60 Claude Certified Associate questions with explanations

QUESTION 25 · D2 · OUTPUT EVALUATION

A customer-success director had Claude build a board deck from a subscription export. Slide 2 says annual churn fell to 4.1%, while the appendix table, built from the same export, shows 4.6%. Two other slides use the churn figure in their own calculations. The deck is due on Friday. What should she do?

• A. Keep the slide 2 figure, since board members read the summary most closely, and delete the appendix table to remove the conflict.
• B. Recalculate churn from the export, correct every place the wrong figure appears, and recheck the two slides that build on it.
• C. Ask Claude which of the two figures is correct, and change the other so the whole deck uses the figure Claude chooses.
• D. Average the two figures to 4.35% and use it everywhere, with an appendix note that small rounding differences may exist.

Correct: B. Two values for one metric from the same data mean at least one is wrong. The fix is to go back to the source, establish the correct figure and trace everything that depends on it. Anthropic’s documentation notes that models can produce text inconsistent with the given context, and its help center advises checking important calculations with dedicated software or by hand. A hides the conflict without knowing which figure is right. C lets the model that produced the conflict settle it without evidence. D invents a third number and calls the gap rounding. See Anthropic’s guidance on reducing hallucinations.

In practice, start from the export. Recompute the figure there, then find every place it appears in the deck, including calculations that use it. The same discipline applies to any document written with AI: nine errors in a publicly funded report written with AI show what happens when nobody does this step.

QUESTION 16 · D6 · GOVERNANCE

A federal contractor’s agreement requires its project documents to be processed only in the United States. Its usage-based Enterprise organization has the US-only inference setting turned on. A team member wants to enable a third-party translation connector from the directory to translate those documents inside Claude. What is the accurate assessment?

• A. The US-only setting covers the whole request chain, so the translation service will also process the documents inside the United States.
• B. Connectors are switched off automatically while US-only inference is on, so the team member will not be able to enable the connector.
• C. The setting covers only where Claude’s models run. The connector’s service processes data on its own infrastructure, so check its location first.
• D. Data sent through connectors is encrypted in transit, so the place where the translation service processes the documents does not matter.

Correct: C. US-only inference keeps Claude’s model processing in the United States. It does not control connectors or third-party services, which process data on their own infrastructure under their own terms, possibly outside the US. The team should confirm where the translation service runs before sending contract documents through it. A assumes the setting reaches services it does not cover. B is false: the setting governs where inference runs, not whether connectors can be used. D confuses protection in transit with the place of processing, and the contract restricts the place. See Anthropic’s help article on US-only inference.

Read a data setting the way a lawyer reads a clause: what exactly does it cover? Here it covers model inference and nothing else. Before you turn on any connector for restricted documents, find out who processes the data, where and under which terms. Write the answer down for whoever approves the change.

QUESTION 13 · D1 · PROMPTING

A facilities coordinator asked Claude to “make a spreadsheet for meeting-room bookings” and received an .xlsx file. Three problems stop her team from using it. Nothing shows when two bookings for the same room overlap. Dates are in MM/DD/YYYY while her office uses DD/MM/YYYY. Both buildings share one list, although each building manager maintains only their own rooms. Which THREE additions to her request address these problems?

• A. Begin the request with “You are a senior Excel consultant with twenty years of experience in facilities management.”
• B. State that two bookings for the same room with overlapping times must be flagged automatically in a Conflict column.
• C. Add “Make it professional, comprehensive and as good as possible, with no mistakes” at the end of the request.
• D. Specify that every date in the file uses the DD/MM/YYYY format her office uses in its calendars and reports.
• E. Ask for a separate sheet for each building, listing only that building’s rooms, so each manager edits only their own.
• F. Ask Claude to add a chart of how often each room was booked per week, to help with planning room capacity.

Correct: B, D, E. Each states a requirement the first request left Claude to guess, and each matches one problem: a rule for flagging overlaps, the office’s date format, one sheet per building. Claude’s help center on file creation advises describing the structure, content and formatting you want. A gives a persona but no requirements, so Claude still knows nothing about overlaps, dates or buildings. C asks for quality without saying what quality means for this file. F adds a chart, which may help later but fixes none of the three problems. See Anthropic’s help article on creating files with Claude.

List what went wrong with the first file, then turn each problem into one sentence of the new request. An addition that fixes none of the problems does not belong in this request. Then open the new file and test it: enter two overlapping bookings and a date such as 03/04, and see what the spreadsheet does.

Question 1 · D7 · Troubleshooting and Optimization

Scenario: Proposal review rounds

A bid team on a Team plan keeps a 25-page proposal as a Claude Docs document. Seven of its nine sections are approved. After each review round, the coordinator pastes every reviewer comment into the conversation and asks Claude to regenerate the whole proposal, and each round then costs a day of re-checking because wording and figures in approved sections keep shifting. Only sections 4 and 7 still need changes. What is the most efficient way to run the next round?

• A. Regenerate the whole proposal once more, adding "Do not change any approved section" at the top of the request
• B. Highlight the passages to change in sections 4 and 7 and use Edit with Claude or an @Claude comment, then re-check only those
• C. Start a new chat each round, paste in the latest proposal and every comment, and ask Claude for a complete clean version
• D. Ask each reviewer to regenerate their own section in a separate chat, then merge the returned sections into the doc by hand

Question 2 · D4 · Workflow Integration and Solution Design

Scenario: Employment-law questions

A department head asks whether her team can simply ask Claude for the current employment-law rules in three countries and stop checking with the legal team. The associate needs to explain what Claude can and cannot be relied on for. Which two points should the explanation include?

• A. Claude can write confident text that is wrong, so answers with legal consequences need checking against authoritative sources
• B. When Claude cites a source, the claim generally matches it, so answers that come with citations can skip the legal team's review
• C. The most capable model at its highest effort level is accurate enough on law that expert review is no longer needed
• D. Claude's built-in knowledge has a cutoff, so current rules need web search or Research, and the cited sources still need reading
• E. Claude's legal errors mostly come from vague questions, so a precise, detailed prompt can take the place of the legal review

Question 3 · D3 · Product and Model Selection

Scenario: Terminology agreed in another project

A consultant on a Pro plan, with chat search turned on, spent several chats in her "Client A" project agreeing the client's terminology rules with Claude. In a chat inside her separate "Client A phase 2" project, she asks Claude which terminology they agreed, and Claude cannot find the earlier discussion. What explains this, and what should she do?

• A. Chat search in a project covers only that project's chats; ask in the original project and add the rules to the new project's knowledge
• B. Memory is off by default on Pro plans; turning it on in Settings will make the rules available in every project and chat she uses from now on
• C. Chat search runs only while Research is turned on; she should switch Research on in the new project and ask the same question again
• D. Paste the full transcripts of the earlier chats into each new chat in the phase 2 project whenever she needs the agreed terminology

Question 4 · D6 · Governance, Risk, and Responsible Use

Scenario: Thumbs-down on a settlement

On her company's Team plan, a legal operations analyst got a poor answer in a chat that contains a confidential settlement agreement. She wants to click thumbs down and paste the key clause into the feedback comment. Company policy allows confidential client documents in Claude only on terms that keep them out of model training. What should she know before submitting?

• A. Feedback is fine, because on Team plans inputs and outputs are never used for model training, and that rule covers feedback as well
• B. Feedback stores the whole conversation for up to five years and may be used for training, so she should not submit it on this chat
• C. Feedback is fine if she deletes the chat straight afterwards, because deleting a chat also removes the copy kept with her feedback
• D. Feedback is fine if she leaves the comment box empty, because only the written comment is sent to Anthropic, not the conversation

Question 5 · D6 · Governance, Risk, and Responsible Use

Scenario: Customer database connector

A marketing coordinator wants Claude to answer questions from the company's customer database, which sits on the internal network behind the firewall. A developer friend offers to publish a custom connector for it on a public server over the weekend. Company policy requires an IT security review before any new integration touches customer data. What should she do?

• A. Accept the offer, because a custom connector can only return the records that her own database account is already permitted to see
• B. Raise it with IT through the integration review, noting that custom connectors are reached from Anthropic's cloud, not from her laptop
• C. Ask for a weekly export of the full customer table and upload it to a project shared with the marketing team, which avoids a connector
• D. Accept the offer, but set the chat's tool access mode to On demand, so the connector loads only when a question needs customer data

Question 6 · D1 · Prompting and Task Execution

Scenario: Workshop plan rework

A team lead asked Claude to "plan a workshop on giving feedback" without saying how long it would be, and Claude produced a three-day program. Over five follow-up messages she wrote "shorter", "remove day two", "it's only one session" and similar corrections. The latest plan is 90 minutes long but still refers to a role-play from "yesterday" and a homework review. The workshop is a single 90-minute session for 12 newly promoted managers. What should she do next?

• A. Start a new chat with a complete brief: one 90-minute session, 12 newly promoted managers and the goals, and let Claude plan from it
• B. Reply once more listing every leftover reference to earlier days, and ask Claude to remove each one from the current version of the plan
• C. Start a new chat and paste her original first message unchanged, since a fresh conversation will not carry over the three-day design
• D. Ask Claude to explain why it keeps referring to earlier days, so that she understands the cause before sending any more corrections

Question 7 · D4 · Workflow Integration and Solution Design

Scenario: Customer-facing order assistant

A customer service manager wants a Claude-powered assistant on the public website. It should answer order questions for about 8,000 customers a day, look up each logged-in customer's orders in the company's order system, and issue refunds below a set amount. She asks an associate to deliver it. What is the associate's most appropriate role?

• A. Build an artifact that calls Claude, publish it, and link it from the website so that customers can start chatting with it straight away
• B. Set up a scheduled task that reads the support inbox every hour and issues the refunds below the threshold without approval
• C. Put the order policies in a project and share the project with customers, so they can ask their questions in project chats
• D. Use Claude to draft the requirements, sample dialogues and escalation rules, then hand the build to developers or an architect

Question 8 · D1 · Prompting and Task Execution

Scenario: Renewal email prompt

A customer-success lead keeps the prompt that drafts contract-renewal emails in her team's project instructions. Last week she changed the tone guidance, added a 120-word limit and pasted in two example emails, all at the same time. Some drafts improved, but drafts for accounts with open support tickets now leave out the ticket status, which they must mention. She does not know which change caused this. What is the best way to iterate from here?

• A. Keep all three changes, add the line "Always mention open support tickets", and roll the prompt out to the team without more testing
• B. Undo all three changes and go back to the old prompt for good, since the previous version never left out the ticket status
• C. Build a small fixed set of test accounts, including some with open tickets, then reapply the changes one at a time and compare drafts
• D. Ask Claude which of the three changes made it leave out the ticket status, and keep or drop each change based on its explanation

Question 9 · D5 · Configuration and Knowledge Management

Scenario: HR policy Q&A project

An HR operations lead runs a project that answers staff policy questions. Its knowledge holds the current Employee Handbook and 60 past HR email replies, added so that answers match the team's friendly tone. Several answers have quoted leave rules from those old emails that the handbook has since changed. The team wants to keep the friendly tone. Which change to the project's configuration fits best?

• A. In the instructions, say policy answers come only from the Employee Handbook, citing the section, that the emails are tone examples only, and that gaps go to HR
• B. Delete the 60 email replies from project knowledge and add 'Write in a friendly tone' to the instructions, leaving only the Employee Handbook to answer from
• C. Upload a second copy of the Employee Handbook to project knowledge, so that its rules outweigh the old emails whenever Claude looks up a leave rule for staff
• D. Ask every employee to begin each question with 'According to the handbook', so that Claude knows which of its sources to rely on when it answers them

Question 10 · D5 · Configuration and Knowledge Management

Scenario: Tenant repair replies

A property management team uses a project to draft replies to tenants' repair requests. Its instructions read: "NEVER give a repair date. DON'T use jargon. Keep it short." Replies now say only "We cannot provide timing information", which upsets tenants, and some still use terms like "HVAC". Many tenants read English as a second language, and the contractor, not the team, sets visit times. Which rewrite of the instructions is most likely to fix this?

• A. Keep the three rules and add "Be empathetic and friendly" at the end, so that Claude softens the tone of the replies it already writes for tenants
• B. Replace the rules with "Use your judgment on dates, wording and length, as an experienced coordinator would", to give Claude more freedom
• C. Keep the rules as they are and ask the coordinators to rewrite the opening and closing lines of each reply by hand before it goes to the tenant
• D. Ask for warm, plain-English replies under 120 words, and explain that the contractor sets visit times, so replies say a coordinator will confirm one

Question 11 · D7 · Troubleshooting and Optimization

Scenario: Recurring NDA corrections

A legal operations team uses a project to draft first-pass summaries of incoming NDAs. Memory is turned off for their Team organization. For six weeks, the reviewing lawyer has made the same three corrections on most drafts: governing law is missing, the term is given in months rather than years, and whether the NDA is mutual is not stated up front. Each paralegal fixes these by replying in their own chat. What is the best adjustment?

• A. Keep correcting in each chat, since Claude takes in these corrections and will stop making the same three mistakes over time
• B. Switch the project to the most capable model, since the repeated errors show that the current one cannot handle legal documents well
• C. Ask the lawyer to stop writing notes and fix the drafts directly, so that the paralegals spend less time on each correction
• D. Add the three rules and an approved sample summary to the project instructions, test on recent NDAs, and track the corrections

Question 12 · D7 · Troubleshooting and Optimization

Scenario: Five deliverables, one prompt

A project manager uses a single prompt to turn a 40-page steering-committee pack into five deliverables: an executive summary, a risk log, an email to sponsors, a slide outline and an FAQ. The summary is strong, but the risk log misses several risks named in the pack, and the FAQ is thin and repeats the summary. What is the most likely cause, and the best fix?

• A. The pack is too long for Claude to read in full; split it into five parts and send each part to Claude as a separate message
• B. One prompt is carrying too many goals; run each deliverable as its own focused step, checking each output before using it next
• C. The model lacks the capability for this task; switch to the most capable model available and then run exactly the same prompt again
• D. Claude needs stronger emphasis; add BE THOROUGH IN EVERY SECTION in capital letters at the very top of the same prompt

Question 13 · D1 · Prompting and Task Execution

Scenario: Room-booking spreadsheet

A facilities coordinator asked Claude to "make a spreadsheet for meeting-room bookings" and received an .xlsx file. Three problems stop her team from using it: nothing shows when two bookings for the same room overlap, dates are in MM/DD/YYYY while her office uses DD/MM/YYYY, and both buildings share one list although each building manager maintains only their own rooms. Which THREE additions to her request address these problems?

• A. Begin the request with "You are a senior Excel consultant with twenty years of experience in facilities management"
• B. State that two bookings for the same room with overlapping times must be flagged automatically in a Conflict column
• C. Add "Make it professional, comprehensive and as good as possible, with no mistakes" at the end of the request
• D. Specify that every date in the file uses the DD/MM/YYYY format her office uses in its calendars and reports
• E. Ask for a separate sheet for each building, listing only that building's rooms, so each manager edits only their own
• F. Ask Claude to add a chart of how often each room was booked per week, to help with planning room capacity

Worked through step by step above

Question 14 · D6 · Governance, Risk, and Responsible Use

Scenario: Wrong penalty in supplier email

A procurement officer asked Claude, through the Gmail connector, to draft a reply to a supplier, and approved the send when Claude asked. The email stated a late-delivery penalty of 2% per week; the signed contract says 0.5%. She notices the next morning, before the supplier has replied. Which two actions reflect responsible use?

• A. Write to the supplier promptly, correct the figure to the contract's 0.5% and take responsibility for the error as the sender
• B. Tell the supplier that the figure came from an AI tool, so the company is not answerable for what the email said about penalties
• C. Switch to the most capable model at maximum effort, so that checking contract figures before approving a send is no longer needed
• D. Before approving future drafts that quote contract terms, check every figure in them against the signed contract itself
• E. Submit thumbs-down feedback on the chat and wait for Anthropic's reply before contacting the supplier about the figure

Question 15 · D6 · Governance, Risk, and Responsible Use

Scenario: Store chat assistant persona

An online retailer's customer-service lead is writing the persona instructions for a Claude-powered chat assistant on the store's website, set up through the store's helpdesk vendor. Her draft names it "Emma, Customer Care" and says: "If asked, say you are part of our human support team." What should she change?

• A. Keep the draft, and add a sentence to the website's terms of service saying that some customer chats may be handled by automated tools
• B. Keep the draft, but have a human agent review a 10% sample of chats each week to confirm that Emma's answers are accurate and polite
• C. Have the assistant say it is an AI at the start of every chat, and remove the instruction to claim it is part of the human support team
• D. Replace the name Emma with a neutral "Support" label, but keep the instruction to say the team is human if a customer asks about it

Question 16 · D6 · Governance, Risk, and Responsible Use

Scenario: US-only processing clause

A federal contractor's agreement requires its project documents to be processed only in the United States. Its usage-based Enterprise organization has the US-only inference setting turned on. A team member wants to enable a third-party translation connector from the directory to translate those documents inside Claude. What is the accurate assessment?

• A. The US-only setting covers the whole request chain, so the translation service will also process the documents inside the United States
• B. Connectors are switched off automatically while US-only inference is on, so the team member will not be able to enable the connector
• C. The setting covers only where Claude's models run; the connector's service processes data on its own infrastructure, so check its location first
• D. Data sent through connectors is encrypted in transit, so the place where the translation service processes the documents does not matter

Worked through step by step above

Question 17 · D5 · Configuration and Knowledge Management

Scenario: Sales pitch project access

On a Team plan, a sales enablement manager is setting up a project with the approved pitch guide, pricing rules and answer instructions for 40 account executives. Two enablement specialists will keep the content current. A similar project drifted last year after several executives edited its instructions. How should she configure access to the new project?

• A. Share it with all 42 people as Can edit, and add a line to the project instructions asking the executives not to change the instructions or the knowledge
• B. Keep the project private, set everything up in one chat with the instructions pasted in, and share that chat's link with each of the 40 executives
• C. Create a separate private copy of the project for each executive, so that any change one of them makes stays inside that person's own copy only
• D. Share it with the 40 executives as Can view and with the two specialists as Can edit, so only the specialists can change instructions or knowledge

Question 18 · D5 · Configuration and Knowledge Management

Scenario: Friday media summary

Each Friday a communications officer asks Claude, with the Gmail and Google Drive connectors turned on, for a media summary. The inputs are PDF press-clipping reports that arrive as Gmail attachments, and a Google Doc into which colleagues paste screenshots of social-media charts. The summaries cite the right emails but none of the clipping figures, and they miss the chart data entirely. Which two changes address these gaps?

• A. Switch the chat's tool access mode to Always available, so that both connectors are loaded at the start and can read every file in full
• B. Save the PDF attachments to Drive or upload them to the chat, because the Gmail connector reads attachment metadata, not attachment content
• C. Ask an Owner to set the Gmail and Drive tools to Always allow, so that Claude can open attachments and images without asking for approval
• D. Have colleagues add the key chart figures as text or a table in the Google Doc, because images embedded in Drive documents are not processed
• E. Turn on extended thinking for the Friday summary, so that Claude reasons longer over the emails and the Google Doc and works out the figures

Question 19 · D2 · Output Evaluation and Validation

Scenario: Forklift battery bulletin

A communications specialist at a distribution company used Claude to draft a safety bulletin on charging forklift batteries, to be posted in three warehouses. The steps look plausible and are well written, but she has no electrical or safety training. The company employs a certified site safety officer. What is the right next step?

• A. Ask Claude to add a reference to a relevant safety standard for each step, and post the bulletin once every step has one
• B. Have the safety officer review and approve the technical steps before posting, while she checks clarity and layout herself
• C. Search the web to confirm two or three of the steps, and post the bulletin if those steps match what she finds online
• D. Regenerate the bulletin with the most capable model available and post it, since that model is least likely to make errors

Question 20 · D6 · Governance, Risk, and Responsible Use

Scenario: Bank's six planned uses

A retail bank's operations team lists six planned uses of Claude. Under the High-Risk Use Case Requirements in Anthropic's Usage Policy, which three require a qualified professional to review Claude's output before it is finalized or sent?

• A. Summarizing last week's internal IT incident tickets into a one-page update for the operations manager
• B. Drafting the letters that tell loan applicants whether they were approved, and the reasons for the decision
• C. Recommending to individual customers, through the bank's app, which investment fund suits their savings goals
• D. Writing an internal staff wellbeing newsletter with general tips on sleep, stress and regular exercise
• E. Translating the internal guide on resetting staff passwords into Spanish for the bank's branch employees
• F. Answering customers' questions about whether their loan contract lets them repay early without a penalty

Question 21 · D5 · Configuration and Knowledge Management

Scenario: Diverging procedure copies

A finance team of nine analysts at a 300-person company on a Team plan keeps one project per client. Each project's instructions hold a copy of the team's 7,000-character procedure for writing monthly variance commentary, and the copies have drifted into nine different versions. Code execution is enabled, and the organization allows skill sharing. What is the best way to maintain the procedure from now on?

• A. Ask an Owner to paste the procedure into the organization instructions, so that every conversation in the company follows a single version of it
• B. Keep a copy in each project, add a version number at the top, and have each project owner paste in the newest version at the start of every month
• C. Package the procedure as one skill shared with the nine analysts, so an update reaches all of them, and remove the copies from the project instructions
• D. Have each analyst paste the procedure into their own Instructions for Claude, so it applies in all of their projects without any separate copies

Question 22 · D2 · Output Evaluation and Validation

Scenario: Kickoff plan and a quick question

A project coordinator at a company on a Team plan has two requests for Claude. First, she wants to know the difference between a RACI and a RASCI matrix, for her own understanding. Second, she needs a project kickoff plan that six colleagues will edit over the next month and that will later go to the client as a Word file. Which output formats fit best?

• A. An inline answer for the RACI question, and a Claude Docs document for the plan, edited by colleagues and exported to Word
• B. A Claude Docs document for the RACI question, and an inline answer for the plan that she pastes into an email to her colleagues
• C. An inline answer for the RACI question, and a PDF file of the plan that colleagues download and mark up before it goes out
• D. An inline answer for the RACI question, and a Claude Docs document for the plan that she later shares with the client by link

Question 23 · D2 · Output Evaluation and Validation

Scenario: Five outputs, which to verify

An operations coordinator used Claude for five outputs this week. Which TWO need checking by a qualified person or independent verification before they are used?

• A. Five subject-line ideas for the internal staff newsletter, from which the newsletter editor will choose one
• B. A summary of a new data-retention law that will set how long the company keeps its customer records
• C. A tidier version of her own meeting notes, which she uses only as her personal to-do list for the week
• D. A first-draft agenda for next week's team meeting, which she will adjust as the discussion goes along
• E. Import duty amounts Claude calculated from 40 supplier invoices, to be entered on a customs declaration

Question 24 · D3 · Product and Model Selection

Scenario: Outdated travel-rules document

A travel coordinator on a Team plan maintains an internal Google Doc listing entry and visa requirements for the nine countries staff visit most. It was last updated two years ago, and she must produce a corrected version with a source for every change. Web search is enabled for her organization and her Google Drive is connected. Which approach fits the task best?

• A. Use Research with her Drive connected to check each country against current official sources, citing every change, then open the citations
• B. Ask in a regular chat with web tools turned off, because Claude's training data already covers visa rules for common destinations
• C. Upload the doc to a Project and ask Claude to flag every entry that looks out of date, based on what it already knows about each country
• D. Use extended thinking without web access, so Claude reasons carefully through which of the nine countries' rules have most likely changed

Question 25 · D2 · Output Evaluation and Validation

Scenario: Churn figures disagree

A customer-success director had Claude build a board deck from a subscription export. Slide 2 says annual churn fell to 4.1%, while the appendix table, built from the same export, shows 4.6%. Two other slides use the churn figure in their own calculations. The deck is due on Friday. What should she do?

• A. Keep the slide 2 figure, since board members read the summary most closely, and delete the appendix table to remove the conflict
• B. Recalculate churn from the export, correct every place the wrong figure appears, and recheck the two slides that build on it
• C. Ask Claude which of the two figures is correct, and change the other so that the whole deck uses the figure Claude chooses
• D. Average the two figures to 4.35% and use it everywhere, with an appendix note that small rounding differences may exist

Worked through step by step above

Question 26 · D4 · Workflow Integration and Solution Design

Scenario: Leave calculator artifact

An HR coordinator with no developer support built an artifact that estimates each employee's remaining annual leave. It gives correct results for full-time staff, but HR testing found three wrong results: two for part-time staff and one for someone who joined mid-year. She wants colleagues to start using it next week. What should she do next?

• A. Share it now with a note asking colleagues to report errors, since many users will find the remaining edge cases faster
• B. Give Claude the three failing cases, the correct results and the policy rule, ask for a fix, then re-test them and the passing cases
• C. Start a new chat with her original prompt and generate the whole tool again, in the hope that the new version handles these cases
• D. Reply that there are still errors and ask Claude to check its calculations carefully, then share the tool once Claude says it is fixed

Question 27 · D2 · Output Evaluation and Validation

Scenario: Parent pick-up notice

A school office manager asked Claude for two versions of a notice about new pick-up times starting 3 November. Version 1 is 420 words, formal and complete: the new times, a gate change, the 24 October deadline to register for late pick-up, and a contact. Version 2 is 120 words, warm and easy to read on a phone, but it leaves out the registration deadline. Most parents read school messages on their phones, and many speak English as an additional language. What should she send?

• A. Version 1, because it is the only complete version, and parents who find it too long can call the office with questions
• B. Version 2 as it stands, because parents prefer short messages and the deadline can follow later in a separate reminder
• C. Both versions in one message, so that each parent can choose the one that suits their reading preference and needs
• D. Version 2 revised to include the registration deadline, with every date and time then checked against the source details

Question 28 · D4 · Workflow Integration and Solution Design

Scenario: Expense-approval requirements

A business analyst has notes from 14 stakeholder interviews about a new expense-approval process. Finance wants every claim approved within 48 hours, regional managers want to approve small amounts themselves, and internal audit needs a full approval trail. The steering group will sign off a requirements list next week. How should she use Claude to analyze the requirements?

• A. Ask Claude for the requirements of a typical expense-approval process from its general knowledge, attaching the notes as background
• B. Have Claude settle conflicting requests by keeping whatever most interviewees asked for, so the list goes to sign-off with no open issues
• C. Ask Claude for a polished requirements document first, and check it against the notes only if a stakeholder raises an objection
• D. Have Claude list each requirement with its stakeholder and quote, merge duplicates, and flag conflicts and gaps for her to resolve

Question 29 · D2 · Output Evaluation and Validation

Scenario: Outage notice for customers

An IT service manager has an internal post-incident review of a six-hour outage of the customer portal. It names the engineers involved, lists internal server names, mentions a dispute with a vendor that is still being negotiated, and includes unconfirmed theories about the root cause. She needs a notice for business customers. Which approach best adapts the review for this audience?

• A. Ask Claude to simplify the language of the review so non-technical readers can follow it, then send that version to customers
• B. Ask Claude to make the notice as reassuring as possible and to promise customers that an outage like this will not happen again
• C. Brief Claude on the audience and what to include or omit, then check the draft against the review for accuracy and confidentiality
• D. Send the full review unchanged, because complete transparency about the incident builds the most trust with business customers

Question 30 · D2 · Output Evaluation and Validation

Scenario: Analyst quote in a deck

A marketing manager asked Claude for talking points on customer-loyalty trends, and Claude answered from its own knowledge without searching the web. One point quotes a named analyst at a well-known research firm, with a report title and year. She searched the firm's website and the wider web and found neither the analyst nor the quote. The deck goes to a client tomorrow. What should she do?

• A. Ask Claude for the report's web address, and keep the quote if the link opens a page on the research firm's website
• B. Keep the point but soften the attribution to "as one analyst reportedly said", which avoids naming a specific person
• C. Ask Claude whether the quote is genuine, and keep it in the deck if Claude confirms the wording and the report title
• D. Treat the quote as a likely fabrication and leave it out unless she can find it in the original report herself

Question 31 · D2 · Output Evaluation and Validation

Scenario: FAQ from support emails

A customer-service lead has exported 300 customer emails from the last quarter and wants Claude's help to build an FAQ page. Many emails ask the same things in different words, and in a few cases agents gave customers different answers to the same question. Which approach best organizes this material?

• A. Have Claude group the questions into themes with counts, merge duplicates, and flag conflicting answers for the product owner
• B. Have Claude turn every email into its own question-and-answer entry, in the order received, so nothing customers asked is lost
• C. Use only the 25 most recent emails, since they reflect current issues best, and have Claude write an FAQ entry for each one
• D. Have Claude write a standard FAQ for this type of product from its general knowledge, skipping the emails to save time

Question 32 · D3 · Product and Model Selection

Scenario: One-off proposal summary

A facilities manager has 45 minutes before a meeting and needs a one-page summary of a single 20-page cleaning-services proposal: price, scope, contract length and anything unusual. She will not need the proposal again after the meeting. A colleague suggests setting up a Project first, because Project content is cached and uses less of the usage limit. What is the best way to do this task?

• A. Create a Project, add the proposal to its knowledge, write project instructions for the summary, then ask for it in a project chat
• B. Turn on Research so Claude can check the supplier's claims across the web before it writes the one-page summary for the meeting
• C. Attach the proposal to a regular chat and ask for the one-page summary, naming the four points and the meeting it is for
• D. Upload the proposal once and ask Claude to save it to memory, so the summary can be requested again from any later chat

Question 33 · D4 · Workflow Integration and Solution Design

Scenario: Vendor shortlist research

A procurement lead must shortlist three of eight project-management tool vendors for a 200-person company. Single sign-on, EU data residency and a per-seat budget are must-haves. He plans to use Claude's Research feature. Which approach is best?

• A. Run Research on the best project-management tools first, then adopt the criteria that appear most often in the report
• B. Ask Claude in a regular chat with web search off to rank all eight vendors, since it already knows this software market
• C. Set the must-haves and weighted criteria first, ask Research for a cited comparison against them, then confirm deal-breakers
• D. Ask Research to choose the single best vendor for the company, so that the shortlisting and scoring steps can be skipped entirely

Question 34 · D7 · Troubleshooting and Optimization

Scenario: Support-reply pilot results

A support team piloted Claude-drafted first replies for four weeks, with agents reviewing every draft. For password resets and delivery questions, 85% of drafts went out with light edits. For billing disputes, agents rewrote 70% of drafts from scratch, mostly because the drafts misread the account history pasted into the chat. What is the best adjustment based on these results?

• A. Keep Claude drafting resets and delivery replies, pause drafting for billing disputes, investigate why, and measure again
• B. End the pilot for all categories, because a 70% rewrite rate shows that Claude cannot be relied on for customer support work
• C. Keep every category as it is, add "take extra care with billing" to the prompt, and stop tracking the rewrite rates by category
• D. Remove agent review for resets and delivery replies to save time, and leave the billing dispute drafting exactly as it is now

Question 35 · D6 · Governance, Risk, and Responsible Use

Scenario: Confidential deal project

At a consultancy on a Team plan, an associate creates a project for a client's confidential acquisition and uploads the target company's financial statements. When creating it, she chose to share it with her broader organization so that colleagues could find it. Only the five-person deal team is covered by the client's confidentiality agreement. What should she do?

• A. Leave it shared with the organization, because chats inside a project shared with everyone stay private to each user unless they are shared
• B. Rename the project "CONFIDENTIAL: deal team only" and add an instruction telling Claude not to discuss the files with anyone else
• C. Leave it shared, but move the financial statements out of project knowledge and attach them to her own chat, where others cannot open them
• D. Change general access to "Only people invited" and invite the five deal-team members, so only they can see and use the project's knowledge

Question 36 · D1 · Prompting and Task Execution

Scenario: Per-diem policy update

An HR manager must update the company travel policy to new government per-diem rates, supplied as a PDF, and then produce a staff FAQ and an announcement email. Finance must approve the policy wording before anything goes to staff. She asked Claude for all three in one message, and the FAQ gave a daily meal allowance of 45 euros while the updated policy said 52 euros. What is the best way to structure the work?

• A. Keep one request but add "make sure all three documents are fully consistent with each other", then regenerate all three together
• B. Run three chats in parallel, one per document, each given the old policy and the rates PDF, so that each document gets full attention
• C. Draft the FAQ and the email first, since staff will read those, then update the policy text so that it matches what the FAQ says
• D. Have Claude tabulate old and new rates, revise the policy, get finance sign-off, then draft the FAQ and email from the approved policy

Question 37 · D5 · Configuration and Knowledge Management

Scenario: Three rules, three layers

On a Team plan, an operations analyst who is a Member, not an Owner, collects three rules at a workshop: (1) no response anywhere in the company should include a customer's account number; (2) chats in the grant-writing project must follow the funder's five-section structure; (3) she personally wants every answer she gets to open with a three-bullet summary. Where should each rule live?

• A. Put all three rules in the grant-writing project's instructions, since that project is where the workshop's work happens and all its members can see them
• B. Ask an Owner to add (1) to the organization instructions, put (2) in the grant-writing project's instructions, and put (3) in her own Instructions for Claude
• C. Put (1) and (3) in her own Instructions for Claude because they apply everywhere, and add (2) to the grant-writing project's knowledge as a document
• D. Ask an Owner to put all three rules in the organization instructions, because organization instructions take precedence over personal instructions

Question 38 · D1 · Prompting and Task Execution

Scenario: Warehouse site decision

A logistics manager must recommend one of two sites for a new warehouse. She has lease costs, drive times to the 20 largest customers, labor availability figures and flood-risk ratings for both. She already leans toward Site A, and her first idea is to ask Claude to "write a memo explaining why Site A is the better choice". Which approach best fits this task?

• A. Keep the request as it is, because a persuasive memo is what leadership will read, and Claude can still mention Site A's downsides
• B. Ask Claude to brainstorm twenty factors that could matter for any warehouse location, then choose the five that seem most important
• C. Give Claude both sites' data, ask it to propose weighted criteria for her to confirm, then compare the sites and state its assumptions
• D. Ask Claude to role-play the two landlords negotiating with each other, so that each site's strengths come out in the conversation

Question 39 · D6 · Governance, Risk, and Responsible Use

Scenario: Sales rep mood scoring

A sales director asks an operations associate to use Claude, through the company's Slack connector, to read each sales rep's messages every week and give each rep a 1-to-10 score for frustration and for likelihood of quitting. Only managers would see the scores, and the reps would not be told. What is the most appropriate response?

• A. Decline: inferring named employees' emotions from their messages is a prohibited use; suggest raising retention concerns with reps directly
• B. Run the scoring as asked, because the Slack connector only reaches channels the associate can already see, so Claude accesses no data beyond existing permissions
• C. Proceed, but restrict the analysis to public channels and report the scores in low, medium and high bands rather than as exact numbers per rep
• D. Add a clause on AI monitoring to the employee handbook first, then run the scoring once the reps have acknowledged it at their next annual review

Question 40 · D2 · Output Evaluation and Validation

Scenario: Complaint reply tone

A support team lead uses a Claude project to draft replies to customer complaints, which agents edit and send. One agent remarks that the drafts seem curter for customers with certain surnames. Before changing anything, the team lead wants to find out whether the drafts actually differ by customer name. What is the best check?

• A. Run identical complaints through the project, changing only the customer's name, and compare the drafts for tone and offers
• B. Ask Claude in the project whether its replies treat customers differently depending on their names, and act on its answer
• C. Add "Treat every customer equally and without bias" to the project instructions and consider the problem resolved
• D. Pull ten recent sent replies at random and read them for tone, since a random sample gives an unbiased picture

Question 41 · D1 · Prompting and Task Execution

Scenario: Field technician handbook

A training manager asked Claude in one message to "write an onboarding handbook for new field technicians" and attached 14 documents, including safety procedures, the van inventory list and the job-ticket guide. The result was a generic 6-page text whose sections did not follow the technicians' first week. The finished handbook should run to about 30 pages, and the safety lead must approve its structure before any content is written. How should she restructure the work?

• A. Send the same request again with "make it at least 30 pages long and very detailed" added, so Claude has room for all 14 documents
• B. Ask Claude for an outline that follows the first week, get the safety lead's approval, then draft one section at a time from its sources
• C. Open a separate chat for each of the 14 documents, have Claude turn each one into a chapter, and combine the chapters into one file
• D. Use Research so that Claude can model the handbook on published onboarding guides from other field-service companies before writing

Question 42 · D2 · Output Evaluation and Validation

Scenario: Lease cost check

A finance coordinator asked Claude in chat for the total cost of a five-year office lease: a base rent that rises 3% each year, the first four months rent-free, and a one-off fit-out charge. Claude gave a total and a short explanation. The figure will go on a contract approval form that the CFO signs. Which validation technique fits best?

• A. Ask Claude to double-check its arithmetic in the same chat, and accept the total if the second answer matches the first
• B. Have Claude build the calculation as an .xlsx file with visible formulas, then check the inputs and one year by hand
• C. Compare the total with what the company paid for its previous office lease, and accept it if the two are reasonably close
• D. Ask the same question in a new chat and, if the two totals differ, use the larger one to stay on the safe side

Question 43 · D4 · Workflow Integration and Solution Design

Scenario: Quarterly business reviews

Customer success managers spend about six hours preparing each quarterly business review deck. They pull usage figures from exported reports, summarize the quarter's support tickets, and write recommendations for the client's next quarter. The team lead wants Claude to cut preparation time without weakening the client relationship. Which redesign fits best?

• A. Claude drafts the usage and ticket slides from the exports in the team template; the manager checks figures and writes the recommendations
• B. Claude builds the complete deck, recommendations included, and sends it to the client automatically on the day before each meeting
• C. Keep the current process unchanged and use Claude only to proofread each finished deck for spelling and formatting before it goes to the client
• D. Claude writes the recommendations for each client, while the managers keep assembling the usage figures and ticket summaries by hand

Question 44 · D1 · Prompting and Task Execution

Scenario: Funder form description

A nonprofit communications officer must write a 300-word description of her organization's after-school literacy program for a funder's application form, which asks for the program's aims, who it serves and last year's results. She has last year's program report as a PDF. She used Research, and Claude returned a well-cited but generic text built on national literacy statistics and other charities' programs. Which approach fits this task better?

• A. Run Research again with her organization's name in the query, so that Claude finds published material about her own program
• B. Ask Claude to brainstorm thirty possible descriptions in different styles, then pick the one that sounds most like her organization
• C. Treat it as drafting: give Claude the program report, the funder's three questions, the word limit and a past description she likes
• D. Ask Claude to write the description from its general knowledge of literacy programs, leaving gaps that she fills with her own figures

Question 45 · D2 · Output Evaluation and Validation

Scenario: E-scooter rules report

A city policy officer used Research to produce a report on e-scooter rental rules in five named cities, as they stand in 2026, for a council committee. The report is 11 pages long and has 46 citations. What is the best first step in evaluating it?

• A. Treat the report as verified, because Research cites a source for every claim and the committee can follow those citations
• B. Check it against her brief: all five cities covered, each rule current for 2026, and key claims confirmed in the cited sources
• C. Ask Claude in the same chat whether the report is complete and accurate, and fix only the sections where it admits a weakness
• D. Ask Research for a one-page executive summary of the report, then check that the summary and the full report agree

Question 46 · D4 · Workflow Integration and Solution Design

Scenario: Choosing a first pilot

A marketing director asks an associate to recommend one use case for a six-week Claude pilot that must show a measurable result. Four candidates are on the list. Which one is the best first pilot?

• A. Letting Claude choose which customer segments receive discount offers each week, and then sending those offers automatically
• B. Drafting the weekly product-update email from approved release notes, edited by the content lead, measured on editing time
• C. Rewriting the company's brand strategy, measured by how inspiring the leadership team finds the final version of it
• D. Answering journalists' press inquiries directly on the company's behalf, without review, to cut the response time

Question 47 · D3 · Product and Model Selection

Scenario: Cross-workstream dependency check

A program manager has spent the morning in one chat on Sonnet, making routine edits to status notes. The chat also holds the plans for six workstreams she pasted earlier. She now needs an analysis of hidden dependencies and scheduling conflicts across all six before a launch decision; a miss would be costly, and she can wait several minutes for the answer. She already raised Sonnet's effort level, and the result still missed two conflicts her team had spotted. What should she do?

• A. Start a new chat on Opus and paste all six plans again, because a conversation keeps the model it was started with
• B. Switch the chat to Haiku and run the analysis several times, since the fastest model can repeat it cheaply and catch more
• C. Stay on Sonnet and split the work into six new chats, one per workstream, so each analysis is smaller and easier to check
• D. Switch this chat to Opus for the dependency analysis, then switch back to Sonnet when she returns to the routine edits

Question 48 · D4 · Workflow Integration and Solution Design

Scenario: Proposal-time claim for the board

After a 20-minute demo in which Claude drafted a sales proposal in five minutes, a sales VP wants to tell the board that Claude will cut proposal time by 80% from next quarter. Proposals currently take about two days, including legal and pricing review. What should the associate recommend?

• A. Present the demo's five-minute draft against the two-day average as the expected saving, because the board wants a clear number
• B. Tell the board Claude should not be used for proposals at all, because it can produce errors that would reach clients in the documents
• C. Quote Anthropic's published benchmark results instead of the demo, since benchmarks are a more objective measure of the speed gain
• D. Pilot Claude on real proposals, measure total time with review and rework against the two-day baseline, then report what was found

Question 49 · D3 · Product and Model Selection

Scenario: Model guide for a team

An operations manager is writing a one-page internal guide to help her team choose between the Haiku, Sonnet and Opus model families in Claude. Which two statements are accurate enough to include in the guide?

• A. Sonnet is an older family kept only for compatibility, so new work should start on either Haiku or Opus instead
• B. Haiku is the fastest, lowest-cost family, a good fit for simple high-volume work such as quick lookups and summaries
• C. Haiku suits the longest documents best, because a faster model can read more text in a single message than the others
• D. A chat stays on the model it started with, so pick Opus for any chat that might later become complex
• E. Opus suits complex, multi-step work where quality matters more than speed, and it uses more of a plan's usage

Question 50 · D5 · Configuration and Knowledge Management

Scenario: Stale risk log export

A project manager on a Pro plan keeps a private project for weekly steering-committee updates. In March she uploaded a PDF export of the project's risk log, which is a Google Doc the team updates daily. In May, Claude's drafts still list risks that were closed weeks ago. Her Google Drive connector is already connected. What is the best fix?

• A. Export a fresh PDF before each weekly update and attach it to that chat, keeping the March export in project knowledge as a historical reference
• B. Add a line to the project instructions telling Claude to leave out closed risks, and to treat the March export as current unless told otherwise
• C. Remove the March export from project knowledge and add the risk log's Google Doc from Drive, so the project uses the latest version as it changes
• D. Ask Claude at the start of each chat to search the web for the latest version of the risk log, so that every draft works from the newest copy

Question 51 · D2 · Output Evaluation and Validation

Scenario: Regional KPI summary

An operations analyst asked Claude to turn a 14-tab KPI workbook into a one-page summary for the COO. The brief: all six regions, four KPIs for each, and the month-over-month change for every figure. The summary reads well. Which TWO checks best establish whether it is accurate and complete before it is sent?

• A. Ask Claude to rate its confidence in each figure from 1 to 10, and look again only at the figures it rates below 8
• B. Have a colleague read it to confirm that the tone and length suit the COO and that the headings are clear
• C. Recompute a sample of the month-over-month changes directly from the workbook and compare them with the summary
• D. Ask Claude to regenerate the summary, then send whichever of the two versions reads more clearly and concisely
• E. Tick the summary off against the brief to confirm all six regions and four KPIs appear, each with a change figure

Question 52 · D1 · Prompting and Task Execution

Scenario: Translated service bulletins

A city transit agency's communications officer uses Claude to draft weekly service-change bulletins in English. Each bulletin then goes through the agency's machine-translation tool to produce Spanish, Vietnamese and Tagalog versions, which bilingual staff check before posting. The staff keep rewriting the same kinds of phrases, such as "hit the ground running" and "the 14 bus is out of action", because they translate badly. Her prompt already says "Don't use idioms." Which change to the prompt is most likely to fix the problem in future bulletins?

• A. Explain that every bulletin goes through machine translation into three languages, so it needs literal, plain wording that translates cleanly
• B. Make the existing rule stronger by writing it in capitals, repeating it at the end of the prompt and adding that it is very important
• C. Attach the list of 40 idioms the translators have flagged so far and tell Claude that none of those expressions may appear again
• D. Open the prompt with "You are an award-winning public transport copywriter" so that Claude writes in a clearer, more professional register

Question 53 · D3 · Product and Model Selection

Scenario: Peer annual-report analysis

A research coordinator on a Team plan must answer questions across 28 long annual reports from peer organizations over the next month. When she attaches all of them to one chat, Claude tells her the conversation has reached its length limit. What is the best way to keep working with all 28 reports?

• A. Ask the plan owner to upgrade the team's plan, which raises the context window so all 28 reports fit in one chat
• B. Paste the reports' text into the message instead of attaching the files, so that the same 28 reports take up less space in the chat
• C. Add the reports to a project's knowledge with clear, descriptive file names, and ask focused questions in project chats
• D. Resend all 28 reports as the first message of a new chat, relying on automatic context management to summarize them

Question 54 · D2 · Output Evaluation and Validation

Scenario: Survey statistic for a report

A communications consultant used web search in Claude to find a statistic on flexible work. Claude's answer says "62% of employees would change jobs for more flexibility" and cites a business-news blog. The figure will headline a report the client is publishing. Which validation step is most appropriate?

• A. Search for the figure elsewhere and accept it if at least three other websites repeat the same percentage in a similar sentence
• B. Ask Claude whether the blog is a reliable source, and keep the figure if Claude describes the blog as trustworthy and established
• C. Keep the figure and credit the blog, because Claude's answer included a citation that readers of the report can follow themselves
• D. Trace the figure to the original survey and check its exact question, sample, country and year before it goes into the report

Question 55 · D7 · Troubleshooting and Optimization

Scenario: Daily product-title rewrites

A merchandising assistant rewrites about 40 product titles every weekday. She sends each title as a separate message in one chat she has used since July, restating the same five style rules in every message. She runs out of usage by mid-afternoon, although the work itself is simple. Which two changes would cut her usage most directly without lowering quality?

• A. Turn on Research so that Claude can check how competitors title similar products before it rewrites each of her titles
• B. Raise the effort level to Max so that fewer of the rewritten titles need a second attempt later in the working day
• C. Group each day's titles into one or a few messages, stating the five style rules once rather than in every message
• D. Start a new chat for each day's batch instead of continuing the single conversation she has been using since July
• E. Attach the full 30-page brand guide to every message so that Claude never misses one of the rules while rewriting

Question 56 · D4 · Workflow Integration and Solution Design

Scenario: Interview scheduling

A recruiting coordinator spends hours each week finding interview slots for four interviewers and emailing candidates. Her organization has enabled the Google Calendar and Gmail connectors, she has authenticated both, and the interviewers share their calendars with her. Last month a wrong interview time sent to a candidate embarrassed the team. How should she bring Claude into this workflow?

• A. Ask an owner to let Gmail sends run without approval, so that Claude can schedule interviews and email candidates end to end
• B. Copy each interviewer's weekly calendar into a chat every Monday and ask Claude to suggest the free slots for the coming week
• C. Have Claude find mutual availability in the shared calendars and draft each candidate email, and approve every send herself
• D. Move availability out of the calendars into a shared project where interviewers type their free times for Claude to match

Question 57 · D3 · Product and Model Selection

Scenario: Monthly regulatory digest costs

A regulatory affairs team runs a monthly workflow in a Project: Claude, on Opus, turns about 30 regulator updates into a two-page digest. A reviewer signs off each digest against a six-point checklist, and every digest for the past four months has passed. The team now runs short of usage near month-end and wants to spend less on the digest without risking its accuracy. What should they do?

• A. Move the digest to Sonnet from next month, since it handles everyday analysis well and the reviewer's checklist will catch anything it misses
• B. Re-run last month's updates on a lighter setup, such as lower effort or Sonnet, score them against the checklist, and switch only if they pass
• C. Leave the workflow unchanged, because lowering the effort level or changing the model of a workflow that already passes review will reduce quality
• D. Ask Claude which model and effort level it recommends for this digest, and apply its recommendation from the next monthly run onward

Question 58 · D4 · Workflow Integration and Solution Design

Scenario: Office move planning

An office manager must move 60 staff to a new floor. The old lease ends on 31 March, IT needs three weeks on the new floor before anyone moves in, furniture arrives five weeks after it is ordered, and no more than 20 people can move in one weekend. How should she use Claude to build the plan?

• A. Give Claude the dates, lead times and limits, ask for a dependency-ordered plan with latest start dates and its assumptions, then check them
• B. Ask Claude for a standard office-move checklist, then add the lease date, lead times and weekend limits herself once the checklist is finished
• C. Turn on Research to find how other companies scheduled their office moves, then copy the most common timeline into her own project plan
• D. Ask Claude for the final move schedule and send it to all staff straight away, so that every team has as much time as possible to prepare

Question 59 · D7 · Troubleshooting and Optimization

Scenario: Announcements copying one example

To set the house style, a communications officer added one past announcement, about a new parking policy, as an example in her prompt. Now every new announcement, whether about a training course or a system outage, opens with "Starting Monday", uses the same three sections even when one does not apply, and sometimes mentions parking. What is the best fix?

• A. Make the parking example longer and label it mandatory, because Claude is clearly not following it closely enough yet
• B. Remove the example and add a list of phrases and sections Claude must never use, so the outputs stop repeating them
• C. Add a few varied example announcements and state which features are fixed and which should change with each topic
• D. Start a new chat and paste the same prompt with the same parking example, because the current chat has grown too long

Question 60 · D6 · Governance, Risk, and Responsible Use

Scenario: Read-only email policy

A company's AI policy says Claude may search and summarize staff email but must never send, reply to or forward messages. The company is on a Team plan with the Gmail connector enabled. As the Owner responsible for the setup, which control best enforces the policy?

• A. In the Gmail connector's tool permissions, set send, reply and forward to Blocked, which applies org-wide and members cannot override
• B. Add "Never send, reply to or forward email" to the organization instructions, which take precedence over each member's own instructions
• C. Rely on the Gmail connector's default behavior, in which Claude asks each member for approval before it sends, replies to or forwards a message
• D. Disable the Gmail connector for the whole organization, since a connector that can send email cannot be limited to reading staff mail only

PREP PLAN

1. 1. First pass in study mode. No time pressure. Read every explanation, including the ones you got right. The point is to see why the other options fail.
2. 2. A full sitting in exam mode. One session, 120 minutes, no breaks. You practise working within a time limit.
3. 3. Study the domain breakdown. Go back to Anthropic’s help center for your two weakest domains. Output evaluation carries the most weight.
4. 4. Explain the decision without the key. A high score after several attempts may mean you remember answers. Change one fact in a scenario, such as the plan or the setting, and check whether the same option still wins.