's dairy category manager asked Claude to write a negotiation brief for the dairy category and received a generic document that could apply to any supplier. Each brief must reflect that supplier's current contract terms and the manager's walk-away price, and because briefs are sometimes forwarded to the supplier by mistake, internal margin figures must never appear in them. Which TWO elements of the prompt would do most to fix the output? (Select two.)

   Select exactly 2. 0/2 selected.

   AAn instruction to write in the confident voice of an experienced negotiator, so the brief carries authority with whoever reads it and does not read as a template filled in by hand.

   BThe specific inputs the brief must draw on, meaning the supplier's current contract terms and the walk-away price, attached or pasted, with an instruction to use no figures from anywhere else.

   CA request that the brief be as long as necessary to anticipate every objection the supplier could raise, so the manager is never caught without a prepared response in the negotiating room.

   DAn explicit statement of who might read the brief and what must not appear in it, naming internal margin figures as excluded content that Claude must neither include nor imply.

   EAn instruction for Claude to fill any gap in the supplied terms with typical terms for the dairy category, so that every brief is complete and ready to use without any further editing by the manager.
   Check answerShow the answer
6. Question 6Domain 1: Prompting and Task ExecutionSelect 2

   Ardent Robotics wants Claude to produce a monthly supplier quality report from inspection data: calculate defect rates, flag suppliers above a two percent threshold, write the narrative, and propose corrective actions. Done as one request, the narrative discussed two suppliers the table had not flagged. Corrective actions can trigger a supplier audit, and the quality director must approve any before it is drafted. Which TWO decomposition choices are most appropriate? (Select two.)

   Select exactly 2. 0/2 selected.

   ASeparate the calculation and flagging step from the narrative step, and pass the narrative step only the verified flagged table so it cannot describe suppliers that were not flagged.

   BAsk Claude to complete the whole sequence in one response but number the steps clearly, so a reader can see where the calculation ends and the narrative begins when checking the finished report.

   CPlace a checkpoint after the flagged list is produced and stop there for the quality director's approval, so no corrective action is drafted until a human with authority has agreed the list.

   DMove the corrective actions to the beginning of the request, since they are the most consequential output and should receive the most attention from Claude before the context fills with data.

   EAsk Claude to produce the narrative and the defect-rate table in one response and to reconcile them against each other before finishing, so that the two outputs can never disagree in the finished report.
   Check answerShow the answer
7. Question 7Domain 1: Prompting and Task ExecutionSingle answer

   Oakhurst Hospital Group's training coordinator asked Claude for ten quiz questions on the hand-hygiene policy for new starters. The questions were accurate, but in a pilot with twelve staff every participant scored ten out of ten, so the quiz cannot show who has understood the policy. The induction runs tomorrow morning and the coordinator has one afternoon left. Which iteration is most likely to produce a usable quiz in that time?

   Select the best answer.

   ATell Claude the pilot result, state the intended result (roughly seven of ten correct for a typical new starter), and ask for questions that require applying the policy to a described ward situation rather than remembering a rule.

   BAsk Claude to make the questions harder, run the twelve-person pilot again, and repeat that loop until the average score falls into a range the coordinator considers acceptable for an induction quiz of this kind.

   CAsk Claude to add two further answer options to every question, since a five-option question is harder to guess correctly and the pilot score will fall without changing the substance of any question.

   DReplace the local policy document with the full national guideline so Claude has more material to draw on, and ask for the same ten questions again with no other change to the wording of the request.
   Show the answer
8. Question 8Domain 1: Prompting and Task ExecutionSingle answer

   Vantage Point University's research office asked Claude to research the funding landscape for climate adaptation and received a fluent overview with no sources, two funders the office could not find anywhere, and no indication of which programmes were current. The output will decide which three funding calls the office pursues next quarter, so currency and sourcing matter more than breadth. Which prompting approach best fits a research task with these stakes?

   Select the best answer.

   AAsk for a brainstorm of every funder Claude can think of, then have staff verify the top twenty by hand, on the basis that breadth matters most at the first stage and accuracy can be added later.

   BFrame the request as drafting and ask Claude to write the application narratives for the three most likely funders directly, since the narrative is the eventual deliverable and the office is short of staff time.

   CFrame it as research: state the decision it supports, require a source and date for every funder and deadline, have Claude mark anything unsourced as unverified, and use a mode that searches current sources.

   DAsk the same question in three separate conversations and keep only the funders that appear in all three answers, treating agreement between independent runs as verification of each funder's existence and currency.
   Show the answer
9. Question 9Domain 2: Output Evaluation and ValidationSingle answer

   Pennant Mutual Bank's compliance analyst asked Claude to condense a sixty-page internal audit report into a one-page brief for the risk committee. The report contains twelve numbered findings. The brief reads well and the analyst has confirmed the figures for the nine findings she has checked so far, but the committee pack closes in thirty minutes. Which evaluation step best addresses both accuracy and completeness in the time available?

   Select the best answer.

   ARead the brief twice for fluency and confirm it fits on one page, since the committee has complained about long papers and length compliance is the constraint most often breached in these packs.

   BList the twelve finding numbers from the report's contents page, tick each one against the brief to confirm it appears, and spot-check the figures in the three findings not yet verified.

   CAsk Claude whether it covered all twelve findings and whether every figure matches the report, and accept the brief for the committee pack if it answers yes to both of those questions.

   DSend the brief to the head of internal audit for a full read-through before it goes into the pack, since a committee paper should always be signed off by the function that wrote the underlying report.
   Show the answer
10. Question 10Domain 2: Output Evaluation and ValidationSingle answer

    Loomis Software's marketing lead asked Claude for a competitor comparison to publish on the company website. The draft includes the line that industry analysts report a 43 percent reduction in onboarding time, attributed to a named research firm. The lead has searched the firm's publications and the wider web for twenty minutes and cannot find the figure or any report matching the description. Publication is scheduled for tomorrow. What should she conclude, and do?

    Select the best answer.

    AThe figure is probably a paraphrase of a real finding, so ask Claude for the exact report title, year and page number so the citation can be completed and the line kept in the published piece.

    BKeep the statistic but soften the attribution to "some analysts suggest", which removes the need to name a source while preserving a point that is broadly true of the product category as a whole.

    CAsk Claude to rate its confidence in the statistic and keep the line if the confidence is above eighty percent, since the model has been exposed to far more industry reports than the lead has read.

    DTreat the figure as fabricated until a source is found, remove the line from the draft, and check every other specific number and attribution in the piece by the same standard.
    Show the answer
11. Question 11Domain 2: Output Evaluation and ValidationSelect 2

    Hollins Recruitment used Claude to screen two hundred applications for an engineering role against the client's job description and asked for a shortlist of twenty. Nineteen of the twenty attended one of three universities and one is a woman, while the applicant pool spanned forty universities and was thirty-eight percent women. The client expects the shortlist by the end of the day. Which TWO findings are most important to raise before the shortlist is used? (Select two.)

    Select exactly 2. 0/2 selected.

    AThe shortlist can go to the client as it stands, because Claude applied the same job description to all two hundred applications, and a screen that holds every candidate to one standard is fair by construction.

    BThe recruiter should ask Claude to add women to the shortlist until the proportion matches the applicant pool, which corrects the imbalance without needing to re-run the whole screen.

    CThe output shows a pattern that suggests bias relative to the pool, so the shortlist should not go to the client until the criteria Claude actually applied are made explicit and reviewed by a person.

    DThe result should be checked by running the same prompt again, and if the second shortlist closely resembles the first, the pattern reflects the candidate pool rather than the screening tool.

    EThe prompt probably contained or implied proxies, such as institutional prestige, that the job description did not require, and those should be identified and removed before any re-screen.
    Check answerShow the answer
12. Question 12Domain 2: Output Evaluation and ValidationSingle answer

    Greyfriars Museum Trust's curator asked Claude to draft display labels for forty objects in a new gallery, including the date and maker of each. The labels will be engraved at a cost of about ninety pounds each, and the trust holds an accession register that is the authoritative record for every object. In three labels Claude's dates differ from the register by a decade or more. Which validation technique should the curator apply before sending the labels for engraving?

    Select the best answer.

    ACheck every date and maker in all forty labels against the accession register before anything is engraved, treating the register as the controlling source even where Claude's version seems more plausible.

    BAsk a second curator to read all forty labels for plausibility, since a specialist will notice an implausible date faster than a line-by-line comparison with the register would, and time before engraving is short.

    CAsk Claude to re-read its own labels and list any dates it is unsure about, then check only those flagged labels against the register to protect the curator's limited time before the engraving deadline.

    DSearch online for each object and compare Claude's dates with the leading results, since agreement between two independent sources is stronger evidence than a single entry in the trust's own accession register.
    Show the answer
13. Question 13Domain 2: Output Evaluation and ValidationSingle answer

    Northgate Housing Association's finance officer pasted a table of twelve estates with rent due and rent collected, and asked Claude for arrears, meaning rent due minus rent collected, as a percentage of rent due for each estate. The returned table shows one estate at 104 percent and two others that do not match her rough mental arithmetic. The figures go to the housing regulator on Friday and will be relied upon. Which validation technique is appropriate?

    Select the best answer.

    ARecompute by hand the three percentages that look wrong, and if Claude's figures for those rows turn out to be correct, accept the table since the remaining rows were produced by the same method.

    BAsk Claude to explain how it reached 104 percent and, if the explanation is coherent, keep the figure, since arrears can exceed a period's rent when older debt is included in the balance.

    CRecompute all twelve percentages in a spreadsheet from the original table she pasted, then compare row by row with Claude's output and correct any discrepancy from her own calculation.

    DAsk Claude to perform the calculation again in a new conversation and adopt whichever set of figures the second run produces, on the basis that a fresh attempt is less likely to repeat the same slip.
    Show the answer
14. Question 14Domain 2: Output Evaluation and ValidationSingle answer

    Fairhaven Pharmacies' regional manager used Claude to draft two documents on the same afternoon: the weekly staff rota emails for eight branches, and a customer leaflet explaining common interactions between over-the-counter medicines. Both drafts read well and neither contains an obvious error. The manager has about an hour before both need to go out and one pharmacist is on shift. How should human review be allocated?

    Select the best answer.

    AReview both drafts personally for tone and typographical errors, since the manager is accountable for everything issued under the pharmacy's name and should not add to the pharmacist's workload on a busy shift.

    BSend the rota emails after a quick read, and route the interaction leaflet to the pharmacist for clinical review before any customer sees it, regardless of how confident and complete the draft reads.

    CSend both once Claude has been asked to double-check its own content for accuracy, since the manager's time is the scarce resource and the drafts have shown no sign of error on a careful read.

    DRoute both documents to the pharmacist, since having a clinician sign off every staff-facing and customer-facing communication is the safest standing policy for a healthcare retailer to adopt.
    Show the answer
15. Question 15Domain 2: Output Evaluation and ValidationSelect 2

    Calder Valley Water's duty communications officer used Claude to draft a website notice about a boil-water advisory. The draft, produced in under a minute from the incident log, lists the affected postcodes and gives an estimated time by which normal supply will be restored. The notice must be live within the hour, and the officer could publish it directly. Which TWO features of this situation most strongly indicate that a qualified person must review the draft before publication? (Select two.)

    Select exactly 2. 0/2 selected.

    AThe draft makes factual claims, the postcodes and the restoration time, whose error would cause public harm and could not be detected by the residents reading the notice.

    BThe draft is longer than the utility's usual advisory notices, which suggests that Claude has added content that did not come from the incident log and should be trimmed before release.

    CThe draft was produced in less than a minute, which is too fast for the model to have considered the incident log carefully, so the speed itself indicates that the content is unreliable.

    DPublishing a restoration estimate commits the utility to a public statement with regulatory and legal consequences, and that decision belongs to a named accountable role rather than to the drafter.

    EClaude did not express uncertainty about the restoration time, so the estimate is probably reliable and any review can be limited to checking formatting and house style before publication.
    Check answerShow the answer
16. Question 16Domain 2: Output Evaluation and ValidationSingle answer

    Whitlock Engineering's technical author asked Claude for a release note covering a firmware update to an industrial controller and received a single nine-hundred-word document. It must serve field engineers, who need the exact installation and rollback steps; customers, who need to know what changed and whether any action is required; and the sales team, who need a one-line statement of the benefit. Which approach to adapting the output is most appropriate?

    Select the best answer.

    ASend the nine-hundred-word note to all three groups unchanged, since a single document avoids inconsistencies between versions and each reader can skip the sections that do not apply to them.

    BAsk Claude to cut the note to three hundred words so every audience can read it in full, keeping the technical installation steps and removing any language that reads like marketing copy.

    CSend the note to the field engineers only, and leave the customer and sales versions to be written by those teams, since they understand their own audiences better than the technical author does.

    DAsk Claude to produce three versions from the same verified content, each framed for its audience and its next action, and check that the facts are identical across all three before release.
    Show the answer
17. Question 17Domain 2: Output Evaluation and ValidationSingle answer

    Ironbridge Savings' product manager asked Claude for two drafts of a letter announcing a fee change. The first, from a short prompt, is warm and easy to read. The second, from a detailed prompt with the fee schedule attached, is stiffer but states the new fee, its effective date and the customer's right to close the account without charge, exactly as the regulator requires. The letter is a regulated communication and must be sent next week. How should she compare the two and proceed?

    Select the best answer.

    ATake the second draft as the base because it carries the mandatory content, ask Claude to improve its readability while instructing it not to alter the fee, date or account-closure wording, and re-verify those three elements afterwards.

    BChoose the first draft because customer comprehension is the whole point of the letter, and ask the compliance team to insert any required regulatory wording after the tone and structure have been settled.

    CMerge the two by hand, taking paragraphs from each, since the manager knows which sentences read best and which paragraphs contain the facts the regulator requires the letter to state to every customer.

    DAsk Claude which of the two drafts is the better letter and adopt its recommendation, since it produced both and can judge them against the same criteria the manager would use to decide.
    Show the answer
18. Question 18Domain 2: Output Evaluation and ValidationSingle answer

    Harrowgate Airports' operations analyst asked Claude to analyse three thousand passenger complaints from the last quarter and summarise them. The result will be handed to the data team, who must load complaint categories and counts into the airport's business intelligence dashboard by Monday. The data team has said they will not hand-transcribe anything from a document. Which output format should the analyst request?

    Select the best answer.

    AAn inline narrative summary in the conversation that describes the main complaint themes and their rough proportions, which the data team can read and interpret before setting up the dashboard categories.

    BA formatted document artifact with a heading for each complaint category and a paragraph of analysis beneath each, suitable for circulating to the leadership team alongside the finished dashboard.

    CA structured data output, such as a table or CSV with one row per category giving the count and an example complaint, plus a short inline note on how the categories were assigned.

    DA slide-style artifact with a chart for each complaint category, since the dashboard is a visual tool and the analyst's output should match the way leadership will eventually see the complaint data on screen.
    Show the answer
19. Question 19Domain 2: Output Evaluation and ValidationSelect 2

    Coastline Ferries' marketing coordinator needs three things from Claude this week: a forty-page brand guidelines document that four colleagues will edit and comment on over the next month; a quick answer about the approved punctuation of the company tagline; and a list of five hundred approved image files with their licence expiry dates, to be imported into the digital asset system. Which TWO format choices are correct? (Select two.)

    Select exactly 2. 0/2 selected.

    AThe tagline punctuation answer as a document artifact, so that it can be kept alongside the guidelines and referred to whenever the question comes up again across the team.

    BThe brand guidelines as a document artifact, because a long work product that will be revised by several people over weeks needs to be edited and exported rather than read in a chat stream.

    CThe image list as inline prose grouped by expiry month, because a human will need to read through the list before it is imported and prose is easier to scan than a five-hundred-row table.

    DThe brand guidelines as a series of inline chat responses, one section per message, so that each section can be approved by a colleague separately before the next section is written.

    EThe image list as a structured format, a table or CSV with fixed columns for file name and expiry date, because it will be consumed by a system rather than read by a person.
    Check answerShow the answer
20. Question 20Domain 2: Output Evaluation and ValidationSelect 2

    Ravenscourt Hotels' general manager asked Claude to draft a reply to a three-page guest complaint that raised five separate issues and requested a refund. The draft is courteous, addresses three of the five issues, and offers "appropriate compensation" without stating an amount. The manager has authorised a specific goodwill credit and wants the reply sent today. Which TWO evaluation checks matter most before it is sent? (Select two.)

    Select exactly 2. 0/2 selected.

    ACheck that the reply is at least as long as the complaint, since a guest who wrote three pages will read a shorter reply as dismissive regardless of what it actually says.

    BAsk Claude to grade the draft's empathy on a ten-point scale and revise it until the score reaches at least nine, since tone is what determines whether a complaint is closed.

    CCheck each of the five issues in the complaint against the draft and confirm that every one is either addressed or deliberately deferred with a stated reason.

    DConfirm that the compensation offer states the specific credit the manager has authorised, since a vague offer of "appropriate compensation" creates an open commitment the hotel may not intend.

    EConfirm that the draft opens with an apology in its first sentence, since the guest's first impression of the reply is what decides whether the complaint is closed or escalated to a review site.
    Check answerShow the answer
21. Question 21Domain 2: Output Evaluation and ValidationSingle answer

    Marlow Insurance's underwriting assistant asked Claude to summarise a twenty-five-page property survey for an underwriter who will price the policy from the summary alone. The summary opens by stating that the roof was replaced in 2019, and its risks section states that the roof retains original 1970s tiles requiring replacement within five years. The assistant has the survey but has not read it. What should she conclude, and do?

    Select the best answer.

    ABoth statements can be true if the roof replacement was only partial, so leave the summary exactly as it stands and let the underwriter resolve the nuance when pricing the policy from the summary alone.

    BThe summary is internally inconsistent on a material fact, so find both roof statements in the survey, decide which the source supports, and correct the summary before the underwriter sees it.

    CAsk Claude which of the two statements is correct and adopt its answer, since it processed the full survey and the assistant has not yet read the document herself.

    DDelete the statement in the risks section, since the opening summary is where the model states its principal findings and is therefore the more reliable of the two passages.
    Show the answer
22. Question 22Domain 3: Product and Model SelectionSingle answer

    Penrose Consulting's engagement lead is fourteen weeks into a client programme. Every Monday she opens a new conversation, uploads the same six background documents, re-explains the client's structure and reporting conventions, and only then asks Claude for that week's analysis. The documents change perhaps once a quarter; the weekly analysis is different each time. She estimates the set-up costs forty minutes a week. Which product feature best fits this pattern of work?

    Select the best answer.

    AA single long-running conversation for the remainder of the engagement, so that all fourteen weeks of context accumulate in one place and nothing needs to be uploaded or explained a second time.

    BResearch mode at the start of each week, so that Claude assembles the client's background from public sources itself and the lead does not need to upload the six documents or explain the structure.

    CAn artifact containing the client context that she copies into each new conversation, which keeps the context in one editable place without requiring her to set up or maintain a Project.

    DA Project holding the six background documents as knowledge and the client's structure and conventions as instructions, with a fresh conversation inside it each Monday for that week's analysis.
    Show the answer
23. Question 23Domain 3: Product and Model SelectionSingle answer

    Whitby Port Authority's digital team is choosing models for three new uses: extracting five fixed fields from thirty thousand standardised customs declarations a month; drafting a weekly operations summary from structured shift reports; and preparing the reasoning for contested berth-allocation decisions that shipping lines can appeal. The team lead proposes a single model for all three to simplify procurement and support. Which understanding of the Claude model families should guide the decision?

    Select the best answer.

    AThe three families trade capability against speed and cost: Haiku suits the high-volume field extraction, Sonnet suits the weekly summaries, and Opus suits the contested reasoning that shipping lines can appeal.

    BSonnet is the general-purpose family, so it is the sound single choice for all three uses: capable enough for the appeals, cheap enough for the extraction, and simplest for procurement and support to standardise on.

    COpus is the most capable family and therefore the most accurate on every task, so it should be used for all three uses to remove any risk of error in customs data or in berth-allocation decisions.

    DPut Opus on the customs extraction, since thirty thousand records a month is where any error would be repeated most often, and Haiku on the berth-allocation reasoning, since only a handful of decisions are contested each month.
    Show the answer
24. Question 24Domain 3: Product and Model SelectionSingle answer

    Lindqvist Architecture set every staff member's default to the most capable Claude model in January. By April the monthly bill had tripled, and a usage review showed that most requests were reformatting meeting notes, summarising emails and tidying specifications. Meanwhile the partners' one real quality complaint concerns planning-appeal arguments, where drafts from a mid-tier model had been thin. The managing partner wants the bill back under control without weakening the appeal work. Which model allocation is appropriate?

    Select the best answer.

    AMove all work to the mid-tier model to bring the cost down at once, and rely on more careful prompting by the planning team to close the quality gap that they have seen on appeal arguments.

    BKeep the most capable model for all work, since the tripled bill is still smaller than the fee at risk on a single lost planning appeal and simplicity of policy has real value for a firm of this size.

    CRoute routine tasks such as note reformatting to a faster, lower-cost model, reserve the most capable model for planning-appeal reasoning, and sample the routine outputs afterwards to confirm quality held.

    DLet each staff member choose the model for each task, since they understand their own work best, and review the total cost again at the end of the following quarter before making any firm-wide policy decision.
    Show the answer
25. Question 25Domain 3: Product and Model SelectionSingle answer

    Everline Rail's timetable planner has spent three hours in one conversation building a public consultation document: about sixty exchanges and five uploaded planning PDFs. In the last half hour Claude has twice contradicted decisions the planner agreed earlier and has dropped the document structure they settled on. Roughly two more days of drafting remain before the document goes to the consultation board. Which approach to managing context is most appropriate?

    Select the best answer.

    AContinue in the same conversation but paste the agreed structure and key decisions before each new request, since the history already contains every decision and starting again would lose that record.

    BAsk Claude to write a summary of the agreed decisions, the settled structure and the open questions; check it against the planner's own notes; then start a new conversation seeded with that summary and the current draft.

    CStart a completely fresh conversation with only the five PDFs and rebuild the consultation document from the beginning, since the accumulated history is the source of the contradictions and none of it can be trusted.

    DSwitch to the most capable Claude model within the same conversation, since a stronger model will hold the long history in mind more reliably and stop contradicting the decisions the planner agreed earlier.
    Show the answer
26. Question 26Domain 3: Product and Model SelectionSelect 2

    Trelawney Foods has two tasks this month. First, a thirty-page tender response that four people will draft and revise over two weeks, drawing on the tender documents and the company's house style. Second, before committing to the tender at all, a quick but sourced picture of a regional market the company has never served, so the board can decide within days whether to bid. Which TWO feature selections are appropriate? (Select two.)

    Select exactly 2. 0/2 selected.

    AFor the tender response, work inside a Project holding the tender documents and house style as knowledge, and produce the response as an artifact that is revised across the two weeks and exported when the contributors are done.

    BFor the tender response, use a plain conversation with no uploaded documents, so that the response is written fresh and is not biased toward wording from earlier tenders the company has submitted.

    CFor the market picture, ask Claude from its own general knowledge in a plain conversation, since research mode takes longer and the board only needs enough information to decide whether to bid.

    DFor both tasks, use one shared conversation so that the market findings flow directly into the tender text without anyone having to copy information between two separate places.

    EFor the market picture, use a research-oriented mode that searches and cites current sources, and treat the result as a starting point that the board's analyst verifies before the decision.
    Check answerShow the answer
27. Question 27Domain 3: Product and Model SelectionSingle answer

    Halloran Community College's course lead uses Claude every day for lesson planning. Each session she re-enters the same module structure, the assessment rubric and her preferred tone before any planning can start, which she estimates costs her ten minutes a day. She has access to a memory feature and to Projects, and other staff on the course will soon plan lessons with the same rubric. Which persistence choice fits this situation?

    Select the best answer.

    ARely on the memory feature to carry everything, since it persists automatically across conversations and does not require her to set up a Project or maintain instruction text as the rubric changes over time.

    BPaste the module structure, rubric and tone at the start of every conversation as she does now, since explicit context in the prompt is more reliable than any persistence feature the platform offers her.

    CKeep one conversation open permanently for all lesson planning, so the preferences stay in context throughout the term without needing either the memory feature or a Project to hold them for her.

    DPut the module structure, rubric and tone into a Project's instructions and knowledge so every conversation there starts with them, and let the memory feature hold only her personal working preferences.
    Show the answer
28. Question 28Domain 3: Product and Model SelectionSelect 2

    Fennimore Software's support team wants Claude for two streams under a fixed monthly budget. The first is about five thousand first-line acknowledgement messages a day that follow a template and vary only in the customer's details. The second is roughly two hundred escalated tickets a day, each requiring reasoning across several product documents before a reply can be drafted. Which TWO selection decisions correctly align models with these requirements? (Select two.)

    Select exactly 2. 0/2 selected.

    AUse the most capable model for both streams, since a consistent voice across every customer message matters more than the difference in cost between the model tiers over a month.

    BUse a faster, lower-cost model for the five thousand acknowledgement messages, after checking a sample of its output against the template standard before switching the stream over.

    CUse a more capable model for the two hundred escalated tickets, since multi-document reasoning is where the quality difference between tiers shows and the volume is small enough to afford it.

    DUse the fastest model for both streams, since the budget cap is the binding constraint and any quality shortfall on escalated tickets can be recovered by tuning the prompt more carefully.

    EUse the most capable model for the acknowledgements, since any shortfall there is repeated five thousand times a day, and a faster model for the escalations, whose small volume limits the harm of a weaker reply.
    Check answerShow the answer
29. Question 29Domain 4: Workflow Integration and Solution DesignSingle answer

    Halvorsen Freight, a regional logistics firm, receives about 340 delivery complaints a month, and 60 percent of them concern missed delivery windows. The operations director asks the improvement team to 'use Claude to fix the complaints problem' and wants a proposal for the steering group in two weeks. Nobody has yet established which stage of the delivery process produces the missed windows. Which first use of Claude best serves the requirements analysis?

    Select the best answer.

    AAsk Claude to draft a rollout plan for an assistant that answers every complaint email within one hour, because faster replies are the improvement customers will notice first and the steering group can approve it quickly.

    BGive Claude the 340 complaints and ask it to classify each by delivery stage and apparent cause, returning a table of candidate problems with the complaint count and quoted evidence behind each one.

    CAsk Claude in research mode to summarise published practice on reducing late deliveries and recommend the three measures most often cited, so the steering group sees options with an external track record.

    DAsk Claude to turn the director's description into a requirements document with a target of 30 percent fewer complaints, so the team has a measurable objective before it looks at any individual case.
    Show the answer
30. Question 30Domain 4: Workflow Integration and Solution DesignSelect 2

    Pellworth Council's planning department has a backlog of 1,900 applications. A time study shows officers spend about 40 percent of their week summarising the documents applicants submit and working out which statutory consultations each application triggers. By law, the decision on every application must be made by a qualified planning officer. The department wants to use Claude to cut the backlog without breaching that rule. Which TWO uses of Claude fit the requirement? Select TWO.

    Select exactly 2. 0/2 selected.

    ASummarise each application's submitted documents into a structured brief with page references, so the officer can verify any point against the source before relying on it.

    BIssue the approval or refusal recommendation on the applications Claude assesses as straightforward, so that the qualified officers spend their limited time only on the contested cases.

    CDraft the list of statutory consultations each application appears to trigger, marked for the officer to confirm against the regulations before it is acted on.

    DReply to applicants automatically with the likely outcome based on how similar past applications were decided, to reduce the volume of progress enquiries officers handle.

    EReplace the pre-application advice meeting with a Claude chat interface that answers applicants from the published planning guidance, so they get guidance without taking officer time.
    Check answerShow the answer
31. Question 31Domain 4: Workflow Integration and Solution DesignSingle answer

    Brackenridge University's library service must consolidate twelve reading-list systems into one within a year. The project lead wants Claude to help research the options and draft the migration plan. The plan must respect 40 pages of internal procurement rules, and vendor capabilities have changed substantially in the last two years. A first attempt asked Claude for the plan from general knowledge and produced confident descriptions of features two vendors have since withdrawn. How should the lead use Claude for the planning work?

    Select the best answer.

    AAsk Claude for a generic migration plan template with headings and milestones, and have the librarians research the vendors and fill in the procurement constraints themselves, since Claude's knowledge of vendors has proved unreliable.

    BRun a single long session that covers all twelve systems in one request and asks Claude to select the winning vendor and write the plan around it, so the analysis and the plan stay consistent with each other.

    CProvide the procurement rules and the current vendor documentation as sources, ask Claude to draft the plan with its assumptions listed separately, and have the team verify each vendor claim against the supplied documents.

    DKeep asking Claude from general knowledge but instruct it to flag any feature it is not certain about, so the procurement team can concentrate its checking on the flagged features rather than the whole plan.
    Show the answer
32. Question 32Domain 4: Workflow Integration and Solution DesignSingle answer

    Corvane Apparel, an online clothing retailer, processes returns through an eleven-step workflow. The measured time from a parcel arriving to the refund being issued averages 6.2 days against a target of three. The returns manager asked Claude how to speed up returns and received a generic list of tips about automation and customer communication. The manager has the step-by-step process map with the recorded time and the team responsible for each step. What is the best next use of Claude?

    Select the best answer.

    AAsk Claude to design a new three-day returns process from a blank sheet, deliberately withholding the current eleven steps so that legacy constraints do not limit the redesign and the team sees what an ideal process looks like.

    BAsk Claude to write a business case for a new returns management system, since a generic answer suggests the problem lies in tooling rather than in the process, and a system purchase is where the manager can act.

    CRepeat the same question in research mode so Claude can draw on published benchmarks for apparel returns, and use the benchmark figures to show the team how far the 6.2 days is from industry practice.

    DGive Claude the process map with the time and owner for each step, and ask it to identify where waiting time accumulates and to propose changes with the assumption behind each one stated.
    Show the answer
33. Question 33Domain 4: Workflow Integration and Solution DesignSingle answer

    Marlowe & Tate, a commercial law firm, issues about 500 engagement letters a month. Paralegals produce each first draft by choosing one of 35 templates and filling it from the matter intake form, and a supervising solicitor signs every letter before it is sent. A review found that 4 percent of letters went out with the wrong fee clause, almost always because a paralegal picked an unsuitable template. The firm wants to integrate Claude into this workflow without removing the solicitor's sign-off. Which integration best addresses the measured error?

    Select the best answer.

    AReplace the template library with Claude drafting each letter from the intake form in its own words, so every letter is tailored to the matter, and rely on the retained solicitor sign-off as the control against errors.

    BAdd Claude at the drafting step to select and populate the template from the intake form, showing the fee clause it chose and the intake fields behind the choice, so the paralegal and solicitor can check it.

    CAdd Claude after sending as a weekly scan of issued letters that reports fee-clause mismatches to the practice manager, so the firm can see which templates cause errors and retrain the paralegals who used them.

    DUse Claude to redesign the intake form so that fee terms are captured more precisely, and leave the drafting step unchanged, since better inputs will reduce the number of unsuitable template choices over time.
    Show the answer
34. Question 34Domain 4: Workflow Integration and Solution DesignSelect 2

    Tessaly Foods, a chilled-food manufacturer, wants Claude to draft the weekly non-conformance summary that the plant manager uses to decide whether a production line is stopped for investigation. Each week produces 60 to 90 incident log entries. A quality engineer currently writes the summary by hand in about five hours. The pilot has six weeks and covers one of the company's four plants. Which TWO elements should the pilot design include? Select TWO.

    Select exactly 2. 0/2 selected.

    AA prompt that asks Claude to rate its own confidence in each summary, with summaries rated above 80 percent released to the plant manager without an engineer's check.

    BA baseline measured before the pilot starts, recording how long the manual summary takes and how often it misses or misstates an incident, so the pilot result can be compared with something.

    CExtension of the summary to all four plants after the first week the plant manager accepts a summary unchanged, since early acceptance shows the design is working.

    DA rule that a named quality engineer checks each Claude summary against the source log entries before it reaches the plant manager, for the whole six weeks.

    ERemoval from the incident log of the free-text fields the model finds hardest to summarise, so the weekly output reads more cleanly and the engineer's check goes faster.
    Check answerShow the answer
35. Question 35Domain 4: Workflow Integration and Solution DesignSingle answer

    Ridgemont Community Bank's loan operations team has capacity to introduce Claude into one step of its mortgage workflow this quarter. Two candidates are on the table. The first is drafting the 'documents still required' letter to applicants from the file checklist, about 400 letters a week, each read by a case officer before sending. The second is recommending a decline on marginal applications, a decision that lending policy reserves to a qualified underwriter. Which candidate should go first, and on what grounds?

    Select the best answer.

    AThe decline recommendations, because a marginal application carries the greatest cost per case, and the underwriter remains the final decision maker under policy, so the risk of an incorrect recommendation is fully controlled.

    BBoth candidates at once, because the quarter's pilot budget is fixed and running the two in parallel produces twice the evidence in the same period, letting the bank choose the stronger use case sooner.

    CThe documents-required letters, because the inputs are a stable checklist, an error is caught by the case officer before anything is sent, and 400 letters a week gives a measurable saving.

    DNeither candidate until a full requirements study has been completed and signed off, because introducing Claude anywhere in a regulated lending workflow requires policy approval before any pilot can begin.
    Show the answer
36. Question 36Domain 4: Workflow Integration and Solution DesignSingle answer

    Sunward Insurance piloted Claude-drafted triage summaries on 200 motor injury claims. Handler reading time fell by an average of 22 minutes per claim. In three of the 200 summaries, an injury detail was omitted that would have changed the claim category. The head of claims asks for a single slide for the executive committee, which will decide next week whether to roll the summaries out across 40,000 claims a year. What should the slide say?

    Select the best answer.

    AThe time saving projected across the 40,000 annual claims, with the three omissions described in an appendix, on the grounds that three cases in 200 sits comfortably within the error rate handlers already show.

    BThe three omissions and a recommendation to pause the rollout, on the grounds that a tool that can change a claim category by leaving out an injury cannot be trusted with triage at any scale.

    CThe time saving with a statement that the model's accuracy will improve as it processes more claims, so the committee can see the trajectory as well as the current result.

    DThe measured time saving with its sample size, the three omissions with their consequence, and the review control proposed before rollout, so the committee weighs the value and the limitation together.
    Show the answer
37. Question 37Domain 4: Workflow Integration and Solution DesignSelect 2

    Oakhaven Hospital Group has piloted a Claude Project that drafts patient information leaflets. Across 25 leaflets, first-draft time fell from about four hours to about 40 minutes, and the clinical review found no errors that reached publication, although two drafts contained dosage wording the reviewers corrected. The communications lead must brief department heads before the Project is offered to them. Which TWO statements belong in the briefing? Select TWO.

    Select exactly 2. 0/2 selected.

    AEvery draft must be reviewed by a clinician before publication, because the model can state incorrect medical detail in fluent, confident wording, as the two corrected dosage drafts showed.

    BThe Project checks its own medical facts against the trust's clinical databases before it drafts, so reviewers can concentrate their limited time on tone and readability rather than on accuracy.

    CThe pilot result shows the Project can also be used for consent forms and discharge instructions, since those are patient-facing documents drafted by the same communications teams.

    DBecause the pilot found no clinical error reaching publication in 25 leaflets, departments can move to reviewing a sample of one leaflet in five once rollout begins across the group.

    EFirst-draft time fell from about four hours to about 40 minutes across 25 leaflets, and the clinical review step is unchanged, so the saving is in drafting, not in review.
    Check answerShow the answer
38. Question 38Domain 4: Workflow Integration and Solution DesignSingle answer

    Quillon Software, a subscription analytics vendor, handles 2,400 support tickets a month. An analysis of last quarter shows 55 percent are password resets and billing address changes that an existing self-service portal can already complete, 30 percent are how-to questions answerable from the published help centre, and 15 percent are suspected bugs with no reference material, each needing an engineer to investigate. Agent time per ticket is highest on the bugs. Where should Claude be applied first?

    Select the best answer.

    AThe how-to questions, drafting answers from the help centre content for an agent to check and send, because the source is authoritative and the volume is large enough to measure.

    BThe password resets and billing changes, drafting the replies with Claude, because they are the largest share of tickets and therefore remove the most agent time from the queue.

    CThe suspected bugs, having Claude write a reproduction summary for engineers, because they consume the most agent time per ticket and any saving there is worth more than elsewhere.

    DAll three categories through a single Claude reply assistant, because one integration is cheaper to build and maintain than three targeted ones, and the analysis shows every category involves writing a reply.
    Show the answer
39. Question 39Domain 5: Configuration and Knowledge ManagementSingle answer

    Larkspur Property's marketing team built a Claude Project to draft property listing descriptions. Its knowledge holds 140 past listings, the brand voice guide, and three years of monthly sales reports. The descriptions match the brand voice well, but in a sample of 50 drafts, eight quoted a price taken from an old sales report, and responses have become noticeably slow. Company policy is that a listing description never states a price. What configuration change best fits?

    Select the best answer.

    AKeep all the uploaded knowledge and add a Project instruction to ignore the sales reports when drafting, because the reports may still be useful for other questions the team occasionally asks the Project.

    BRemove the sales reports, keep the voice guide and about 20 exemplary listings, and add an instruction that descriptions carry no price and that any request for one is flagged for the agent.

    CSplit the Project into one Project per property type so that each holds fewer documents, keeping the sales reports in every Project so that drafts stay grounded in the local market context.

    DReplace the uploaded files with a Google Drive connector to the marketing folder, so the Project always reads the current files and the team stops maintaining a separate set of uploads.
    Show the answer
40. Question 40Domain 5: Configuration and Knowledge ManagementSelect 2

    Dunmore Recruitment is setting up a shared Claude Project that 30 consultants will use to write candidate screening summaries against job specifications. The agency needs the summaries to follow one structure regardless of who requests them, and its inclusion policy prohibits any inference about a candidate's protected characteristics. Consultants have complained that summaries from an earlier trial varied in format and occasionally speculated about a candidate's age. Which TWO configuration choices should the Project include? Select TWO.

    Select exactly 2. 0/2 selected.

    AEach consultant's personal formatting preferences appended as extra paragraphs to the shared instructions, so the one Project serves everyone's working style without separate configurations.

    BEvery CV the agency received in the last two years uploaded as knowledge, so the model learns what a strong candidate in each sector looks like before it summarises a new one.

    CProject instructions that fix the summary structure, such as skills match, gaps against the specification and questions for interview, and state that protected characteristics are never inferred or mentioned.

    DAn instruction that Claude ranks the candidates for each role and names the five to interview, so consultants review a shortlist rather than every summary.

    EThe agency's inclusion policy and its standard job-specification template uploaded as Project knowledge, so every consultant's summaries are produced against the same reference documents.
    Check answerShow the answer
41. Question 41Domain 5: Configuration and Knowledge ManagementSingle answer

    A partner at Fennimore Accounting wants to connect Gmail to Claude so it can draft replies to client queries. The connector would give Claude access to the partner's whole mailbox, which holds confidential correspondence for roughly 90 clients. The firm's data policy states that a client's information may be used only for that client's engagement, and the policy owner has not yet considered mailbox connectors. The partner needs to start drafting replies this week. What should the partner do?

    Select the best answer.

    AConnect Gmail and, for each reply, point Claude at the specific thread by subject line, adding an instruction that it must not read or use any other thread in the mailbox while drafting.

    BCreate a Gmail label for each client, connect Gmail, and rely on the labels to keep the model's reading within a single client's correspondence whenever a reply is drafted.

    CAsk the policy owner whether a mailbox-wide connector can be permitted and under what conditions, and until then paste the individual client thread into the chat for drafting.

    DDecline to use Claude for client email at all, since a mailbox containing 90 clients' confidential correspondence can never be reconciled with a policy that separates clients' information.
    Show the answer
42. Question 42Domain 5: Configuration and Knowledge ManagementSingle answer

    Wexcombe Telecom's HR team maintains a Claude Project answering staff questions from 24 uploaded policy documents. The parental leave policy changed in March, and the new document was uploaded in April. In a June check of 40 sample answers, six still quoted the old entitlement, and the Project's knowledge list shows two parental leave files with similar names and no dates. What is the right way to manage the knowledge?

    Select the best answer.

    AAdd the new parental leave entitlement directly into the Project instructions as a stated fact, so that the answer no longer depends on which of the two uploaded files the model happens to retrieve first.

    BUpload the new policy document a second time under a different name, so that the current version outweighs the old one whenever the Project retrieves material on parental leave.

    CAsk HR staff to check the entitlement figure by hand whenever a Project answer mentions parental leave, and correct the figure before it is passed on to the employee who asked.

    DRemove the superseded file, keep only the current version with its effective date in the title, add an instruction to cite the document and date used, and re-run the 40 sample questions.
    Show the answer
43. Question 43Domain 5: Configuration and Knowledge ManagementSingle answer

    Greyfell Energy's customer communications team wrote a 1,800-word Project instruction containing 45 numbered rules. Several conflict: one says every message must be under 120 words, another says every message must carry the full regulatory disclaimer, which alone runs to 90 words. Staff report that outputs vary from request to request and that they append corrections almost every time. Which redesign of the system-level instruction is best?

    Select the best answer.

    ACut the instruction to the rules that govern every output, such as audience, tone, the required disclaimer and prohibited claims, state which rule wins when two conflict, and move task-specific detail into individual requests.

    BKeep all 45 rules, renumber them in priority order, and add a line at the top saying that every rule must be followed exactly, so the model treats each rule as binding and stops choosing between them.

    CReplace the written rules with 20 example messages the team considers exemplary, so the model infers the style and the disclaimer from the examples rather than from rules that can contradict one another.

    DSplit the configuration into two Projects, one for short operational messages and one for regulated messages that carry the disclaimer, each with a shorter instruction set written for that message type alone.
    Show the answer
44. Question 44Domain 5: Configuration and Knowledge ManagementSelect 2

    Kestrel Airways is writing the system-level instruction for a Claude Project that drafts replies to passengers claiming delay compensation under the applicable passenger rights regulation. Agents will paste each claim file into their request. Compensation amounts come from an uploaded schedule. In a trial, four of 30 drafts promised sums above the schedule, and the airline wants drafts to be consistent across 60 agents. Which TWO elements belong in the system-level instruction rather than in each agent's request? Select TWO.

    Select exactly 2. 0/2 selected.

    AThe passenger's name, flight number, booked departure time and the length of the delay for the specific claim being answered, so the model has every fact it needs.

    BThe role, the audience and the rule that every draft names the regulation article relied on and marks any fact it could not find in the claim file as unverified.

    CA summary of the compensation policies of three competing airlines, so that the drafts can explain to passengers how Kestrel's offer compares with what the rest of the market currently pays them.

    DThe rule that compensation figures come only from the uploaded schedule and that a case the schedule does not cover is routed to an agent rather than answered with an estimate.

    EA directive to resolve the greatest possible number of claims without an agent needing to review the draft, so the team's handling time falls and the backlog clears faster.
    Check answerShow the answer
45. Question 45Domain 5: Configuration and Knowledge ManagementSingle answer

    A communications officer at Thornbury School Trust drafts a parent newsletter with Claude every week. Each time she opens a fresh chat, pastes the trust's tone guide and the current term calendar, and describes the structure again, which takes about 15 minutes before drafting starts. The tone guide has not changed in two years; the calendar changes once a term. Which configuration removes the repeated setup with the least ongoing maintenance?

    Select the best answer.

    AContinue with fresh chats but keep the tone guide and calendar in a single text file to paste, since a Project adds configuration overhead that a weekly task of this size does not justify.

    BA Project holding the tone guide and the current term calendar as knowledge, with instructions on the newsletter's structure and audience, updated once a term when the calendar is issued.

    CA Project holding the tone guide and the structure in its instructions, with the term calendar pasted into each weekly request so that the dates the newsletter relies on are never stale.

    DA Project holding three years of past newsletters as knowledge, so the model reproduces the established structure and tone from the archive without the officer maintaining a tone guide.
    Show the answer
46. Question 46Domain 6: Governance, Risk, and Responsible UseSingle answer

    Coleridge Housing Association, which manages 12,000 tenancies, has shortlisted four uses of Claude for its operations team. The association's governance lead must decide which one is unsuitable before any pilot begins. Rent arrears cases can end in enforcement action that affects a household's home, and the association's policy requires such action to follow a documented assessment by a housing officer. Which use should be declined?

    Select the best answer.

    ADrafting the quarterly tenant newsletter from content the communications team has already approved, with a communications officer editing every draft before it is sent.

    BPrioritising the daily repairs queue by predicted urgency from the request text, with the repairs supervisor confirming or changing the order each morning before any job is allocated to a contractor.

    CRanking tenants in arrears for enforcement action using the tone of their correspondence as a signal of willingness to pay, so officers begin with the least cooperative households.

    DTranslating maintenance notices into the six languages most spoken by tenants, with a bilingual member of staff checking each translation before it is posted in the building.
    Show the answer
47. Question 47Domain 6: Governance, Risk, and Responsible UseSelect 2

    Ostrander Pharma Distribution wants Claude to decide which alternative product is substituted when a pharmacy orders a medicine that is out of stock. Orders arrive as structured lines with product codes and quantities, about 5,000 lines a day, more than the two duty pharmacists can each review by hand. Under the company's licence, substitution decisions are reserved to a pharmacist. Which TWO features of this task show it should not be automated with Claude? Select TWO.

    Select exactly 2. 0/2 selected.

    AA wrong substitution may only come to light after a patient has already taken the substituted medicine, so an error is discovered after harm rather than before it.

    BThe daily volume of 5,000 lines exceeds what the two duty pharmacists can review by hand, so the workload itself is beyond what a human process can absorb.

    CThe licence reserves the substitution decision to a pharmacist, so the authority to decide cannot be handed to a tool however accurate its output appears to be.

    DThe two duty pharmacists have not used Claude before and would need training before they could interpret, trust or challenge the substitution choices it produced.

    EThe orders arrive in a structured format with product codes and quantities, so the inputs the tool would work from are stable, machine-readable and easy to validate.
    Check answerShow the answer
48. Question 48Domain 6: Governance, Risk, and Responsible UseSingle answer

    An analyst at Castellan Payroll Services wants to find out why 7 percent of the 60,000 monthly payslips generate a query. The available data has the employee's name, national insurance number, salary, and the free text of each query. Company policy permits personal data in Claude only in anonymised form and requires minimisation. The analysis needs to link each query to the payslip fields that may explain it. What should the analyst upload?

    Select the best answer.

    AThe full dataset unchanged, because Claude is an approved tool at Castellan, the analysis is entirely internal to the payroll team, and removing any fields could hide the very pattern the analyst is looking for.

    BA random sample of 500 records with the names removed but the national insurance numbers kept, because the numbers are needed to match each query to the correct payslip line.

    COnly the free text of the queries with no payslip fields at all, because the text is enough to categorise the queries and leaving out the fields avoids any question of personal data.

    DThe query text and payslip fields with names and national insurance numbers removed and salaries banded, with a pseudonymous key held outside Claude so records can be traced back if needed.
    Show the answer
49. Question 49Domain 6: Governance, Risk, and Responsible UseSingle answer

    An adviser at Coriander Wealth Management writes about twelve suitability reports a month, each taking three hours, and wants Claude to draft them from meeting notes that contain clients' financial circumstances. The regulator requires the firm to retain a record of the advice given for a set period. The firm's policy allows client data only in systems on its approved register; the firm's enterprise Claude deployment is on the register and personal consumer accounts are not. Which approach meets the regulatory and policy requirements?

    Select the best answer.

    AUse the enterprise deployment for drafting, and file the meeting notes and the final report in the firm's advice record system, where the retention requirement is met.

    BUse a personal Claude account with the client's name removed from the notes, since notes without a name fall outside the client data policy and the personal account is quicker to reach.

    CUse the enterprise deployment and treat the chat history as the retained record of advice, since the conversation captures both the notes that went in and the report that came out.

    DAvoid Claude for suitability reports altogether, since a document the regulator can inspect should be written entirely by the adviser who is accountable for the advice it contains.
    Show the answer
50. Question 50Domain 6: Governance, Risk, and Responsible UseSingle answer

    Ellery College's admissions office holds five years of applicant records collected under a privacy notice that describes admissions processing. The marketing team asks to analyse those records in the college's enterprise Claude workspace to target next year's recruitment campaign by school and postcode. The data protection officer has not been consulted, and the marketing lead points out that the data is already held and the workspace is approved. What should happen before any analysis?

    Select the best answer.

    AProceed with the analysis, because the data is already lawfully held by the college and the enterprise workspace is an approved system, so no new processing decision arises from analysing it there.

    BAsk the data protection officer whether the marketing purpose is compatible with the notice applicants received, and if not, update the notice or use aggregated data that cannot identify anyone.

    CPseudonymise the records by replacing applicant names with codes and then proceed, because pseudonymised records are no longer personal data and therefore fall outside the scope of the notice altogether.

    DEmail every applicant from the five years to ask for consent to the marketing analysis, and proceed with the records of those who reply positively within two weeks of the message.
    Show the answer
51. Question 51Domain 6: Governance, Risk, and Responsible UseSingle answer

    A bid manager at Danforth Construction must submit a tender response by Friday. The company's AI policy requires that any AI-generated content in a client-facing document is logged in the AI-use register when work starts and reviewed by a named content owner before release; registering takes about a day to be acknowledged. It is Monday, the draft will take two days, and the review usually takes one. What should the bid manager do?

    Select the best answer.

    ARegister the use today, draft with Claude on Tuesday and Wednesday, and book the named owner's review for Thursday, so the tender goes out on Friday within the policy as written.

    BDraft with Claude immediately and record the use in the register after submission, since the deadline leaves no slack and a late register entry still gives the company an accurate record.

    CWrite the tender without Claude to avoid the register and review steps altogether, since a policy that adds two days of process makes AI slower than manual drafting for urgent work.

    DHave Claude review and improve the finished manual draft rather than generate it, since reviewing is not generating content and therefore falls outside what the register is meant to capture.
    Show the answer
52. Question 52Domain 6: Governance, Risk, and Responsible UseSingle answer

    Averline Airport Authority approved a Claude Project last year for drafting public notices from published documents, and it has produced about 120 notices without incident. The operations team now wants to add a connector that pulls internal incident reports into the same Project so notices can reference recent events. The owner of the incident report system has agreed to grant read access, and the Project owner is willing to make the change today. Under the authority's AI governance standard, what is required before the connector is added?

    Select the best answer.

    ANothing further, because the Project is already approved and a connector is an approved product feature, so adding it is a configuration change within the Project owner's existing remit.

    BThe Project owner's sign-off recorded in the Project's change log, because the owner is accountable for the configuration and the data owner has already agreed to grant the access.

    CA new submission through the AI governance process, because the Project would now process a different class of data for a different purpose, even though the tool and the Project are approved.

    DNotification to the governance team at the next quarterly review, because the connector can be added now on the data owner's approval and reviewed retrospectively alongside the quarter's other configuration changes.
    Show the answer
53. Question 53Domain 6: Governance, Risk, and Responsible UseSelect 2

    Halcyon Hotels' HR team wants to pilot Claude to score the 3,000 applications it receives each recruitment season and produce a shortlist for two recruiters. The scoring would read CVs and covering letters against the job description. The head of HR asks what ethical questions must be answered before the pilot starts, not after. Which TWO belong on that list? Select TWO.

    Select exactly 2. 0/2 selected.

    AWhether Claude, asked directly before the pilot starts, confirms that the scoring criteria drawn from the job description are fair to every group of applicants it will assess.

    BWhether applicants will be told that an automated tool contributes to screening, and how they can ask for a person to review a decision that the tool influenced.

    CWhether the tool can score all 3,000 applications inside the fortnight the season allows, so the two recruiters are not left waiting for a shortlist while interviews are being booked.

    DWhether the scoring could disadvantage applicants through proxies for protected characteristics, such as career gaps or the names of schools, and how the pilot will test for that before anyone is affected.

    EWhether the tool should also draft the rejection emails, so unsuccessful applicants hear back within days rather than after the current three-week wait for a reply.
    Check answerShow the answer
54. Question 54Domain 6: Governance, Risk, and Responsible UseSingle answer

    Beacon Hill Charity's fundraising team uses Claude to draft donor appeal letters that describe the people the charity has helped. The drafts consistently add emotive details, such as a specific illness or a quotation, that do not appear in the consented case files the team supplies. Early tests suggest the more vivid letters raise about 15 percent more per mailing. Which course of action is ethically sound?

    Select the best answer.

    AKeep the added details, because the extra income is spent on the same beneficiaries the letters describe, so the people portrayed gain more from the vivid version than they lose by it.

    BConstrain drafts to the details present in the consented case files, and mark any passage the model has embellished for removal before a letter is sent, whatever the effect on income.

    CKeep the added details but change every beneficiary's name and location, so that no real person is misrepresented and donors are moved by a composite rather than a specific case.

    DStop using Claude for donor letters, because fundraising copy about vulnerable people is too sensitive for any draft that a model has produced to be sent to supporters.
    Show the answer
55. Question 55Domain 7: Troubleshooting and OptimizationSingle answer

    Pryor Dental Group's reception team uses a Claude Project to answer patient questions about treatment prices from an uploaded price list. In a check of 100 answers, 15 gave a price that differs from the list. The team has rewritten the Project instructions three times, each time adding stronger wording about quoting prices exactly, without improvement. Inspection shows the uploaded price list is a scanned image saved as a PDF with no text layer. What is the correct diagnosis and next step?

    Select the best answer.

    AThe failure is at the source: a scanned image gives the model no reliable text to retrieve, so replace it with a text-based price list and re-run the 100 checks before touching the instructions again.

    BThe failure is in the instructions: add a rule that Claude must quote the price list verbatim and must decline to answer any question whose price it cannot find word for word in the uploaded document itself.

    CThe failure is model capability: switch the Project to a higher-tier model, because a stronger model reads scanned documents more accurately and the wording changes have already been exhausted.

    DThe failure is context length: split the scanned price list into several smaller PDFs, one per treatment category, so that each retrieval returns a single page and the model sees less material at once.
    Show the answer
56. Question 56Domain 7: Troubleshooting and OptimizationSelect 2

    Orwell & Bright, a marketing agency, runs a Claude Project that produces campaign briefs. Three account managers report that the same request yields briefs with different section orders, different lengths and a noticeably different tone, and clients have started to notice. The Project has been in use for four months without changes to its configuration. The head of delivery wants the cause found before anything is altered. Which TWO checks should come first? Select TWO.

    Select exactly 2. 0/2 selected.

    AMove the Project to a larger model immediately, since output that varies across users indicates the current model lacks the capability to hold a consistent format.

    BDelete the Project and recreate it with the same files and instructions, since four months of use may have left it in a state that produces inconsistent output.

    CCompare the exact prompts the three managers used to see whether each adds instructions about format or tone that differ from or contradict the Project's instructions.

    DRead the Project instructions to check whether they define the brief's section order, length and tone explicitly or leave those choices to the model on each request.

    EAsk each manager to rate the briefs they received and adopt the highest-rated brief as the template for future requests, so the Project converges on the preferred version.
    Check answerShow the answer
57. Question 57Domain 7: Troubleshooting and OptimizationSingle answer

    Kingsmere Veterinary Group's reception team drafts appointment reminder messages with Claude. After a month, 40 of 1,200 clients complained that the messages were too long to read on a phone. The team lead added the words 'be concise' to the prompt; a fortnight later the complaints continued at the same rate and the messages averaged 110 words. The lead wants the next change to work. What should the team do?

    Select the best answer.

    AAsk Claude why its reminder messages are too long and apply whatever changes it suggests to the prompt, since the model has the clearest view of how it interprets the current instructions.

    BSwitch the Project to a faster model, since concise output is a speed characteristic and the smaller models are designed to produce shorter, more direct replies than the larger ones.

    CAdd three examples of short reminder messages to the prompt and roll the change out to every client immediately, since examples are the strongest form of instruction the team has not yet tried.

    DDefine the target explicitly, such as under 40 words with the appointment details first, test the revised prompt on 30 past reminders, measure length and content against the target, then roll out.
    Show the answer
58. Question 58Domain 7: Troubleshooting and OptimizationSingle answer

    Redfern Medical Supplies' sales team sends Claude-drafted follow-up emails to prospects. Three weeks ago the prompt was changed to add a line offering a first-order discount. Since then the reply rate has fallen from 18 percent to 12 percent. The same three weeks included a public holiday and the launch of a competitor's product. The sales lead wants to revert the prompt today. What is the best response to the results?

    Select the best answer.

    ARevert immediately across all sends, since the fall began when the discount line was added and every week spent testing costs the team replies it will not get back.

    BSend half the follow-ups with the discount line and half without for the next two weeks, then compare reply rates, since the holiday and the competitor launch coincided with the change.

    CKeep the discount line and ask Claude to make the emails more persuasive, since a first-order discount should logically raise the reply rate and the wording rather than the offer itself is the likely fault.

    DAsk the sales team which version of the email they prefer to send and adopt that one, since the people who talk to prospects daily have the best sense of what works.
    Show the answer
59. Question 59Domain 7: Troubleshooting and OptimizationSingle answer

    Northwick Facilities Management introduced Claude to draft replies for its helpdesk, which handles about 900 tickets a week. Every draft is reviewed by an agent before sending. After six weeks the measured end-to-end time per ticket has risen from six minutes to seven, because agents rewrite roughly 30 percent of drafts from scratch, and the rewrites cluster in tickets about contractor invoices. The operations manager wants the workflow to save time without weakening the review. Which optimisation is best?

    Select the best answer.

    AMeasure which ticket categories produce the rewritten drafts, keep Claude drafting for the categories agents accept, and route the invoice tickets straight to agents until the drafting for them is fixed.

    BRemove the agent review for the 70 percent of drafts that go out unchanged, so the time saved on those tickets more than offsets the time spent rewriting the other 30 percent.

    CMove the whole workflow to a more capable model so that the rewrite rate falls across all categories, accepting the higher cost per ticket as the price of a lower end-to-end handling time.

    DAsk agents to send drafts that have only minor problems rather than rewriting them, since a 30 percent rewrite rate suggests the review standard is stricter than the tickets require.
    Show the answer
60. Question 60Domain 7: Troubleshooting and OptimizationSelect 2

    Verity Market Research runs a weekly analysis in which an analyst pastes 40 interview transcripts into a single Claude chat, asks for the main themes, and then asks a series of follow-up questions. By the end, responses are slow, later answers sometimes contradict the themes identified earlier, and the whole exercise takes about three hours. The coding frame and the report template are the same every week. Which TWO changes would make the workflow faster and more reliable? Select TWO.

    Select exactly 2. 0/2 selected.

    AProcess the transcripts in batches of ten against a fixed coding frame, then consolidate the batch summaries in a fresh chat, so no single conversation carries all 40 transcripts.

    BKeep one long chat open for the whole quarter so that Claude retains every transcript and theme it has seen and can compare this week's findings with earlier weeks directly.

    CMove every step of the workflow to the most capable model available, so that the contradictions between early and late answers disappear without changing how the chat is run.

    DStore the coding frame and the report template in a Project so they are not re-pasted each week, and start a new chat for each weekly run.

    EState the three-hour deadline at the top of the prompt and ask Claude to work more quickly, so the model prioritises speed over thoroughness in each of its responses.
    Check answerShow the answer