Claude Certified Associate — Foundations

• What it tests
• Who it’s for
• The shape of the paper
• The seven domains
• How to prepare
• Common mistakes

What this exam actually tests

Look at the weights before you read anything else about this exam. Output Evaluation and Validation is the largest domain, at 21%. Prompting — the thing everyone assumes an entry-level AI certification is about — is 14%.

That ordering is the whole exam in one fact. This is a credential for business professionals using Claude without writing code, and the competency it weights most heavily is knowing when not to trust the output: spotting hallucinations and inconsistencies, fact-checking, judging when something needs human review before it leaves your desk.

Which is the correct emphasis, and unusual. Getting a plausible answer out of Claude is not hard. Knowing that a confident answer citing a specific regulatory subsection needs verifying before it reaches the compliance team is the skill that separates someone who can be trusted with the tool from someone who cannot.

Who it’s for

Business analysts, project managers, operations leads, marketing, communications, HR and education professionals, consultants, knowledge workers. Anthropic is explicit that no software-development or API experience is needed and there are no mandatory prerequisites.

What it does assume is a practical understanding of AI limitations — hallucinations, context constraints, data sensitivity. Not theory. The kind of understanding you get from having relied on Claude for real work and occasionally been let down by it.

If you write code and build with the API, this is the wrong exam: CCDV-F covers the developer track and CCAR-F the architecture one.

The shape of the paper

Sixty items in a hundred and twenty minutes — two minutes each, which is generous for standalone items and means pacing is unlikely to be your problem. Multiple-choice and multiple-response, with each item stating how many responses to select.

The pass mark is 720 on a 100–1,000 scale, the same bar as the Architect and Developer exams. Worth saying plainly: the “Associate” in the title describes the audience, not a lowered standard. Your score report gives percent-correct by domain alongside the total, and only the total determines the result.

The seven domains, and what each is really asking

Output Evaluation and Validation — 21%

The largest domain, with six published objectives — evaluating for accuracy and completeness, identifying hallucinations and bias, fact-checking, deciding when human review is required, editing and comparing outputs for an audience, and choosing output formats.

Prepare by building the habit rather than reading about it. For a week, every time you use Claude for something that matters, write down what you checked and how. That list is most of this domain.

Workflow Integration and Solution Design — 16%

Applying Claude to requirements and use cases, research and planning, supporting design and iteration, integrating it into existing workflows — and communicating its value and limitations to stakeholders. That last objective is a reminder that this credential is partly about being the person in the room who can say what the tool will not do.

Governance, Risk, and Responsible Use — 15%

Appropriate and inappropriate use cases, data sensitivity and privacy, organisational AI policy, ethical implications. If your employer has an AI usage policy, read it — this domain is largely the general form of the judgement it encodes.

Prompting and Task Execution — 14%

Effective prompts for business and technical tasks, task decomposition for complex requests, iteration, and adapting strategy by task type — analysis, research, drafting, brainstorming. Smaller than most candidates expect, and the task-type objective is the one to notice: the exam wants you to prompt differently for a research task than for a drafting one.

Product and Model Selection — 12%

Claude’s product surface — Projects, research mode, chat, artifacts — and the model family, Haiku through Sonnet to Opus, aligned to cost, speed and quality. Plus context limits and memory: when to restart a conversation, when to summarise, when to persist.

This is the most factual domain on the paper and the easiest to revise. Spend an hour with the product documentation and you have most of it.

Configuration and Knowledge Management — 12%

Configuring Projects with instructions and knowledge sources, managing uploaded knowledge and connectors, writing system-level instructions, and keeping all of it current.

Hands-on ground. If you have never set up a Project with its own instructions and documents, do that before sitting — the objectives here describe actions, and they are quicker to perform than to read about.

Troubleshooting and Optimization — 10%

Diagnosing underperforming prompts and poor outputs, adjusting on feedback, optimising workflows. The smallest domain, and largely the practical consequence of the other six.

How to prepare

Anthropic’s guidance is unusually actionable for this exam, and it is worth following as written rather than replacing it with reading.

• Self-assess against the 30 objectives above. They are published, they are specific, and the low-scoring ones are your study plan.
• Configure a real Project. Instructions, knowledge sources, connectors. That single exercise touches Configuration (12%) and much of Workflow Integration (16%).
• Practise validation deliberately. Take five Claude outputs you would normally have sent on, and verify every factual claim in them. Note what you found. This is the 21% domain.
• Work responsible-use judgement. Data sensitivity, appropriate use cases, when to escalate to a human. Less revisable than the rest, so give it real thought rather than a skim.

Three things candidates get wrong

Over-preparing on prompting. It is 14%. Output evaluation is 21% and governance 15%. The exam is more interested in your judgement about Claude’s output than in your ability to elicit it.

Assuming “Associate” means easier. Same 720 pass mark, same two hours, same scale as the Architect exams. Different subject matter, not a lower bar.

Studying without using the product. Half the blueprint describes actions — configure a Project, manage knowledge sources, compare outputs, select a model for a task. Reading about them is a poor substitute for an afternoon of doing them.

What Rubric will have for this exam

Mock exams built to the seven published weightings, with every answer citing the documentation behind it and dated against the current blueprint, and results reported per domain so a score tells you what to go and fix.

None of that exists yet. This page is the blueprint; the exams follow.