Test your exam readiness
Pick a domain to practise, or take a full mixed exam. Each set is exam-style with instant scoring, a per-task-statement breakdown, and an explanation for every answer. Pass mark: 70%. Free to take — a free account unlocks the full answer review, gap analysis and certificate.
Sample Claude Certified Associate practice questions
6 free sample questions from the practice bank, with the correct answer and the reasoning.
Question 1 · Domain 1 — Prompting and Task Execution
Northwind Retail Support Team
A prompt engineer at Northwind Retail is drafting a reusable system prompt for a returns-eligibility assistant and decides to put the required JSON output schema at the very top, before describing the assistant's role. A reviewer flags this. What is the correct fix?
- A.Leave the output format first — the shape of the response is the most important constraint and should be seen immediately.
- B.Reorder to role → criteria → boundaries → output format, since role gives the model context before it reads the rules, and format is the final shape constraint once everything else is established. Correct answer
- C.Move the output format to be the second element, right after the role, since format is nearly as important as role.
- D.Order does not matter as long as all four elements are present somewhere in the prompt.
Why: Every reliable prompt has four parts in a fixed order: role → criteria → boundaries → output format. Role first gives the model context before it reads the rules; output format comes last as the final shape constraint once role and rules are already established. Why the other options are wrong: putting format first strips away the context the model needs to interpret the rules that follow. Format second is still premature — criteria and boundaries need to be established first. Order matters precisely because scrambling it gives later instructions less context to work from, producing inconsistent gap-filling.
Question 2 · Domain 2 — Output Evaluation and Validation
Beacon Health Claims Intake
Beacon Health's claims-summary assistant occasionally states a diagnosis code that contradicts the source chart. Which metric is required to measure this failure, and against what?
- A.p95/p99 latency, since slow responses correlate with rushed, less accurate summaries.
- B.Accuracy measured against a labeled dataset of charts with known-correct diagnosis codes. Correct answer
- C.Cost per request, since cheaper calls tend to under-invest in accuracy.
- D.Safety violation rate on an adversarial test set.
Why: Wrong or ungrounded answers require accuracy measured against a labeled dataset — inputs with known-correct answers. Latency, cost, and safety violation rate are each the right metric for a different failure mode, not for factual incorrectness.
Question 3 · Domain 3 — Product and Model Selection
Alden & Vance Realty
A 6-person leasing team at Alden & Vance re-pastes the same 300-word instruction block into a fresh Claude.ai conversation every week for their listing-comparison report. None of them can write code. What is the best fit?
- A.The Claude API with a config file their internal tool reads before each call.
- B.Claude.ai Projects, which stores shared, persistent instructions with no coding required. Correct answer
- C.Claude Code, since it supports reusable session instructions.
- D.A shared spreadsheet macro that calls the API directly.
Why: Claude.ai Projects gives a non-technical team with a repeated workflow shared, persistent instructions with zero integration work — exactly this team's situation. The API and a macro both require engineering resources the team doesn't have.
Question 4 · Domain 4 — Workflow Integration and Solution Design
Vantage Municipal Utilities
During discovery for a Claude-powered billing-inquiry assistant, a Vantage Municipal stakeholder describes desired capabilities in detail but never mentions what the system should refuse to do on its own. Which discovery category is being skipped, and why does this matter?
- A."What it must cost" — because budget always gets skipped first.
- B."What it must NOT do" — stakeholders almost never volunteer this category; it must be asked for explicitly, or boundaries surface expensively later. Correct answer
- C."What it must prove" — because audit requirements are always the last thing discussed.
- D."What it must do" — because capabilities are typically under-specified.
Why: Of the four discovery categories, "what it must NOT do" — boundaries and human-escalation triggers — is the one stakeholders almost never volunteer on their own. It has to be asked for explicitly ("what should Claude never decide by itself?"), or the gap surfaces later, typically during compliance review.
Question 5 · Domain 5 — Configuration and Knowledge Management
Tarragon Health Engineering
A developer at Tarragon Health writes new logging conventions into their personal `~/.claude/CLAUDE.md`. A teammate clones the repository the next day and Claude Code shows no sign of the new conventions. What is the root cause?
- A.`~/.claude/CLAUDE.md` is user-level configuration, personal to one developer and never version-controlled — team-wide standards must be committed in project-level `.claude/CLAUDE.md`. Correct answer
- B.CLAUDE.md files only take effect once an explicit `@import` statement is added to a second file.
- C.The teammate needs to run `/memory` once to pull down the new conventions.
- D.An org admin must approve new CLAUDE.md content before it reaches other developers.
Why: User-level CLAUDE.md is personal to one machine and is never version-controlled or shared. Team-wide conventions must live in the project-level `.claude/CLAUDE.md` (or root CLAUDE.md), which is committed and reaches every developer on checkout. This is the single most-tested distinction in this domain. Why the other options are wrong: CLAUDE.md files at project or directory level load automatically without `@import` — that directive is for pulling in extra files, not for basic loading. `/memory` is a diagnostic command that shows what's active; it does not fetch or activate anything. Nothing about CLAUDE.md distribution requires admin approval — committing the file is sufficient.
Question 6 · Domain 6 — Governance, Risk, and Responsible Use
Solandra Consumer Goods
A colleague at Solandra Consumer Goods says: "We don't need an acceptable use policy for Claude — its built-in safety handles everything." What is wrong with this statement?
- A.Nothing — Claude's trained safety behaviors cover every business scenario automatically.
- B.Claude's trained safety reduces obviously harmful, general-purpose output, but it does not know Solandra's specific rules about data handling or approvals — a request can clear Claude's safety checks completely and still violate the organization's own acceptable use policy. Correct answer
- C.It's only wrong for companies that handle health data or EU personal data.
- D.It's only wrong if Solandra operates outside the United States.
Why: Claude's trained safety behaviors apply automatically everywhere but only address obviously harmful, general-purpose output — they know nothing about a specific organization's rules on data handling, required approvals, or approved tools. Those two layers are separate, and the organization is always responsible for the second one, regardless of industry or geography.
What this practice exam covers
321 questions · 7 exam domains · 1 mock exam
- Domain 1 — Prompting and Task Execution · weight 14%
- Domain 2 — Output Evaluation and Validation · weight 21%
- Domain 3 — Product and Model Selection · weight 12%
- Domain 4 — Workflow Integration and Solution Design · weight 16%
- Domain 5 — Configuration and Knowledge Management · weight 12%
- Domain 6 — Governance, Risk, and Responsible Use · weight 15%
- Domain 7 — Troubleshooting and Optimization · weight 10%
More practice for this exam
Questions about this practice exam
- Are these the official Claude Certified Associate exam questions?
- No. They are practice questions written against the published exam domains. This site is not affiliated with or endorsed by Anthropic.
- How many practice questions are there?
- 321 questions across 7 exam domains including 1 full mock exam.
- Do I need an account to practise?
- No. You can take the practice questions without an account and your scores save in your browser. A free account adds the full answer review, gap analysis and a completion certificate.
- Do the questions explain the answers?
- Yes. Every question comes with an explanation of the correct answer.
Practice questions written against the published exam domains. Not official exam questions. Not affiliated with or endorsed by Anthropic.