CCAO-F practice questions

Question 1.1Select one

A marketing associate types “write something about our new product” and is disappointed with the generic result. Which revision best reflects effective prompting for a business task?

• A“Write something better about our new product.”
• B“Write a 150-word LinkedIn post announcing our scheduling app to small-business owners; friendly, professional tone; end with a waitlist call to action.”
• C“Write about our product” repeated three times in a row so the model understands the request is serious and takes it seriously.
• D“You are the world’s greatest copywriter, with decades of award-winning experience. Write something about our product that really captures how great it is.”

Question 1.2Select one

An operations lead must produce a quarterly review covering budget variance, project status, staffing, and next-quarter risks. A single prompt asking for the whole document produces shallow, uneven coverage. Which technique addresses this?

• AAsk for the whole document again, but double the requested word count so every section has enough room to be covered in proper depth
• BAsk Claude to try harder on the weak sections
• CSwitch to bullet points only, so gaps are less visible
• DDecompose it: work through each section as its own request, review each one, then ask for a combined, consistent draft

Question 1.3Select one

Claude’s first draft of a client email is too formal and buries the key request in the final paragraph. What is the most effective way to iterate?

• AGive specific feedback: “Make it warmer and more conversational, and move the request to reschedule into the opening two sentences”
• BSend the identical prompt again without any changes and hope the natural variation between responses produces a better draft
• CStart a brand-new conversation and retype the original prompt
• DAccept the draft as it stands and edit everything manually yourself, since iterating with the model rarely improves anything much

Question 1.4Select one

A project manager wants to use Claude for two different tasks: brainstorming names for an internal program, and drafting the program’s formal charter. How should the prompting strategy differ between the two?

• ABoth tasks should use identical prompts, since prompting principles are universal
• BThe brainstorm should specify exact formatting rules and strict limits; the charter should ask for maximum creativity and give the model complete freedom over structure
• CFor the brainstorm, ask for many varied options and defer judgment; for the charter, provide structure, required sections, tone, and constraints
• DBrainstorming should be avoided, as models cannot generate original ideas

Question 1.5Select one

An analyst pastes a table of regional sales figures and asks Claude to “analyze this.” The response is a bland restatement of the numbers. Which revised request produces genuinely useful analysis?

• A“Take another look at this data and analyze it again, but this time go much deeper and be far more thorough about it.”
• B“Identify the three regions with the largest quarter-over-quarter decline, suggest explanations to investigate, and flag anything unusual — we’re deciding where to focus retention next quarter.”
• C“Analyze this table exactly the way a world-class data scientist would: bring the full weight of statistical expertise, professional intuition, and years of hard-won industry experience to bear on it.”
• D“Summarize this table in half the length.”

Question 1.6Select one

A communications officer has a detailed technical incident report and must brief the executive team. Which prompt element matters most for getting a usable result on the first pass?

• AAsking Claude to keep every technical detail from the report in the briefing so that absolutely nothing is lost in translation
• BRequesting the longest possible output to demonstrate thoroughness
• CAsking for the briefing to be written in a language the executive team does not speak, so that confidentiality of the incident details is fully preserved
• DStating the audience and needs: “for non-technical executives — one page, lead with customer impact and cost, end with the three decisions we need”

Question 1.7Select two

Which TWO elements most improve a prompt for drafting a customer-facing announcement? (Select TWO.)

• ALeaving the request deliberately vague so the model can be creative
• BNaming the audience and the purpose the announcement must achieve
• CWriting the entire prompt in capital letters for emphasis
• DSpecifying format, length, and tone requirements for the deliverable
• EAsking the model to guess what the company probably wants to say

Question 1.8Select one

An operations associate must turn a 60-page vendor audit into a board-ready one-page summary, using Claude throughout. The work involves five activities:

1. Review the extracted findings against the source report for accuracy
2. Upload the report and have Claude extract the key findings for each audit area
3. Define the summary’s audience, purpose, and required structure
4. Have Claude draft the one-page summary from the verified findings
5. Refine the draft’s tone and emphasis through targeted feedback

In which order should the activities be carried out?

• A2 → 3 → 1 → 4 → 5
• B3 → 2 → 1 → 4 → 5
• C3 → 2 → 4 → 1 → 5
• D2 → 4 → 5 → 1 → 3

Question 2.1Select one

Claude produces a market summary that includes a precise statistic — “the sector grew 14.2% in 2025, according to the annual industry report” — which will appear in a board paper. What must happen before this is used?

• ANothing — the precision of the number and the named source together indicate that the claim is reliable enough for the paper
• BAsk Claude whether it is sure, and proceed if it confirms
• CVerify the figure against the actual industry report, or another authoritative source, before it goes in the board paper
• DRound the number to 14% so any error is smaller

Question 2.2Select one

An associate asks Claude to summarize a spreadsheet. The narrative says “headcount grew in every region,” but the accompanying table Claude produced shows one region declining. What does this signal, and what should the associate do?

• AAn internal inconsistency — a reliability warning. Check both claims against the source spreadsheet and correct the output before use
• BTrust the narrative, since prose reflects the model’s more considered overall reading of the data than a mechanically produced table does
• CTrust the table, since numbers are always more reliable than prose
• DAverage the two claims and move on

Question 2.3Select one

HR asks Claude to draft a job advertisement, and the result describes the ideal candidate as a “young, energetic digital native.” What is the appropriate response?

• APublish it — AI-generated text is neutral by definition
• BAsk Claude to make the ad more exciting
• CKeep the wording but add a prominent disclaimer stating that the company is an equal-opportunity employer welcoming applicants of all ages, then publish otherwise unchanged
• DRecognize the age-biased framing, revise the ad to describe skills and competencies, and review AI-drafted HR content for bias as standard practice

Question 2.4Select one

Claude drafts a supplier contract clause covering liability caps. The draft reads well and the associate cannot see anything wrong with it. What determines the next step?

• ASend it to the supplier — if nothing looks wrong, it is fine
• BThe stakes: legally binding language requires review by qualified legal counsel before use, however polished the draft appears
• CAsk Claude to double-check its own clause carefully against common contract pitfalls and rely on its confirmation that the wording is sound
• DShorten the clause so there is less that could be wrong

Question 2.5Select one

An associate wants to fact-check the key claims in a Claude-generated research brief. Which approach constitutes genuine validation?

• AAsking Claude to confirm that its brief is accurate
• BRegenerating the brief a second time and checking whether the new version says substantially the same thing as the first one did
• CChecking each material claim against independent, authoritative sources — official publications or primary data
• DReading the brief carefully for spelling and grammar errors

Question 2.6Select one

Claude summarizes a 30-page policy document into one page. Before circulating it, the associate wants to confirm the summary is complete, not just accurate. What is the practical check?

• AConfirm the summary is exactly one page as requested, since meeting the length requirement is what the instruction actually asked for
• BCompare coverage against the source — check the document’s section headings and key provisions are each represented or consciously omitted
• CAsk Claude whether it left anything out
• DCount the words in the summary and the source document and confirm the compression ratio stays consistent from section to section throughout

Question 2.7Select one

Claude produces an accurate but dense, jargon-heavy analysis. It must now go to a customer advisory panel of non-specialists. What does good output adaptation look like?

• AAsk Claude to rewrite it for the panel — plain language, defined terms, lead with customer impact — then review the rewrite for accuracy
• BSend the original — accuracy is all that matters
• CDelete the technical sections so nothing needs explaining
• DKeep the original dense version fully intact and add a comprehensive glossary appendix defining every technical term for the panel members to consult

Question 2.8Select one

During one working session an associate gets three things from Claude: a quick factual answer, a comparison of four vendors across six criteria, and a polished one-page brief that colleagues will edit and reuse. Which output formats fit?

• AEverything as artifacts, since artifacts are the premium format
• BEverything inline, to keep the conversation in one place
• CThe polished brief inline in the chat, the vendor comparison as a separate chat message, and the quick factual answer as its own artifact
• DThe quick fact inline; the vendor comparison as a structured table; the reusable brief as an artifact to refine and share from

Question 2.9Select one

An associate has two Claude-drafted versions of a sensitive customer apology and must choose one. What is the soundest way to compare them?

• AChoose the longer one, since it covers more
• BDefine what the message must achieve — acknowledge the failure, take responsibility, state the remedy, match company voice — and assess both drafts against those criteria
• CAsk Claude to weigh its own two drafts against each other, explain in detail which it considers the stronger piece of writing and why, and then send whichever draft the model selects
• DMerge both drafts in full so nothing is lost

Question 2.10Select one

After a long research conversation, the useful findings are scattered across dozens of messages mixed with dead ends. The material is needed next week by three colleagues. What should the associate do?

• AForward the entire conversation transcript to the three colleagues so they can read through everything and find the useful parts themselves
• BRely on memory and re-ask Claude next week
• CCurate it: have Claude consolidate the validated findings into one organized document, review and correct it, and share that
• DCopy every message into a spreadsheet row by row

Question 2.11Select one

An associate asks Claude how to configure a setting in their company’s internal, custom-built expense tool. Claude confidently provides step-by-step instructions. What should the associate recognize?

• AClaude cannot know the internals of a private company tool; the confident steps are likely plausible invention and should be checked against internal documentation
• BThe confidence of the instructions indicates Claude has genuine knowledge of the tool
• CClaude can see the associate’s screen and is describing the actual tool
• DInternal business tools follow universal design standards and conventions, so generic step-by-step instructions of this kind are dependable for any company’s software

Question 2.12Select two

Which TWO characteristics of a Claude output most strongly signal that extra verification is needed before use? (Select TWO.)

• AThe response is well formatted, with clear headings, bullet points, and a professional document structure throughout
• BThe response contains specific figures, quotations, or citations that will be relied upon
• CThe response runs longer than two paragraphs, suggesting the model had to work harder to produce it
• DThe response uses a warm, professional tone
• EThe response makes claims about very recent events or specialized facts the associate cannot confirm themselves

Question 2.13Select two

A team is defining its standard validation practice for Claude-assisted work. Which TWO practices belong in it? (Select TWO.)

• ACheck high-stakes factual claims against authoritative external sources before publication
• BAsk Claude to rate its own confidence in each output and treat consistently high self-ratings as sufficient verification
• CRegenerate each output twice and accept it if the versions agree
• DRoute specialized content (legal, medical, financial) to a qualified human reviewer before it is relied upon
• ETreat any output that survives a spelling check as validated

Question 3.1Select one

Every month, an operations associate produces a supplier scorecard using the same 12 reference documents, the same formatting rules, and the same evaluation rubric — currently re-uploading and re-explaining everything each time. Which feature removes this overhead?

• AResearch mode, so Claude finds the documents on the web
• BA longer first message each month containing everything
• CA Claude Project holding the reference documents as knowledge and the rules and rubric as instructions, so every monthly chat starts equipped
• DAsking Claude at the start of each month to remember the documents and rules from last month’s conversation and simply carry on from there

Question 3.2Select one

A strategy associate needs a comprehensive overview of a market: competitor moves, pricing trends, and recent regulatory changes, drawn from many current sources with citations. Which approach fits?

• AA quick chat question with no tools, relying on training knowledge
• BUse the research capability, which searches and synthesizes multiple current sources into a cited report
• CAsk for the answer as a haiku to force conciseness
• DUpload last year’s market report and ask Claude to infer what has probably changed across competitors, pricing, and regulation since it was written

Question 3.3Select one

An associate is developing a client proposal with Claude and expects many rounds of revision — restructuring sections, tightening language, adjusting tone — before it is shared with colleagues. Which way of working fits best?

• AKeep each revision as a new pasted block in the chat and scroll to compare
• BRetype the full proposal into every message when requesting changes
• CAsk for the proposal one paragraph per conversation, in separate chats
• DDevelop the proposal as an artifact, so there is one evolving document to revise, review, and ultimately share

Question 3.4Select one

An associate has two tasks today: (1) a nuanced synthesis of three conflicting analyst reports into a strategy recommendation, and (2) reformatting 200 addresses into a standard layout. How should model choice differ?

• AThe most capable model for the synthesis, where reasoning quality dominates; a fast, low-cost model for the mechanical reformatting
• BUse the fastest model for both, since speed always wins
• CUse the most capable model for both, since quality always wins
• DThe fast model for the synthesis and the most capable model for the address reformatting, so that spending balances out across the two tasks over the day

Question 3.5Select one

A support team drafts hundreds of short, routine replies daily. Response speed and cost-per-reply drive the business case; the replies require no complex reasoning. Which model tier fits, and what should confirm the choice?

• AThe most capable, highest-cost tier — anything customer-facing always justifies maximum capability, regardless of how simple the individual replies are
• BAlternate tiers by day of week to spread the cost
• CA fast, low-cost tier (such as Haiku-class) suited to high-volume work — confirmed by spot-checking a sample of drafts before standardizing
• DWhichever tier was used most recently, for consistency

Question 3.6Select one

Midway through a long planning conversation, several key decisions have been made that will matter for future work over the coming weeks. What is the right way to make those decisions durable?

• ATrust that Claude will remember them whenever asked in future chats
• BPersist them deliberately: have Claude produce a decision summary, verify it, and save it somewhere durable — such as the Project’s knowledge
• CKeep the conversation permanently open in a browser tab and treat that tab as the team’s system of record whenever the decisions need to be consulted
• DRepeat the entire conversation from scratch whenever the decisions are needed

Question 3.7Select two

Which TWO situations most clearly call for setting up a Claude Project rather than using standalone chats? (Select TWO.)

• AA recurring task where every conversation should draw on the same reference documents and instructions
• BA single, one-off question that needs a quick factual answer
• CMaking the model respond noticeably faster to every request by giving it a dedicated space of its own to work within
• DA team wanting consistent behavior — shared context, tone, and rules — across everyone’s conversations
• ERemoving the model’s context limits altogether by giving it somewhere to store an unlimited amount of conversation history

Question 4.1Select one

A business analyst has 40 pages of raw stakeholder interview notes for a new intranet project. How can Claude best support the requirements-analysis stage?

• AHave Claude analyze the notes and decide the project’s final requirements itself, so stakeholders do not need to be consulted or bothered for input a second time
• BHave Claude organize the notes into themes, extract needs and pain points, and draft a structured requirements list for the analyst to validate with stakeholders
• CSkip the notes and ask Claude what intranets usually require
• DUse Claude only to fix the grammar in the notes

Question 4.2Select one

A project manager is kicking off an office relocation and wants help getting from a blank page to a credible plan. What is an effective use of Claude at this stage?

• AAsk Claude for the final, approved project plan and circulate it unedited
• BAvoid using Claude, since planning requires human intuition only
• CAvoid involving Claude until after the plan has been finished by the team, and then use it only to turn the completed plan into a polished set of presentation slides
• DAsk Claude to draft a first-pass plan — workstreams, sequencing, risks, stakeholders — then refine it with knowledge of the actual organization and budget

Question 4.3Select one

An operations team suspects their invoice-approval process is inefficient but cannot articulate where. How can Claude support process optimization here?

• ADescribe the process step by step and ask Claude to identify likely bottlenecks, redundant approvals, and questions to investigate, then verify against real data
• BAsk Claude to approve the invoices itself from now on
• CHave Claude generate a motivational memo about efficiency
• DRetire the current process immediately and replace it wholesale with whatever alternative process Claude describes as industry best practice for invoice approval workflows

Question 4.4Select one

A content team's workflow is: research → outline → first draft → expert review → publish. The first-draft stage is the chronic bottleneck; expert review works well. How should Claude be integrated?

• AReplace the entire workflow — research, outlining, drafting, and the expert review stage included — with a single consolidated Claude step that does it all
• BInsert Claude at the publishing stage, since that is the final step
• CAugment the bottleneck: use Claude to produce structured first drafts from the outlines, keeping research direction and expert review where they are
• DUse Claude to time each stage more precisely

Question 4.5Select one

A team lead wants to integrate Claude into the department’s weekly client-reporting workflow. Five activities are planned:

1. Scale the approach across the whole team as the standard process
2. Map the current workflow and baseline how long each stage takes today
3. Pilot Claude on the chosen stage for several reporting cycles
4. Select the stage where Claude can add the most value at the lowest risk
5. Evaluate the pilot’s results against the baseline measures

In which order should the activities be carried out?

• A4 → 3 → 2 → 5 → 1
• B3 → 2 → 4 → 5 → 1
• C2 → 4 → 3 → 1 → 5
• D2 → 4 → 3 → 5 → 1

Question 4.6Select two

A knowledge worker’s typical week includes: triaging a shared inbox into action categories, negotiating contract terms in supplier meetings, condensing project updates into a weekly status report, approving invoices for payment, and mentoring a junior colleague. Which TWO parts of this week are the strongest candidates for integrating Claude as day-to-day support? (Select TWO.)

• AInbox triage: Claude drafts the categorisation and suggested actions, which the worker reviews and confirms
• BSupplier negotiations: Claude attends the meetings and negotiates the contract terms
• CThe status report: Claude drafts it from the collected project updates, for the worker’s review and edit
• DInvoice approval: Claude reviews and approves the payments autonomously each week, removing the approval bottleneck entirely
• EMentoring: Claude replaces the one-on-one sessions with generated career advice

Question 4.7Select one

After a successful pilot using Claude for proposal drafting, an associate must present the results to skeptical leadership. Which presentation approach is right?

• AReport concrete outcomes — drafting time cut from six hours to two, quality maintained per reviewer scores — alongside honest limitations
• BClaim the tool is flawless to secure the budget quickly
• CPresent only the limitations to appear rigorous and cautious
• DKeep the presentation deliberately free of specific numbers and measurements, so that leadership has nothing concrete to challenge or pick apart

Question 4.8Select one

An associate's team loves their Claude-assisted triage workflow and now wants it to run automatically: connected to the CRM via API, triggered on every new ticket, with no one initiating chats. What should the associate do?

• ABuild the CRM API integration personally over the weekend, using code generated by Claude, so the team doesn’t have to wait for anyone
• BTell the team automation of this kind is impossible
• CRecognize this has crossed into technical implementation — API integration and automated triggering — and escalate to a developer or Architect resource
• DSimulate the automation by hand: personally copy every new ticket from the CRM into Claude throughout the day and paste the triage results back into the CRM

Question 4.9Select two

An associate is designing the pilot for their department's first Claude workflow. Which TWO practices give the pilot the best chance of producing a trustworthy result? (Select TWO.)

• ARoll it out across the entire organization at once, so the pilot sample is as large as it can possibly be
• BChoose a frequent, low-risk task where outputs are easy to check against reality
• CPick the department's highest-stakes process so the pilot proves maximum value
• DDefine success measures — time saved, quality maintained, adoption — before the pilot starts
• EAvoid measuring the current process, to keep the comparison unbiased

Question 4.10Select two

A research associate is preparing a competitive-landscape briefing. Which TWO uses of Claude add genuine value to this workflow? (Select TWO.)

• ASynthesizing a stack of collected articles and reports into organized themes and points of disagreement between sources
• BServing as the final authority on which competitor claims are true
• CProducing a structured comparison of competitors across chosen dimensions as a reviewable starting draft
• DEliminating the need for anyone to check any of the briefing’s claims before it is circulated to the wider leadership group
• EDeciding the company's competitive strategy autonomously from the briefing

Question 5.1Select one

An associate begins every conversation in their reporting Project by pasting the same three paragraphs: the team's tone rules, the report structure, and formatting requirements. What is the correct configuration fix?

• AKeep pasting the paragraphs at the start of every conversation — the repetition helps the model learn the rules permanently over time
• BShorten the three paragraphs so pasting is faster
• CAttach the three paragraphs as a knowledge file in the Project and hope Claude notices and applies it in every conversation
• DMove the standing rules into the Project’s custom instructions, where they apply to every conversation automatically

Question 5.2Select one

Setting up a Project for contract-renewal support, an associate wonders what to upload as knowledge. The team drive holds ~2,000 files; about 15 are actually relevant: current templates, the pricing sheet, and negotiation guidelines. What should be uploaded?

• AThe entire team drive of roughly 2,000 files, so that no piece of information could ever be missing when it turns out to be needed
• BThe 15 relevant, current documents — a curated knowledge base of what the work draws on, maintained as things change
• CNothing — knowledge uploads are unnecessary if prompts are good
• DOnly the oldest documents on the drive, since they carry the most accumulated history of how the work has been done

Question 5.3Select one

A team's status-report documents live in Google Drive and are updated by several colleagues throughout the week. The associate wants Claude to work from the latest versions without manual re-uploading. What is the right approach, and what should be checked first?

• AUse the Google Drive connector so Claude accesses the current files directly — after confirming the connection complies with data and access policies
• BRe-upload every file into the Project each morning before starting work, treating the routine as a useful discipline-building ritual for the whole team
• CAsk colleagues to stop updating the documents so versions stay stable
• DHave Claude guess what probably changed since the last upload

Question 5.4Select one

Which of the following is the most effective set of system-level instructions for a customer-communications Project?

• A“Be consistently helpful and professional, always do a good job, use sound judgment at all times, communicate clearly, and make every customer as happy as possible.”
• B“Never make mistakes.”
• C“You draft customer-service replies for a software company. Tone: warm, plain English, no jargon. Always: acknowledge the issue, state next steps. Never: promise refunds (escalate instead) or exceed 200 words.”
• DThe company’s complete 30-page policy and procedures manual, pasted in as the instructions in full — every policy, exception, and historical amendment since 2015 — so that no conceivable situation is left uncovered

Question 5.5Select one

An associate is configuring a Project and has two assets: (1) rules about how Claude should behave — tone, structure, escalation triggers; and (2) reference material Claude should draw on — the product catalog and policy documents. Where does each belong?

• ABoth pasted into the first message of every chat
• BBehavioral rules in the Project’s custom instructions; reference material as Project knowledge
• CBoth uploaded as knowledge files, keeping instructions empty
• DBehavioral rules uploaded as knowledge documents, with the product catalog written into the custom instructions field instead

Question 5.6Select one

Sales reps report that the quoting Project keeps citing superseded prices. Investigation shows the Project knowledge contains “pricing_2024.xlsx”, while current prices took effect in January. What does proper knowledge maintenance look like?

• AAdd a chat instruction each time: “ignore the old prices you may have seen”
• BUpload the new pricing file into the Project knowledge alongside the old one, so that both versions remain available for reference
• CTell the reps to mentally adjust the quoted figures
• DRemove the outdated file, upload the current pricing document, and put knowledge sources on a review cycle

Question 5.7Select one

An associate is setting up a Claude Project to support the team’s proposal-writing process. Five setup activities are involved:

1. Test the Project with a realistic sample task and review the output
2. Write the Project’s custom instructions: tone, structure, and escalation rules
3. Upload the curated reference documents as Project knowledge
4. Create the Project
5. Refine the instructions and knowledge based on what the test revealed

In which order should the activities be carried out?

• A4 → 1 → 2 → 3 → 5
• B2 → 3 → 4 → 5 → 1
• C4 → 2 → 3 → 1 → 5
• D4 → 2 → 3 → 5 → 1

Question 6.1Select one

Which of the following tasks is the associate right to treat as an appropriate, low-risk use of Claude without additional approvals?

• AProducing the company’s official written response to a regulator’s formal inquiry, and sending it out without any human review
• BAnalyzing an unredacted export of customer identity records from the CRM
• CTurning the associate’s own rough notes from an internal team meeting into organized minutes, reviewed before sharing
• DGenerating a diagnosis for an employee's described medical symptoms

Question 6.2Select one

An HR associate wants Claude's help analyzing turnover patterns using a spreadsheet that includes employee names, salaries, and health-leave records. Organizational policy restricts sharing sensitive personal data. What is the correct approach?

• AUpload it as-is; internal analysis doesn't count as sharing
• BDe-identify first — remove or pseudonymize identifiers, keep only fields the analysis needs — and confirm compliance with the data policy
• CUpload the file as it is, but instruct Claude at the start of the conversation to treat the contents as strictly confidential and promise not to share them
• DAnalyze only the names column, since names alone reveal nothing

Question 6.3Select one

A consultant working with a healthcare client is asked to run patient-feedback records through Claude to find service-improvement themes. Health data is subject to specific regulation. What must happen before any records are processed?

• AProceed — feedback records don't feel like medical records
• BProceed, but delete the conversation afterwards
• CRun a small sample of the records through first, and wait to see whether the client or anyone else raises an objection before processing the full set
• DConfirm the regulatory and contractual position first — engage the client’s privacy or compliance function and process the data only as they approve

Question 6.4Select one

An associate discovers a teammate has been pasting confidential company documents into a personal, free AI chatbot account because “it's the same model anyway.” What is the governance-correct response?

• AExplain the problem, direct the teammate to the sanctioned workspace, and follow the organization’s process for reporting the data exposure
• BIgnore it — the underlying model quality is indeed similar, so the work product is the same wherever the documents happen to be pasted in the end
• CJoin in, because the free account saves budget
• DQuietly delete the teammate's chat history and say nothing

Question 6.5Select one

A marketing lead asks an associate to have Claude write a dozen enthusiastic customer testimonials, each attributed to a fictional customer name, for the company website. How should the associate respond?

• AWrite them — every company polishes its testimonials
• BDecline and explain that fabricated testimonials presented as real customers are deceptive — then offer legitimate alternatives
• CWrite them, but attribute each testimonial to a first name only, so that no complete fictional identity is created and the risk is reduced
• DWrite them and add a tiny footnote reading “illustrative”

Question 6.6Select one

An associate uses Claude to draft a vendor-evaluation report, edits it, and submits it to their director. The report contains an error that later causes a problem. Who is accountable for the error?

• AClaude, since it generated the sentence containing the error
• BAnthropic, as the maker of the tool
• CThe associate who authored and submitted the report — AI assistance does not transfer accountability for one’s work product
• DNo one — errors that originate from AI-assisted work are an accepted cost of the technology and are nobody’s individual fault in particular

Question 6.7Select one

A colleague going through a divorce asks an associate to have Claude determine “what custody arrangement I'm legally entitled to” so they can act on it directly. What should the associate recognize?

• AThis asks the tool to substitute for professional legal advice on a high-stakes personal matter — the determination belongs with a qualified lawyer
• BClaude's answer will be binding, so it should be worded carefully
• CIt's fine as long as the jurisdiction is specified precisely
• DThe request is acceptable, because family law matters are considerably simpler and more standardized than corporate law and well within the tool’s reach

Question 6.8Select two

Before uploading a dataset to Claude for analysis, which TWO checks should an associate perform? (Select TWO.)

• AWhether the file's formatting is attractive
• BWhether the dataset contains personal, confidential, or regulated information requiring removal, de-identification, or special handling
• CWhether the file was created recently enough for the analysis to be interesting and worth the effort of uploading and running in the first place
• DWhether the organization's AI and data policies permit this category of data to be used with this tool, in this workspace
• EWhether the dataset is large enough to look impressive

Question 6.9Select two

Which TWO practices belong in a team's responsible-AI working norms? (Select TWO.)

• ADisclose AI assistance where organizational policy or the context requires it, rather than presenting AI-drafted work as unaided
• BAllow Claude outputs to be sent onward to external parties automatically, without a review step, so that speed of delivery is maximized for the team
• CKeep humans accountable for decisions that affect people — hiring, performance, customer commitments — with AI limited to support
• DTreat an output as verified once the model has reviewed its own work
• EDiscourage colleagues from questioning or second-guessing AI-produced analysis once it has been circulated, so that team efficiency is maintained

Question 7.1Select one

Claude's answers to an associate's competitive-analysis questions keep coming back generic and surface-level. What should the associate examine first?

• AWhether Claude is having a bad day, by waiting and retrying tomorrow
• BThe prompt itself: does it supply the company’s context, the competitors, the decision it feeds, and the criteria that matter?
• CThe subscription tier, since analytical depth of this kind is presumably a feature reserved for the more expensive paid subscription plans
• DThe time of day the questions are sent

Question 7.2Select one

An associate's mega-prompt contains fourteen requirements for a report. The output consistently satisfies most of them but drops two or three — different ones each time. What is the likely cause and fix?

• AThe model is being deliberately selective; add a line demanding total obedience
• BThe requirements are fundamentally impossible to satisfy together in any single document; permanently delete half of them so that the remaining half can actually be met
• CFourteen is fine — the real problem is not saying ‘please’
• DInstruction overload: too many requirements competing in one request — group and prioritize them, or decompose into sequential steps

Question 7.3Select one

In a team's client-reporting Project, every conversation starts with Claude using the wrong tone, and every user manually corrects it mid-chat. What is the right fix?

• AFix it at the source: update the Project’s custom instructions to specify the required tone, so every conversation starts correct
• BKeep correcting each conversation manually as the team already does — it only takes a message or two at the start of every chat to put right
• CAsk the team to accept the default tone to save effort
• DStart every chat by insulting the previous answer to motivate improvement

Question 7.4Select one

Deep into a marathon conversation that has covered several topics, Claude begins conflating earlier drafts, responding slowly, and missing recent corrections. What is happening, and what is the practical response?

• AThe model is permanently degraded and the account should be reset
• BNothing is wrong — conversations behave identically at any length, so the drafts and corrections are all being tracked exactly as they were at the start
• CAccumulated context has grown long and muddled. Have Claude summarize the current state, verify it, and continue in a fresh conversation seeded with it
• DType the corrections in capital letters so they outweigh the older content

Question 7.5Select one

Every week an associate spends fifteen minutes re-explaining the same report requirements to Claude before starting the actual weekly report. What is the right optimization?

• AAccept the fifteen minutes as the cost of quality
• BMake the setup reusable: capture the standing requirements once in a Project’s instructions, so each week starts at the work
• CStop explaining the requirements and accept lower-quality reports
• DRotate the weekly task among the whole team, so that no single person has to bear the same fifteen minutes of re-explaining every single week

Question 7.6Select two

An associate is systematically improving an underperforming prompt. Which TWO practices make the improvement process reliable? (Select TWO.)

• ARewrite the entire prompt completely from scratch on every attempt, so no weak wording survives
• BChange several elements at once to converge faster
• CChange one element at a time, so each change's effect can be attributed
• DSwitch to a different AI product at the first sign of trouble
• ETest each revision against the same example input, so results are compared like-for-like