CCAO-F exam map

This note maps the Claude Certified Associate – Foundations exam onto this hive. For each domain it shows how much of the exam it carries, where it is taught here, and how the mock papers sample it. Every exam fact below comes from the official exam guide, version 1.0 (effective July 2026). Where this hive makes its own choice, such as how many seats a domain gets on a mock paper, the note says so.

Who can take the exam

Anthropic's certification FAQ says: "Certification is currently available only to organizations in the Claude Partner Network." Registration also needs a partner email address on a recognized company domain. Personal email addresses do not work. Check that your organization is in the network before you plan around a date.

Registration goes through the Anthropic Partner Academy, and Pearson VUE delivers the exam. The guide states that there are no mandatory prerequisites and that no software-development or API experience is needed.

The exam at a glance

The guide sets the fee and the retake waiting periods, and both can change. Check them in the current guide rather than here.

The seven domains

The guide calls its weights "the approximate proportion of scored items drawn from each domain". The mock-paper seats are this hive's own allocation. They are the guide's weights applied to 60 items and rounded by largest remainder, so the seats always add up to 60.

Objective count does not follow weight. Domain 2 has six objectives at 21%, while Domain 6 has four at 15%. A heavy domain can still spread thinly across its objectives, so study the objectives as well as the percentages.

Where each domain is taught

Every domain has:

• a narrated lecture deck for each topic, with a companion quiz;
• flashcards;
• a study note;
• a no-code practice exercise with a reference solution.

How to use the hive

1. Learn a topic. Watch the deck, then take its quiz straight away. The quiz tests only what that deck teaches.
2. Consolidate a domain. Read the domain's study note and review its flashcards.
3. Apply it. Work through the domain's practice exercise in the Claude app, then compare your answers with the reference solution. Steps that need an administrator or a paid plan are written as plans, so every exercise works on any plan.
4. Sit a mock paper once you have covered all seven domains. Take it timed, with no notes.
5. Review by domain. Note which domains your missed questions come from. Go back to the decks for your weakest domain, then sit the next paper.

The four mock papers

• Each paper has 60 questions, is timed at 120 minutes, and includes multiple-response items that state how many to select, as the real exam does.
• The four papers share no questions. Every paper asks about all 30 objectives, in the seat counts shown above.
• The exam-bank questions are separate from the quiz questions. Scoring well on a paper means applying what you learned to new situations, not recognizing a quiz item you have seen before.
• A mock result is the percentage you answered correctly. It is not the exam's scaled score. The guide reports a pass or fail on a scaled score of 100–1,000 with a cut score of 720. It also reports the percentage correct in each domain, and that per-domain breakdown is the part a mock can imitate.

Staying current

Claude's products change faster than the exam. This hive teaches judgement that lasts: when to verify, which kind of model suits which task, and what not to upload. It avoids prices, plan limits and version numbers. Where a feature depends on your plan or your organization's settings, the material says so. If the Claude app on your screen differs from a deck, trust the help center page the deck cites, which is linked from the hive's sources.

Configuration and Knowledge Management — practice exercise

Difficulty: beginner · Estimated duration: 45–60 minutes

You will do this exercise in the Claude app. No code is involved. Steps that need a Team or Enterprise plan, or an owner’s permission (sharing, organization instructions, connectors an owner must enable, and memory controls where memory is off), are written as plans rather than carried out, so every stage works on any plan. One help-center page lists project instructions as a paid-plan feature; if your plan does not offer them, write yours down instead. The scenario is invented for practice: you coordinate volunteers for the Harbour Arts Festival, a three-day community event.

Stage 1 — Build and share the project

Skills: CCAOF-U5.T1.LO1.S1, CCAOF-U5.T1.LO1.S2, CCAOF-U5.T1.LO1.S3 Minutes: 12

1. Create a project called “Festival volunteers”. In its description, write one sentence for people who will see it.

2. Save this briefing as a text file and add it to the project’s files:

   Volunteers sign in at the Market Square tent 30 minutes before each shift. Shifts run 9:00–13:00 and 13:00–17:00. Lost children are taken to the Lantern Tent. First-aid questions go to the St John volunteer at the Quay Stage.

3. Sort these six lines into project instructions or ask in a chat, then write the project instructions:

   • a. You help the Harbour Arts Festival’s volunteer team.
   • b. Write in plain, friendly English.
   • c. Draft the rota email for Saturday’s early shift.
   • d. Never include volunteers’ phone numbers in drafts.
   • e. Summarize Friday’s feedback forms.
   • f. Use the festival’s site names: Quay Stage, Lantern Tent, Market Square.

4. Open a chat in the project and ask where volunteers sign in. Then, in that chat, tell Claude that the sign-in time is now 45 minutes before each shift. Open a second chat and ask for the sign-in time. Record what it says, whether memory and chat search are on for your account, and why.

5. Plan only (Team or Enterprise): two shift leads should keep the briefing up to date, and twelve stewards should only ask questions. Write down the visibility and permission you would give each group.

Stage 2 — Files, retrieval and connectors

Skills: CCAOF-U5.T1.LO2.S1, CCAOF-U5.T1.LO2.S2, CCAOF-U5.T1.LO2.S3 Minutes: 12

1. The festival’s site map is a picture inside a Word document. Write down what Claude will read from that document, and add the three most important locations to the project as text instead.
2. The festival office sends you three files named Doc1.docx (the volunteer code of conduct), scan0032.pdf (last year’s road-closure notice) and notes FINAL (2).txt (first-aid procedures). Write a clear new name for each.
3. Write the question you would ask to get Claude to answer from the first-aid procedures in particular.
4. Plan only: you want Claude to draft replies to volunteers’ emails in Gmail, and to read the rota in Google Drive. Write the steps from “not connected” to “reply sent”, noting any that need an owner, and say which of your chats should have each connector switched on.

Stage 3 — Write and test instructions

Skills: CCAOF-U5.T2.LO3.S1, CCAOF-U5.T2.LO3.S2, CCAOF-U5.T2.LO3.S3 Minutes: 12

1. Decide the level for each rule (account, project or organization):
   • a. “I always want UK spelling.”
   • b. “Use the festival’s site names.”
   • c. “Every staff member’s chats must use the festival’s safeguarding wording.”
2. Rewrite “Be nice to volunteers” as two concrete instructions, and add them to the project instructions.
3. Start a new chat in the project and test the instructions with three different kinds of request: an email, a short summary and a list. Record whether each instruction was followed. If your plan does not offer project instructions, paste them at the start of a new chat and test there.
4. Plan only: the festival’s organization instructions say “Sign off emails ‘The Festival Team’”. Your personal instructions say “Sign off with my first name” and “Use numbered lists for rotas”. Predict how Claude signs off a rota email, and whether the rota uses numbered lists.

Stage 4 — Maintain the setup

Skills: CCAOF-U5.T2.LO4.S1, CCAOF-U5.T2.LO4.S2, CCAOF-U5.T2.LO4.S3 Minutes: 12

Two months later, three things have changed: the Lantern Tent is now called the Harbour Lights Tent; last year’s rota files are no longer used; and the festival is over, though next year’s team may want to look back at the project.

1. Update the project’s instructions and briefing for the new tent name, and remove the old name. Test the change in a new chat.
2. Write down what you would do with last year’s rota files, and with the project once the festival is over. Plan only (Team or Enterprise): one steward who has left should lose access; say how.
3. If memory is available to you, open your memory and look for the old tent name; edit or delete it. If memory is not available, write down how you would do it.
4. Plan only (Team or Enterprise with memory on): you accidentally asked about your own tax return in a chat inside the project. Write down how to keep it out of the project’s memory.

Acceptance checks

• Stage 1: the project has the briefing in its files and short instructions; you sorted all six lines; you explained the second chat’s answer; your sharing plan gives each group one permission.
• Stage 2: you named what Claude reads from the Word document; all three files have descriptive names; your question names a document; your connector steps include signing in, switching on per chat and approval.
• Stage 3: each rule has a level; the rewrite is concrete; you tested in a new chat with three kinds of request; you predicted both parts.
• Stage 4: the old name is gone from instructions and briefing; you have a plan for the files, the project and the steward; you have a memory step for each of 3 and 4.

Reference solution

Stage 1. The description is for people only, because Claude does not have access to a project’s name or description.

• Project instructions: a, b, d and f. They are Claude’s role and guidelines for every chat.
• Ask in a chat: c and e. They are single tasks, and task-specific instructions belong in the chat itself.
• With memory and chat search off, the second chat gives the briefing’s 30 minutes, not the 45 you told the first chat: context is not shared across chats in a project unless it is added to the project knowledge. With them on, it may pick up 45, but the briefing file still says 30. Either way, the change belongs in the briefing file, which every chat uses.
• Sharing plan: keep the project private and invite both groups, the shift leads with “Can edit” and the stewards with “Can view”. Sharing needs a Team or Enterprise plan.

Stage 2.

• For a non-PDF document such as Word, Claude extracts text only and cannot interpret embedded images. So the map’s locations need writing out, for example: “Market Square tent: volunteer sign-in. Lantern Tent: lost children. Quay Stage: first aid.”
• File names, for example: “Volunteer code of conduct”, “Road-closure notice, last year” and “First-aid procedures”.
• The question should name the document: “Using the first-aid procedures document, what should a steward do if…?”
• Connector plan:
  1. On Team and Enterprise plans, an owner or primary owner enables Gmail and Google Drive for the organization.
  2. You sign in to each yourself, unless the organization uses Enterprise-managed auth (in beta).
  3. You toggle Gmail on only in the chats that handle volunteer email, and Drive only where the rota is needed.
  4. Claude drafts, and by default asks for your approval before each action such as sending. On Team and Enterprise plans, owners decide whether members can allow these actions to run without asking.
  5. Claude can reach only what you can already open.
• On Team and Enterprise plans, connectors are available only in private projects, so if you had made the project public (“Everyone at” your organization), you would use them in chats outside it; your invite-only project is private, so they are available there, although the Google Workspace page says the Drive files option is disabled for shared projects.

Stage 3.

• Levels: a is account, since it applies to all your conversations. b is project. c is organization, set by an owner on a Team or Enterprise plan.
• A good rewrite: “Thank volunteers at the start of each email.” and “Keep emails under 120 words, with the shift time in the first line.”
• The test counts only in a new conversation, and should cover different types of request.
• Prediction: the email signs off “The Festival Team”, because organization instructions take precedence and Claude favors them in a direct contradiction (the help center notes that in rare edge cases behavior may vary). The rota uses numbered lists, because your personal instructions still apply where the organization’s are silent.

Stage 4.

1. Replace “Lantern Tent” in the instructions and briefing, remove the old wording rather than adding a second line (contradictory instructions may not be followed reliably), and test in a new chat.
2. Remove the unused rota files, and archive the project when the festival is over. Archiving keeps members and sharing, so remove the steward in the project’s sharing settings.
3. Open the memory topic that mentions the Lantern Tent, and edit or delete it; or tell Claude to forget the old name. Deleting old chats would not remove the memory. On Team and Enterprise plans, memory is off until an owner turns it on.
4. For Team and Enterprise plans using memory, use “Remove from project” on that chat, so it counts toward your non-project memory instead of the project’s.

Sources

• What are projects? — https://support.claude.com/en/articles/9517075-what-are-projects
• How can I create and manage projects? — https://support.claude.com/en/articles/9519177-how-can-i-create-and-manage-projects
• Manage project visibility and sharing — https://support.claude.com/en/articles/9519189-manage-project-visibility-and-sharing
• Upload files to Claude — https://support.claude.com/en/articles/8241126-upload-files-to-claude
• Retrieval augmented generation (RAG) for projects — https://support.claude.com/en/articles/11473015-retrieval-augmented-generation-rag-for-projects
• How do usage and length limits work? — https://support.claude.com/en/articles/11647753-how-do-usage-and-length-limits-work
• Use connectors to extend Claude’s capabilities — https://support.claude.com/en/articles/11176164-use-connectors-to-extend-claude-s-capabilities
• Use Google Workspace connectors — https://support.claude.com/en/articles/10166901-use-google-workspace-connectors
• Understanding Claude’s personalization features — https://support.claude.com/en/articles/10185728-understanding-claude-s-personalization-features
• Set organization instructions — https://support.claude.com/en/articles/14546867-set-organization-instructions
• Use Claude’s chat search and memory to build on previous context — https://support.claude.com/en/articles/11817273-use-claude-s-chat-search-and-memory-to-build-on-previous-context

Configuration and Knowledge Management — study note

Domain 5 of the Claude Certified Associate – Foundations exam. Its four objectives ask you to:

• configure Claude projects with instructions and knowledge sources;
• manage uploaded knowledge and connectors, such as Google Drive and Gmail;
• create effective system-level instructions;
• inform, maintain and update Claude configurations, knowledge sources and instructions.

Projects

Project instructions (listed as a paid-plan feature on one help-center page) work best for general context, key guidelines and Claude’s role. Keep them concise, and reserve task-specific instructions for the chat itself.

Sharing (Team and Enterprise plans). A project shared only with invited people is private. Members get “Can view” (see and chat, no editing) or “Can edit” (change instructions and knowledge, update member settings). A public project can be viewed and used by everyone in the organization, but each person’s chats stay private unless they share them. If an owner turns project sharing off, projects already shared stay shared, and public projects become private.

Knowledge and connectors

Connectors inherit your permissions: Claude reaches only what you can already open. On Team and Enterprise plans they work only in private projects, and the Google Workspace page says the Drive option for adding project files is disabled for shared projects. Owners can set each action to Always allow, Needs approval or Blocked for everyone. Connectors are token-intensive, so switch off the ones a chat does not need and disconnect services you no longer use.

Instructions

• Specific: “respond in formal English, no contractions, slang or emojis”, not “be professional”.
• Consistent: contradictory instructions may not be followed reliably.
• Short: organization instructions are included in every message everyone sends.
• Tested: after saving, start a new conversation and try several types of question.
• Precedence: organization instructions take precedence over personal ones, though in rare edge cases of direct contradiction behavior may vary; personal instructions still apply where the organization’s are silent. Organization preferences can’t disable Claude’s built-in safety guidelines.

Maintenance

Pausing memory keeps it but stops Claude using or adding to it; resetting deletes all memories. Deleting a conversation does not remove memories generated from it. On Team and Enterprise plans, memory is off by default until an owner turns it on.

Sources

• What are projects? — https://support.claude.com/en/articles/9517075-what-are-projects
• How can I create and manage projects? — https://support.claude.com/en/articles/9519177-how-can-i-create-and-manage-projects
• Manage project visibility and sharing — https://support.claude.com/en/articles/9519189-manage-project-visibility-and-sharing
• Upload files to Claude — https://support.claude.com/en/articles/8241126-upload-files-to-claude
• Retrieval augmented generation (RAG) for projects — https://support.claude.com/en/articles/11473015-retrieval-augmented-generation-rag-for-projects
• How do usage and length limits work? — https://support.claude.com/en/articles/11647753-how-do-usage-and-length-limits-work
• How large is the context window on paid Claude plans? — https://support.claude.com/en/articles/8606394-how-large-is-the-context-window-on-paid-claude-plans
• Use connectors to extend Claude’s capabilities — https://support.claude.com/en/articles/11176164-use-connectors-to-extend-claude-s-capabilities
• Use Google Workspace connectors — https://support.claude.com/en/articles/10166901-use-google-workspace-connectors
• Understanding Claude’s personalization features — https://support.claude.com/en/articles/10185728-understanding-claude-s-personalization-features
• Set organization instructions — https://support.claude.com/en/articles/14546867-set-organization-instructions
• Use Claude’s chat search and memory to build on previous context — https://support.claude.com/en/articles/11817273-use-claude-s-chat-search-and-memory-to-build-on-previous-context

Governance, Risk, and Responsible Use — practice exercise

Difficulty: intermediate · Estimated duration: 45–60 minutes

This is a written exercise. No code is involved, and nothing needs to be sent to Claude, so it works on any plan. You may use Claude to discuss your answers, but do not paste the invented personal details below into a real chat. The scenario is invented for practice: you are the operations lead at a mid-sized lettings and property-management agency that has just adopted Claude on a Team plan.

Every answer can be worked out from the rules card below. Most rows restate the course pages; the rows marked as house practice are this course’s advice.

Stage 1 — Sort the requests

Skills: CCAOF-U6.T1.LO1.S1, CCAOF-U6.T1.LO1.S2, CCAOF-U6.T1.LO1.S3 Minutes: 12

Label each request prohibited, high-risk, extra guidelines or ordinary, and give the rule in one line:

1. A negotiator wants Claude to write an email that looks as if it came from the local council, telling a tenant to leave.
2. The agency wants a public chat on its website that tells applicants whether they qualify for a tenancy.
3. A manager wants a checklist for the annual fire-alarm test.
4. A staff member wants a copy of a tenant’s passport with the date of birth altered.
5. The website will get a chat that answers questions about viewing times.

Then answer: a property manager’s legitimate question about gas-safety rules is blocked by a filter, and the next day she gets a warning she thinks is a mistake. What should she do, and what should she not do?

Stage 2 — Handle the data

Skills: CCAOF-U6.T1.LO2.S1, CCAOF-U6.T1.LO2.S2, CCAOF-U6.T1.LO2.S3 Minutes: 12

A colleague drafted this prompt:

Summarize the rent arrears for flat 4B. Tenant: J. Rahman, bank sort code 00-00-00, account 00000000, portal password Lettings!2026. She has told us she is in hospital for surgery next month. Arrears: January £400, February £400, March £0.

1. Rewrite the prompt so that it carries only what the summary needs, and say why each detail you removed can go.
2. The colleague suggests using an incognito chat “so nothing is kept”. Write two sentences on what incognito would and would not change.
3. A sister company, a physiotherapy clinic group in the United States, is also on a Team plan. It wants to use its account to summarize patients’ treatment notes. Write what must be in place first, and who can put it in place.

Stage 3 — Set up governance

Skills: CCAOF-U6.T2.LO3.S1, CCAOF-U6.T2.LO3.S2, CCAOF-U6.T2.LO3.S3 Minutes: 14

1. Draft two lines of organization instructions for the agency, one on tone and one on personal data, and then one line you would like to add but cannot. Say why the last one would not work.
2. The partners want file creation unavailable to everyone while a data audit runs. Say which lever does that and who pulls it.
3. A negotiator needs Claude to check a current rule online, but web search is missing from her chats. Write what she should do today and whom she should ask.
4. Choose public or private for two projects, with the reason: (a) the agency’s house style guide; (b) the partners’ draft salary review.
5. A negotiator uses Claude to draft a notice to a tenant that cites housing legislation. Write the three-line record the agency should keep about how the notice was produced and checked.

Stage 4 — Use it honestly

Skills: CCAOF-U6.T2.LO4.S1, CCAOF-U6.T2.LO4.S2, CCAOF-U6.T2.LO4.S3 Minutes: 12

1. Write the opening line the website’s maintenance-request chat should show at the start of each session.
2. Marketing proposes that the chat introduce itself as “Tom from Maintenance” and, if asked, say it is a real person. Say which part can go ahead and which cannot, and why.
3. The agency’s weekly tenant newsletter is written by Claude and published on the website automatically, with nobody reviewing it. This week it included a biased line about one group of tenants. The tool has no report button. Write the report you would make and the change to the agency’s process.

Reference solution

Stage 1.

1. Prohibited: impersonating a real organization to mislead someone.
2. High-risk, and extra guidelines too: a consumer-facing decision about eligibility for housing needs professional review before release and disclosure at a minimum at the beginning of each session, and because it is a public chat it must also tell users it is AI at the start of each chat session.
3. Ordinary.
4. Prohibited: a falsified document, and misuse of private information.
5. Extra guidelines: it is not high-risk, but as a consumer-facing chatbot it must tell users it is AI at the start of each chat session.

For the blocked question: filters are not failsafe, so the block may be a false positive. She should email about the mistaken warning, with her details and account information, as the help center directs; if her account were ever suspended, she would log in and use the appeal form. She should not reword the question to trick the filter or move to another account.

Stage 2.

1. For example: Summarize the rent arrears for flat 4B: January £400, February £400, March £0. The bank details go because the task does not need them, and the password goes because a password is never shared. The health detail goes because it is a health record the summary does not need. The tenant’s name goes too: the flat identifies the account, and, as house practice, a summary carries only what the task needs.
2. Incognito keeps the chat out of history and memory and out of training. It is still retained for a period and, on a work plan, included in the data exports Owners can access, so it is not a way to share what should not be shared.
3. The HIPAA-ready configuration, which is for Enterprise plans only: on a Team plan it can’t be enabled at all, so the clinic group would first need an Enterprise plan. Then only its Primary Owner can accept the BAA, which is click-to-accept in organization settings, and enabling it is a one-way decision. After that, treatment notes go only through covered features. Until then, they stay out.

Stage 3.

1. For example: Write in plain, courteous British English. / Never include tenants’ bank details, dates of birth or health information in responses or files. The line that cannot work is anything that switches off Claude’s safety guidelines or content policies: organization instructions cannot disable them. The personal-data line is prompt-level guidance, so test it and do not treat it as a hard control.
2. A capability setting: an Owner turns code execution and file creation off in organization settings. An instruction would only ask Claude not to.
3. Work within it today, for example by pasting the official text she already has and asking Claude to work from that. Ask an Owner or the Primary Owner, who enables web search for the workspace; once it is on, she switches it on in her chat where the chat has a web-search toggle (in the new Claude experience, Claude searches when it helps). She should not move the work to a personal account.
4. (a) Public: everyone should use it. (b) Private, with only the partners invited.
5. For example: Drafted with Claude on [date] by [negotiator]. / Legislation citations checked against the primary source by [name]. / Reviewed and approved by [qualified person] before sending.

Stage 4.

1. For example: You’re chatting with an AI assistant. A member of our team will pick up anything it can’t handle.
2. The name can go ahead as a custom AI persona, as long as the chat says it is AI at the start of each session. Saying it is a real person cannot: presenting AI output as a person’s is prohibited, and Claude will not claim to be human when directly and sincerely asked.
3. Report the output by email to the user-safety address the Usage Policy gives, since there is no report button. The process change: automatically generated content published for external consumption is a high-risk use, so a qualified person reviews each issue before it is published, readers are told AI helped, and the policy names who reviews and reports problems for the tool.

Sources

• Usage Policy — https://www.anthropic.com/legal/aup
• Our approach to user safety — https://support.claude.com/en/articles/8106465-our-approach-to-user-safety
• Safeguards warnings and appeals — https://support.claude.com/en/articles/8241253-safeguards-warnings-and-appeals
• Who can view my conversations? (sensitive data) — https://support.claude.com/en/articles/8325621-i-would-like-to-input-sensitive-data-into-my-chats-with-claude-who-can-view-my-conversations
• Use incognito chats — https://support.claude.com/en/articles/12260368-use-incognito-chats
• HIPAA-ready Enterprise plans — https://support.claude.com/en/articles/13296973-hipaa-ready-enterprise-plans
• Using Claude for legal work — https://support.claude.com/en/articles/15707726-using-claude-for-legal-work-privilege-confidentiality-and-how-to-think-about-configuration
• Set organization instructions — https://support.claude.com/en/articles/14546867-set-organization-instructions
• Enable and use web search — https://support.claude.com/en/articles/10684626-enable-and-use-web-search
• Create and edit files with Claude — https://support.claude.com/en/articles/12111783-create-and-edit-files-with-claude
• Manage project visibility and sharing — https://support.claude.com/en/articles/9519189-manage-project-visibility-and-sharing
• Claude’s constitution — https://www.anthropic.com/constitution

Governance, Risk, and Responsible Use — study note

Domain 6 of the Claude Certified Associate – Foundations exam. Its four objectives ask you to:

• identify appropriate and inappropriate use cases;
• apply data sensitivity, regulatory and privacy considerations;
• follow organizational AI policies and governance standards;
• understand the ethical implications of AI usage.

Nothing here is legal advice. Where a question turns on the law, your organization’s own legal or compliance owners decide. Domain 2 covers what high-risk uses require (professional review and disclosure) and how to correct flawed output; Domain 5 covers writing organization instructions and sharing projects. This domain takes the governance angle on the same ground.

Which group is the use in?

The Usage Policy applies to anyone who can submit inputs, including through resellers or passthrough access.

Safety filters, warnings and appeals. Detection models can flag prompts, and safety filters may then block a response. Anthropic says these features are not failsafe: false positives and false negatives happen. Repeated violations can bring enhanced filters for a period, removed after a stretch with few or no violations.

Opening another account to get round a ban is itself prohibited.

Is the data appropriate to share?

Governance inside an organization

What an AI policy for professional work should settle (modelled on Anthropic’s legal-work article, which is Anthropic’s view and not legal advice): commercial terms with a DPA and the no-training commitment; client-consent and engagement-letter practices; who verifies output against primary sources and how AI use is documented; and deployment decisions discussed with your own counsel, since the law is moving quickly.

Ethics

• Be open. Consumer-facing chatbots, including any external-facing or interactive AI agent, must tell users they are talking to AI, at a minimum at the beginning of each chat session. Never present AI output as human-generated. The Usage Policy lists submitting AI-assisted work without proper permission or attribution alongside plagiarism.
• Claude’s values. The constitution wants Claude to be exceptionally helpful while honest, thoughtful and caring about the world. It generally prioritizes being broadly safe, then broadly ethical, then following Anthropic’s guidelines, then genuinely helpful, weighed holistically rather than strictly. Claude only sincerely asserts what it believes true and avoids creating false impressions; a persuasive essay on request is a performative assertion, not a lie. Operators may give Claude a custom AI persona, but by default, whatever an operator instructs, it will not claim to be human when sincerely asked, and operators cannot have it deceive users with false information. A user can set up a role-play in which Claude plays a human.
• Use the right route. Report potentially inaccurate, biased or harmful output through the “report issues” thumbs-down or similar feedback, where available, or to the user-safety email. Feedback on the safety measures themselves goes to the address on Anthropic’s safety page. As house practice, a tool that runs unattended needs a named person who reviews and reports.

Sources

• Usage Policy — https://www.anthropic.com/legal/aup
• Our approach to user safety — https://support.claude.com/en/articles/8106465-our-approach-to-user-safety
• Safeguards warnings and appeals — https://support.claude.com/en/articles/8241253-safeguards-warnings-and-appeals
• Who can view my conversations? (sensitive data) — https://support.claude.com/en/articles/8325621-i-would-like-to-input-sensitive-data-into-my-chats-with-claude-who-can-view-my-conversations
• Use incognito chats — https://support.claude.com/en/articles/12260368-use-incognito-chats
• HIPAA-ready Enterprise plans — https://support.claude.com/en/articles/13296973-hipaa-ready-enterprise-plans
• Using Claude for legal work — https://support.claude.com/en/articles/15707726-using-claude-for-legal-work-privilege-confidentiality-and-how-to-think-about-configuration
• Set organization instructions — https://support.claude.com/en/articles/14546867-set-organization-instructions
• Enable and use web search — https://support.claude.com/en/articles/10684626-enable-and-use-web-search
• Create and edit files with Claude — https://support.claude.com/en/articles/12111783-create-and-edit-files-with-claude
• Manage project visibility and sharing — https://support.claude.com/en/articles/9519189-manage-project-visibility-and-sharing
• Claude is providing incorrect or misleading responses — https://support.claude.com/en/articles/8525154-claude-is-providing-incorrect-or-misleading-responses-what-s-going-on
• Claude’s constitution — https://www.anthropic.com/constitution

Output Evaluation and Validation — practice exercise

Difficulty: intermediate · Estimated duration: 50–65 minutes

You will do this exercise in the Claude app, with a spreadsheet or a calculator for one step. No code is involved. Stages 1–4 work on web, desktop or mobile; Stage 5 needs Claude on the web or desktop, with the code execution and file creation capability turned on (organization owners can disable it). Every document you need is provided below and invented for practice, so you can check Claude’s answers against them rather than against facts you would have to look up. Keep a short record as you go.

Practice document — remote working policy (extract), used in Stages 2, 3 and 5:

Staff may work remotely up to three days a week with their manager’s approval. Requests are made in the HR portal at least two weeks in advance. Equipment for home working is provided on request; staff are responsible for keeping their home workspace safe.

Stage 1 — Spot what needs checking

Skills: CCAOF-U2.T1.LO1.S1, CCAOF-U2.T1.LO1.S2, CCAOF-U2.T1.LO1.S3, CCAOF-U2.T1.LO1.S4, CCAOF-U2.T1.LO2.S1, CCAOF-U2.T1.LO2.S3 Minutes: 17

Here is a draft memo, written as if Claude had produced it:

Travel policy update. As the Finance Director said last week, “every trip needs two approvals.” The new mileage rate took effect on 1 August this year. A 2025 survey found 72% of staff prefer rail. This quarter’s budget: 3 trips × $240 = $620. Younger staff tend to book the cheapest options, so they need extra reminders about comfort.

1. List every sentence you would not rely on without checking, and say why for each: an unverified quotation, a recent event, an unsourced statistic, a calculation, or a biased generalization.
2. Recompute the budget line with a calculator or spreadsheet.
3. Decide which of these problems you would also report through the product’s feedback feature, where available, if Claude had written it.
4. A colleague asked Claude for three things about the office move: the move date, which floors move first, and who to contact about IT equipment. Claude replied: “The move is on 3 June, and floors 4 and 5 move first.” Decide whether the reply is complete, and what you would ask next.

Stage 2 — Expose inconsistency

Skills: CCAOF-U2.T1.LO2.S2 Minutes: 8

1. Paste the remote-working extract (above) into a new chat and ask: How much notice do I need to give for a remote-working day?
2. Ask the same question in a second new chat, with the same extract.
3. Compare the two answers. Then, in one chat, ask Claude to explain step by step where in the extract its answer comes from, and to verify it against the text.

Stage 3 — Fact-check and ground the answer

Skills: CCAOF-U2.T2.LO3.S1, CCAOF-U2.T2.LO3.S2, CCAOF-U2.T2.LO3.S3 Minutes: 12

The original document is the remote-working extract above. A summary someone circulated: “Staff can work remotely three days a week.”

1. Compare the summary with the original. Write down every condition the summary dropped.
2. In a new chat, paste the extract and ask Claude to summarize it, citing the supporting sentence for each point and leaving out any point it cannot support.
3. In the same chat, instruct Claude to answer only from the extract, and to reply “not stated” if the extract does not say. Then ask: Can I work remotely from another country?

Stage 4 — Decide who must review

Skills: CCAOF-U2.T2.LO4.S1, CCAOF-U2.T2.LO4.S2, CCAOF-U2.T2.LO4.S3 Minutes: 8

For each draft Claude might produce, decide whether it needs review by a qualified professional before release, a validation of its critical figures, or your own check:

1. A reply sent to a customer telling them whether their insurance claim is covered.
2. Tips on stretching at your desk for the internal staff newsletter.
3. A decision on whether a customer qualifies for a payment plan.
4. A summary of a team meeting.
5. The cost figures in a paper going to the board.

For item 1, say what the customer must also be told, and describe in one sentence how the professional stays in charge of the work.

Stage 5 — Shape it for the reader

Skills: CCAOF-U2.T3.LO5.S1, CCAOF-U2.T3.LO5.S2, CCAOF-U2.T3.LO5.S3, CCAOF-U2.T3.LO6.S1, CCAOF-U2.T3.LO6.S2, CCAOF-U2.T3.LO6.S3 Minutes: 12

Using the remote-working extract:

1. Ask for two versions of a short explainer: one for managers who approve requests, one for new starters. Say who each is for and why.
2. For the new-starter version, describe the style you want (for example, three short, friendly paragraphs), rather than listing what to avoid.
3. Ask for the new-starter explainer as a document artifact. Highlight one sentence and use Edit with Claude to change it. Then edit an earlier message to branch the chat and try a different opening, and compare the two versions.
4. Decide where each of these should live, and ask for one of them: a one-line answer to “How many remote days can I take?”, the explainer, and a tracker HR will update in their spreadsheet tool. For the tracker, describe its columns.
5. Ask for a table comparing the managers’ and new starters’ versions, with columns for audience, length and main message.

Acceptance checks

• Your Stage 1 list names five issues, each with its reason, and the corrected budget total, and you identified the incomplete reply and the missing part.
• Your Stage 2 record shows whether the two answers agreed, and what the step-by-step verification pointed to.
• Your Stage 3 record lists the conditions the circulated summary dropped, a summary with a supporting sentence per point, and a “not stated” answer to the question about working abroad.
• Your Stage 4 answers classify all five drafts, name the disclosure for item 1, and describe how the professional stays in charge.
• Your Stage 5 work includes two versions for two readers, a style described positively, an edited and a branched version, a placement decision for three outputs, and a comparison table.

Reference solution

Stage 1. Five sentences need checking. The quotation from the Finance Director cannot be relied on until you find a record of it; it is the kind of authoritative-looking detail that can be invented. The date the new mileage rate took effect is a recent event, possibly after the model’s knowledge cutoff, and nothing in this exercise supplies the real date, so the right answer is “check it”, not a corrected date. The 72% statistic has no source. The budget line is wrong: 3 × $240 is $720, not $620. The last sentence is a generalization about an age group, which should be removed. For step 3, the criterion is to report output you know to be wrong or biased: from the supplied text alone, that is the wrong sum and the biased generalization. The quotation, the 72% statistic and the date are unverified rather than known to be wrong, so you check them first and report any that turn out to be wrong. For step 4, the reply is incomplete: it gives the date and the floors, but not the IT contact, so ask for that part. The point of the stage is the habit, not the memo: a fluent paragraph can contain several kinds of error, and a good reply can still leave out a part of the request.

Stage 2. Both answers should say two weeks, because the extract says requests are made at least two weeks in advance. If they disagree, that inconsistency is your signal to check the text. The step-by-step explanation should point to the sentence about the HR portal; if it points anywhere else, the answer was not grounded.

Stage 3. The circulated summary dropped several conditions: remote days are “up to” three a week, not a fixed three; the manager must approve; requests go through the HR portal at least two weeks in advance; and staff are responsible for a safe home workspace (equipment is available on request). Claude’s cited summary should attach a sentence from the extract to each point. Asked about working from another country, the grounded answer is “not stated”: the extract does not cover it. That is the correct answer, and it tells you exactly what to ask HR.

Stage 4. (1) Qualified professional review: an insurance coverage decision sent to a customer is a consumer-facing high-risk use case. (2) Your own check: wellness tips for an internal newsletter are outside the healthcare category and are not consumer-facing advice. (3) Qualified professional review: a financial eligibility decision directly affecting a customer. (4) Your own check. (5) Validate the critical figures before the board sees them; safeguards reduce errors but do not eliminate them. For item 1, the customer must also be told that AI helped produce the reply, at a minimum at the beginning of the session, and a qualified claims professional reviews the decision before it is sent. In the way Anthropic describes for professional work, that professional directs the drafting, verifies the decision against the primary sources (the policy terms themselves), and records that AI was used.

Stage 5. The managers’ version leads with the approval rule and the two-week notice; the new-starters’ version leads with how to make a request. A positively described style (“three short, friendly paragraphs”) usually lands better than “don’t be formal.” In a doc or Markdown artifact, Edit with Claude changes exactly the sentence you highlighted; branching an earlier message keeps both openings for comparison. The one-line answer (“up to three days a week, with your manager’s approval”) stays inline, the explainer is an artifact, and the tracker is a downloadable spreadsheet with columns such as name, requested days, date requested and approval status. The comparison table should have one row per version.

Sources

• Claude is providing incorrect or misleading responses — https://support.claude.com/en/articles/8525154-claude-is-providing-incorrect-or-misleading-responses-what-s-going-on
• How up-to-date is Claude’s training data? — https://support.claude.com/en/articles/8114494-how-up-to-date-is-claude-s-training-data
• How does Claude handle mathematical equations and calculations? — https://support.claude.com/en/articles/10366421-how-does-claude-handle-mathematical-equations-and-calculations
• Reduce hallucinations — https://platform.claude.com/docs/en/test-and-evaluate/strengthen-guardrails/reduce-hallucinations
• Usage Policy — https://www.anthropic.com/legal/aup
• Using Claude for legal work — https://support.claude.com/en/articles/15707726-using-claude-for-legal-work-privilege-confidentiality-and-how-to-think-about-configuration
• Prompting best practices — https://platform.claude.com/docs/en/build-with-claude/prompt-engineering/claude-prompting-best-practices
• What are artifacts and how do I use them? — https://support.claude.com/en/articles/17153992-what-are-artifacts-and-how-do-i-use-them
• Create and edit files with Claude — https://support.claude.com/en/articles/12111783-create-and-edit-files-with-claude
• Claude’s response to my prompt is too brief — https://support.claude.com/en/articles/8114518-claude-s-response-to-my-prompt-is-too-brief

Output Evaluation and Validation — study note

Domain 2 of the Claude Certified Associate – Foundations exam, and the heaviest domain. Its six objectives ask you to evaluate outputs for accuracy and completeness, spot hallucinations, inconsistencies and bias, fact-check, decide when human review is required, adapt output for its audience, and choose the right output format. This note summarizes each topic and the Anthropic pages behind it.

Checking accuracy

Fact-checking techniques

• Open the cited sources. Every web-search response includes citations, and the original may hold context the summary dropped.
• A quote for every claim. Ask Claude to cite quotes and sources for each claim, and to retract any claim it cannot support.
• Stay inside the documents. Tell Claude to use only the documents you supplied, and give it permission to say it does not know. For long documents, ask for word-for-word quotes first.

These techniques significantly reduce hallucinations, but they do not eliminate them. Always validate critical information, especially for high-stakes decisions.

When a qualified professional must review

The high-risk areas are legal, healthcare, insurance, finance, employment and housing, academic testing, accreditation and admissions, and media or professional journalistic content.

Anthropic suggests (legal work, not a Usage Policy requirement): keep a lawyer visibly directing the work, verify output against primary sources, and document the AI use. The existing duties of competence, confidentiality, supervision and verification still apply.

Shaping output

Artifacts and file creation need the code execution and file creation capability turned on; organization owners can disable it.

Sources

• Claude is providing incorrect or misleading responses — https://support.claude.com/en/articles/8525154-claude-is-providing-incorrect-or-misleading-responses-what-s-going-on
• How up-to-date is Claude’s training data? — https://support.claude.com/en/articles/8114494-how-up-to-date-is-claude-s-training-data
• How does Claude handle mathematical equations and calculations? — https://support.claude.com/en/articles/10366421-how-does-claude-handle-mathematical-equations-and-calculations
• Reduce hallucinations — https://platform.claude.com/docs/en/test-and-evaluate/strengthen-guardrails/reduce-hallucinations
• Enable and use web search — https://support.claude.com/en/articles/10684626-enable-and-use-web-search
• Usage Policy — https://www.anthropic.com/legal/aup
• Using Claude for legal work — https://support.claude.com/en/articles/15707726-using-claude-for-legal-work-privilege-confidentiality-and-how-to-think-about-configuration
• Prompting best practices — https://platform.claude.com/docs/en/build-with-claude/prompt-engineering/claude-prompting-best-practices
• Usage limit best practices — https://support.claude.com/en/articles/9797557-usage-limit-best-practices
• Claude’s response to my prompt is too brief — https://support.claude.com/en/articles/8114518-claude-s-response-to-my-prompt-is-too-brief
• What are artifacts and how do I use them? — https://support.claude.com/en/articles/17153992-what-are-artifacts-and-how-do-i-use-them
• Create and edit files with Claude — https://support.claude.com/en/articles/12111783-create-and-edit-files-with-claude

Product and Model Selection — practice exercise

Difficulty: beginner · Estimated duration: 40–55 minutes

You will do this exercise in the Claude app (web, desktop or mobile). No code is involved. You make choices, try them, and keep a short record of what you chose and why. The scenario is invented for practice: you are a team coordinator in a mid-sized company. Features and models vary by plan and by your organization's settings (research, for example, is on paid plans); where something is not available to you, write down what you would do and why instead of running it.

Stage 1 — Choose the feature

Skills: CCAOF-U3.T1.LO1.S1, CCAOF-U3.T1.LO1.S2, CCAOF-U3.T1.LO1.S3 Minutes: 12

1. For each task, write the feature you would use and one sentence on why:
   • (a) Is the city's main library open on Sunday this week?
   • (b) Answering staff questions about the same travel policy, several times a week.
   • (c) A cited comparison of four venues' published accessibility information.
   • (d) A one-page event checklist the team will edit and reuse.
2. Create a project called Team travel (any name will do). Paste this short policy into its knowledge as a text file, and add one project instruction: Answer as the team coordinator and quote the policy line you rely on.

Travel policy: Book trains at least 7 days ahead. Hotels up to the city rate need no approval; above it, your manager approves in writing. Keep receipts for 90 days.

1. In two separate chats inside the project, ask two different questions about the policy. Note whether you had to paste the policy again.
2. Ask Claude to turn the policy into a one-page checklist for new starters, as an artifact. Note whether it appears as an artifact, and who can see it.

Stage 2 — Compare models and effort

Skills: CCAOF-U3.T2.LO2.S1, CCAOF-U3.T2.LO2.S2, CCAOF-U3.T2.LO3.S2 Minutes: 12

1. Write one line placing Haiku, Sonnet and Opus on the capability–speed–cost trade-off.
2. Find the model selector in a chat and note which model is selected.
3. Ask for a short reply to a routine question (for example, confirming a meeting room booking). Then, in the same chat, switch to a different available model, or change the effort level if the model offers one, and ask again. Record what changed, if anything.
4. Ask a harder question that needs careful reasoning (for example, a rota for five people across seven days with two constraints each). Try it at the default setting, then with higher effort or thinking if your model offers them. Record any difference in depth and in how long it took.

Stage 3 — Plan and test a model choice

Skills: CCAOF-U3.T2.LO3.S1, CCAOF-U3.T2.LO3.S3 Minutes: 12

1. For each task, write whether you would start efficiency first or capability first, and why: (a) tagging 2,000 short feedback comments by topic; (b) reviewing one complex supplier contract for risks before signature.
2. For task (a), design a small test set: write five example comments and the topic you expect for each.
3. Run your five comments with one prompt, and check the answers against your expected topics.
4. Run the same five comments with a second configuration (a different available model, or a different effort level where offered), and write one sentence comparing the two runs.

Stage 4 — Manage limits and persist context

Skills: CCAOF-U3.T3.LO4.S1, CCAOF-U3.T3.LO4.S2, CCAOF-U3.T3.LO4.S3 Minutes: 12

1. Classify each case and write the fix: (a) a single message is too long to send, even in a new chat; (b) you have used your allowance for now across all chats; (c) a conversation has grown so long that it hits its length limit.
2. Take any conversation from Stage 1 or 2. Ask Claude to summarize the decisions made and the open questions in five bullets. Start a new chat and paste that summary as its first message. Note what carried over and what did not.
3. Write which feature you would use for each need: (i) the same policy in every chat; (ii) Claude remembering that you prefer bullet-point summaries; (iii) finding a discussion from last month; (iv) a conversation you want kept out of memory and chat history.

Acceptance checks

• Stage 1: four feature choices with reasons; a project in which a second chat answered without the policy being pasted again; a note on the checklist artifact and its visibility.
• Stage 2: a correct one-line placement of the three families, and a record of what changed when you switched model or effort.
• Stage 3: a justified starting strategy for each task, a five-item test set, and a comparison of two runs.
• Stage 4: three cases classified with the right fix, a fresh chat started from a summary, and four persistence choices.

Reference solution

Stage 1. (a) Web search: one current fact. (b) A project: the same policy and instructions are needed in many chats, so they go in the project knowledge once. (c) Research: several sources and citations are needed; it is on paid plans, needs web search turned on and can use more of your limits. (d) An artifact: substantial, self-contained output the team will edit and reuse. In the project, both chats should answer from the policy without you pasting it again, because project knowledge is used across all of its chats — for example, “Hotels up to the city rate need no approval.” The checklist should appear as an artifact beside the chat if the code execution and file creation capability is on; artifacts start private to you, so only you can see it until you share it.

Stage 2. A good placement: of these three, Haiku is the fast, low-cost end, Sonnet balances speed and capability for everyday work, and Opus is the capability end (the lineup also includes other models). The chat shows which model is in use, and selecting it opens the selector; changes apply starting with Claude’s next response, in the same chat. For the routine question, a different model or lower effort may give a similar answer, faster or using less, or you may see little difference. For the rota, higher effort or thinking, where offered, may give a more careful answer that takes longer — the trade-off the guidance describes. If a setting was not available to you, your record should say what you would have expected and why.

Stage 3. (a) Efficiency first: high-volume, straightforward work, where a fast, low-cost model is often sufficient; upgrade only if the test shows a gap. (b) Capability first: a complex task where accuracy outweighs cost. A good test set has varied comments with clear expected topics, including one borderline case. Your comparison should rest on how many of the five each run got right, weighed against speed and usage, not on how the model is described.

Stage 4. (a) The message must be shortened: attach fewer or smaller files, summarize or extract the key sections, or split the content into chunks. (b) Usage limit: wait for the reset or, depending on your plan, add usage — a new chat does not help. (c) Length limit for the conversation: start a new chat with a summary, or use a project. In the fresh chat, the summary carries the decisions and open questions; details you left out do not carry over unless memory or chat search brings them in, depending on your settings. (i) Project knowledge; (ii) memory; (iii) chat search; (iv) an incognito chat, started outside a project.

Sources

• What are projects? — https://support.claude.com/en/articles/9517075-what-are-projects
• How can I create and manage projects? — https://support.claude.com/en/articles/9519177-how-can-i-create-and-manage-projects
• When should I use web search, extended thinking, and research? — https://support.claude.com/en/articles/11095361-when-should-i-use-web-search-extended-thinking-and-research
• Use research on Claude — https://support.claude.com/en/articles/11088861-use-research-on-claude
• What are artifacts and how do I use them? — https://support.claude.com/en/articles/17153992-what-are-artifacts-and-how-do-i-use-them
• Choosing the right model — https://platform.claude.com/docs/en/about-claude/models/choosing-a-model
• Change the model, effort, and thinking settings — https://support.claude.com/en/articles/8664678-change-the-model-effort-and-thinking-settings
• How do usage and length limits work? — https://support.claude.com/en/articles/11647753-how-do-usage-and-length-limits-work
• Troubleshoot Claude error messages — https://support.claude.com/en/articles/12466728-troubleshoot-claude-error-messages
• Use Claude’s chat search and memory to build on previous context — https://support.claude.com/en/articles/11817273-use-claude-s-chat-search-and-memory-to-build-on-previous-context

Product and Model Selection — study note

Domain 3 of the Claude Certified Associate – Foundations exam. Its four objectives ask you to choose the right Claude feature, tell the model families apart, match a model to the task's needs for cost, speed and quality, and manage context limits and memory. This note summarizes each topic and the Anthropic pages behind it. Model versions, context sizes and plans change often; the note deliberately teaches the relative positions and the judgement instead.

Choosing a feature

The model families

The lineup also includes other models; the exam's objective names these three.

Choosing a model means balancing capability, speed and cost. You can see the current model by the message box and change it — or the effort and thinking settings — at any point in a conversation; changes apply starting with Claude's next response. On organization plans, an administrator may limit which models appear.

Matching the model to the task

• Efficiency first for high-volume, straightforward work: quick iteration at lower cost, often sufficient; upgrade only for a specific capability gap.
• Capability first for complex work where accuracy outweighs cost: start with the strongest starting point for the task, and optimize to more efficient models later.
• Effort before switching (on models that offer an effort setting; Haiku does not, though you can still switch extended thinking on or off): effort controls how much thinking Claude applies. Lower effort suits routine tasks and stretches usage; raise effort or turn on thinking for complex work, at the cost of time and usage.
• Test it: a good evaluation set specific to your task, run with your actual prompts and data, is the most important step. Speed depends on prompt length, output length and effort, not on the model alone.

Context limits and memory

• Part of the context window is reserved for Claude's reply. On paid plans with code execution enabled, Claude summarizes earlier messages near the limit — but longer conversations use more of your usage.
• Tools and connectors take up context space and usage; switch off ones a chat does not need.
• Restart well: summarize decisions and open points, then start a new chat with that summary.
• Carrying context: a new chat starts without earlier context unless memory, chat search or project knowledge carries it, depending on your plan and settings (memory and chat search are on by default for some users). Project knowledge is shared across a project's chats, and on paid plans, once it nears the context limit, Claude retrieves only what each question needs. Memory carries your role and preferences into new chats, with each project keeping its own. Chat search, on paid plans, finds earlier conversations. Incognito chats, started outside a project, stay out of memory and history.

Sources

• What are projects? — https://support.claude.com/en/articles/9517075-what-are-projects
• How can I create and manage projects? — https://support.claude.com/en/articles/9519177-how-can-i-create-and-manage-projects
• Retrieval augmented generation (RAG) for projects — https://support.claude.com/en/articles/11473015-retrieval-augmented-generation-rag-for-projects
• When should I use web search, extended thinking, and research? — https://support.claude.com/en/articles/11095361-when-should-i-use-web-search-extended-thinking-and-research
• Use research on Claude — https://support.claude.com/en/articles/11088861-use-research-on-claude
• What are artifacts and how do I use them? — https://support.claude.com/en/articles/17153992-what-are-artifacts-and-how-do-i-use-them
• Models overview — https://platform.claude.com/docs/en/models/overview
• Choosing the right model — https://platform.claude.com/docs/en/about-claude/models/choosing-a-model
• Change the model, effort, and thinking settings — https://support.claude.com/en/articles/8664678-change-the-model-effort-and-thinking-settings
• Get started with Claude — https://support.claude.com/en/articles/8114491-get-started-with-claude
• How do usage and length limits work? — https://support.claude.com/en/articles/11647753-how-do-usage-and-length-limits-work
• How large is the context window on paid Claude plans? — https://support.claude.com/en/articles/8606394-how-large-is-the-context-window-on-paid-claude-plans
• Troubleshoot Claude error messages — https://support.claude.com/en/articles/12466728-troubleshoot-claude-error-messages
• Use Claude’s chat search and memory to build on previous context — https://support.claude.com/en/articles/11817273-use-claude-s-chat-search-and-memory-to-build-on-previous-context

Prompting and Task Execution — practice exercise

Difficulty: beginner · Estimated duration: 45–60 minutes

You will do this exercise in the Claude app (web, desktop or mobile). No code is involved. You write prompts, read Claude’s replies, and keep a short record of what you changed and why. The scenario is invented for practice: you are an operations coordinator at a mid-sized company.

What you need: a Claude account and a blank note for your record. Stage 4 asks you to write opening messages, not run them, so every stage works on any plan.

Stage 1 — Write a complete brief

Skills: CCAOF-U1.T1.LO1.S1, CCAOF-U1.T1.LO1.S2, CCAOF-U1.T1.LO1.S3 Minutes: 14

The facts: expenses are now submitted in the finance app within 30 days, with a photo of the receipt, and approved by your line manager.

1. Send this deliberately vague prompt: Write something about the new expense process. Save the reply.
2. Now write a full brief for the same task. Include the facts above, the audience (all staff), the purpose (explain the new steps), and the form (five bullet points, under 150 words).
3. Add one instruction with its reason (for example: most staff read this on their phones, so put the deadline first).
4. Add three short example bullets written in your company’s style, marked as examples.
5. Send the brief. In your record, note two differences between the two replies.
6. Technical task: brief Claude for a spreadsheet formula. Your sheet has unit price in column B and quantity in column C; you want each row’s total in column D, rounded to cents. Write the brief so the inputs, the result and where it goes are all stated.

Stage 2 — Structure a multi-part request

Skills: CCAOF-U1.T1.LO2.S1, CCAOF-U1.T1.LO2.S2, CCAOF-U1.T1.LO2.S3 Minutes: 12

Use these meeting notes:

Ops sync, Tuesday. Priya will book the training room for the onboarding session. Sam to check the new-starter laptops by Friday. Budget for the team lunch still not agreed; Lee to ask finance. Next sync moved to Thursday.

1. Write one prompt that asks for three outputs: the action items, an owner for each, and a short follow-up email to the team.
2. List the three outputs as numbered steps in the order you need them.
3. Wrap the meeting notes in a tag such as <notes> … </notes>, typed as plain text, so they are separate from your instructions.
4. After Claude replies, ask it to review the email against two criteria: it names every owner, and it is under 100 words. Then ask it to refine the email based on that review.

Stage 3 — Iterate against criteria

Skills: CCAOF-U1.T2.LO3.S1, CCAOF-U1.T2.LO3.S2, CCAOF-U1.T2.LO3.S3 Minutes: 12

The facts: a customer returned a jacket on 2 March. Refunds are issued within 14 days of the returned item arriving at the warehouse, to the original payment method.

1. Before prompting, write your own three success criteria for a reply to this customer, who is asking where the refund is. Make each one something you can check by reading the reply.
2. Ask Claude for the reply, giving it the facts. Check the reply against each of your criteria and note which it meets.
3. For any criterion it misses, give specific feedback in the same chat that says what to do instead.
4. Practise continuing a cut-off answer. Open a new chat and paste the prompt and partial reply below. Ask Claude to continue from where the reply stops, without starting again.

Prompt: Write a three-section guide for customers: How returns work, When refunds arrive, What to do if a refund is late. Facts: refunds are issued within 14 days of the returned item arriving at our warehouse, to the original payment method.

Partial reply: How returns work. Pack the item with its label and drop it at any post office. When refunds arrive. Refunds are issued within

Stage 4 — Match the strategy to the task

Skills: CCAOF-U1.T3.LO4.S1, CCAOF-U1.T3.LO4.S2, CCAOF-U1.T3.LO4.S3, CCAOF-U1.T3.LO4.S4 Minutes: 12

Write the opening message you would send for each task, and name the capability you would use:

1. Is tomorrow’s national rail strike still going ahead?
2. A cited overview of three competitors’ published onboarding programs, for your manager.
3. A letter to customers about a price change (you choose the details).
4. Twelve names for a new internal newsletter, from safe to unconventional.
5. Which of five departments drove last quarter’s rise in overtime costs (use any small table you invent).

Acceptance checks

• Your Stage 1 brief includes the facts, the audience, purpose and form, one instruction with its reason, and three marked examples. Your formula brief names the input columns, the result and the destination.
• Your Stage 2 prompt numbers the outputs and tags the notes; the final email passed a review against two stated criteria.
• Your Stage 3 record shows your own criteria, which ones each version met, the exact feedback you gave, and a continued guide that picks up mid-sentence.
• Your Stage 4 answers pick a different strategy for each task and explain why.

Reference solution

Stage 1. The vague prompt usually returns a general article about expense processes. A complete brief reads: Write a short update for all staff explaining our new expense process: submit expenses in the finance app within 30 days, attach a photo of the receipt, and your line manager approves them. Format: five bullet points, under 150 words. Most staff read this on their phones, so put the 30-day deadline in the first bullet. Match the style of these example bullets: followed by three marked examples in your house style, such as Snap your receipt and attach it in the finance app. The reply should now name the app, the photo, the 30-day limit and manager approval, in five bullets with the deadline first. The two differences to record are that it fits the reader and the format, and that it leads with the deadline because you said why. For the formula, a good brief reads: In my sheet, column B is unit price and column C is quantity. Give me a formula for column D that multiplies them for each row and rounds the result to two decimal places.

Stage 2. A strong prompt reads: Using the meeting notes in the notes tag: 1. List the action items. 2. Give the owner of each. 3. Draft a follow-up email to the team. The notes then follow between <notes> and </notes>. Claude should find three actions: Priya books the training room, Sam checks the laptops by Friday, and Lee asks finance about the lunch budget. It should also mention the moved sync. The review step should catch an email that leaves out an owner or runs long, and the refined email should fix it. That is the draft, review and refine loop.

Stage 3. Good criteria are checkable, for example: states that the refund arrives within 14 days of the jacket reaching the warehouse; says it goes to the original payment method; ends with one next step if it has not arrived. A typical first reply meets one or two of them. Good feedback is specific and positive, for example: Add that the refund goes back to the original payment method, and end with one next step: reply with your order number if nothing has arrived after 14 days. Avoid “that’s wrong.” For the continuation, Claude should resume mid-sentence (“…14 days of your item arriving at our warehouse”) and then write the third section. It has the earlier exchange, including the refund facts, only because you pasted it; check that every figure in the continuation comes from those facts. In the original chat you could simply have asked it to continue.

Stage 4. (1) Web search, because it is a quick public fact. (2) Research, ideally with extended thinking so Claude plans before gathering; ask for citations your manager can check. Research needs web search turned on. (3) A drafting brief with the audience (existing customers), the requirements (friendly, under 200 words) and the key points (the new price, the date, what customers need to do). (4) An explicit request for range: twelve names, from safe to unconventional; go beyond the obvious. (5) An analysis request that states the question and the focus (by department, by month) and pastes the whole table in the same message. The skill being tested is the choice itself: five tasks, five different opening messages.

Sources

• Introduction to prompt design — https://support.claude.com/en/articles/7996853-introduction-to-prompt-design
• Prompting best practices — https://platform.claude.com/docs/en/build-with-claude/prompt-engineering/claude-prompting-best-practices
• My prompt isn’t giving me a helpful answer — https://support.claude.com/en/articles/7996857-my-prompt-isn-t-giving-me-a-helpful-answer
• Prompt engineering overview — https://platform.claude.com/docs/en/build-with-claude/prompt-engineering/overview
• Claude’s response to my prompt is too brief — https://support.claude.com/en/articles/8114518-claude-s-response-to-my-prompt-is-too-brief
• Get started with Claude — https://support.claude.com/en/articles/8114491-get-started-with-claude
• Use research on Claude — https://support.claude.com/en/articles/11088861-use-research-on-claude
• When should I use web search, extended thinking, and research? — https://support.claude.com/en/articles/11095361-when-should-i-use-web-search-extended-thinking-and-research
• Usage limit best practices — https://support.claude.com/en/articles/9797557-usage-limit-best-practices