Claude Certification Readiness Report 2026
Anthropic launched its Claude certifications without a published pass rate, so candidates have nothing to measure themselves against. This report is a first benchmark: how people score the first time they sit a Claude certification practice diagnostic, which official domains they miss, and how far a typical candidate starts from the pass line.
Sample 277 first attempts at the free Claude Certified Associate (CCAO-F) diagnostic and 37 at the Claude Certified Developer (CCDV-F) diagnostic, from 343 distinct candidates, 2026-09-25 to 2026-11-14. Practice data from PrepClubs users, not Anthropic's official results.
1Key findings
of first attempts at the CCAO-F practice diagnostic reached a pass-level score
11 of 277 candidates scored 20 or more of 25. Even on a looser 72% line, only 7.6% would clear it.
- The median first attempt scores 44%. That is 11 of 25 questions, about 9 questions short of pass level.
- 63.2% of candidates answer fewer than half correctly on their first sitting.
- Troubleshooting and Optimization is the domain most people miss. First attempts got 24.7% of its questions right, and it was the single weakest domain for 43.5% of candidates.
- Governance is the second gap. Governance, Risk, and Responsible Use scored 34.2%, and its hardest question, on ethical implications, was answered correctly by 9.4% of candidates.
- The Developer exam looks the same. In a small sample of 37, 2 reached pass level (5.4%) and the median was 52%.
2Who would pass on a first attempt
Each candidate counts once, on their first completed attempt. Pass level means 20 or more of 25 correct, explained in Method.
| Exam | Candidates | Median | Middle half | Pass level | At 72%+ |
|---|---|---|---|---|---|
| CCAO-F | 277 | 44% | 28% to 56% | 4% (11) | 7.6% (21) |
| CCDV-F (small sample) | 37 | 52% | 32% to 64% | 5.4% (2) | 10.8% (4) |
Not reported: CCAR-F, 20 first attempts; CCAR-P, 9 first attempts. Below the 30-attempt floor, a percentage would move by several points with each new candidate.
Most candidates start well below the line. The diagnostic is usually the first thing people do after deciding to sit the exam, so this is a picture of where preparation starts. It says little about how candidates do on exam day after weeks of study.
3How first-attempt scores are spread
277 candidates. Bars at or above the dashed line reached pass level. Tenth percentile 20%, ninetieth 68%.
- The bulk sits between 28% and 56%. That range is 7 to 14 questions correct out of 25.
- Only 21.3% reach 60%. A candidate at 60% on a first attempt is in roughly the top fifth of this sample and still five questions from pass level.
- Nobody needs to rush. The median first attempt took 35.1 minutes for 25 questions. Time pressure does not explain these scores.
4Accuracy by official domain
The CCAO-F exam guide lists seven domains. The diagnostic samples every one of them, and the share of each domain answered correctly across first attempts is below.
277 first attempts. The question count is per diagnostic, so a short bar on a two-question domain rests on fewer items than one on a five-question domain.
- Troubleshooting and Optimization: 24.7%. These questions describe a setup that is falling short, such as a long procedure pasted into every new chat or a cheaper model missing detail, and ask for the best change. The right answer turns on plan features and settings as much as on prompting, which everyday users of Claude often have not had to learn.
- Governance, Risk, and Responsible Use: 34.2%. The two hardest single questions on the form are here, on ethical implications (9.4%) and organizational policy and governance (15.9%).
- The rest cluster near 50%. Configuration and Knowledge Management (49.8%), Prompting and Task Execution (50.2%), Output Evaluation and Validation (50.5%), Product and Model Selection (50.7%).
- The easiest topic is data privacy. Data Sensitivity and Privacy questions were answered correctly 70.8% of the time, the highest on the form.
| Skill | Domain | Correct |
|---|---|---|
| Ethical Implications | Governance, Risk, and Responsible Use | 9.4% |
| Organizational Policy and Governance | Governance, Risk, and Responsible Use | 15.9% |
| Feedback-Driven Adjustment | Troubleshooting and Optimization | 21.3% |
| Task Decomposition | Prompting and Task Execution | 21.3% |
| Human Review and Verification | Output Evaluation and Validation | 21.7% |
Each row is one question, answered by all 277 first-attempt candidates.
5Claude Certified Developer: a small sample
37 candidates have completed a first CCDV-F diagnostic. That clears the 30-attempt floor but not by much, so read these as directional.
- Pass level: 2 of 37 (5.4%). Median 52%, middle half 32% to 64%.
- Weakest domain: Prompt and Context Engineering, 31.5% correct. Next: Claude Code (43.2%) and Agents and Workflows (48%).
- Strongest domain: Eval, Testing, and Debugging, 56.8%.
- Architect (CCAR-F, 20 first attempts) and Architect Professional (CCAR-P, 9) are below the floor and not reported. They will be added once each passes 30.
| Domain | Questions | Correct |
|---|---|---|
| 6. Prompt and Context Engineering | 3 | 31.5% |
| 3. Claude Code | 1 | 43.2% |
| 1. Agents and Workflows | 4 | 48% |
| 7. Security and Safety | 2 | 48.6% |
| 5. Model Selection and Optimization | 4 | 51.4% |
| 2. Applications and Integration | 8 | 52.7% |
| 8. Tools and MCPs | 2 | 54.1% |
| 4. Eval, Testing, and Debugging | 1 | 56.8% |
37 first attempts. Domains with one question per diagnostic rest on 37 answers each.
6By country
Country comes from where the account signed up. Only two countries have more than 30 CCAO-F first attempts.
| Country | Candidates | Mean | Median | Pass level |
|---|---|---|---|---|
| India | 125 | 39.4% | 36% | 5 |
| United States | 48 | 48.9% | 52% | 1 |
| All other countries, or not recorded | 104 | 47% | 48% | 5 |
Pass-level counts are too small to compare as rates between countries. The means describe who chose to practise here, not national ability.
7Method
- Source. Completed attempts at the free 25-question PrepClubs diagnostics for each Claude certification, 2026-09-25 to 2026-11-14. Questions are original PrepClubs items written to Anthropic’s published domain outline for each exam.
- One attempt per candidate. Only each candidate’s first completed attempt counts. Re-sits are the subject of Report 2026-06.
- Pass level. Anthropic passes candidates at a scaled 720 of 1,000 and does not publish how raw answers map to that scale. PrepClubs uses 78% correct as its practice target, a margin above the scaled line, and this report uses the same rule: on 25 questions, 20 correct. A looser 72% line (18 correct) is shown for comparison.
- Re-graded. Every attempt was re-scored against the current answer key. All of them matched the stored score.
- Floor. No exam is reported below 30 first attempts.
| Step | Attempts |
|---|---|
| Claude diagnostic and exam attempts on record | 449 |
| Dropped: faster than 10 seconds a question | 24 |
| Dropped: fewer than half the questions answered | 37 |
| Analysed attempts | 388 |
| First attempts (one per candidate) | 343 |
Staff and test accounts are excluded before counting.
8What these numbers are not
- Not Anthropic’s results. This is practice-test data from PrepClubs users on PrepClubs questions. PrepClubs is not affiliated with or endorsed by Anthropic, and no figure here is an official pass rate.
- A starting point, not an outcome. Most candidates take the diagnostic before they study. A low first score is the normal case and says nothing about how they will do after preparing.
- Few questions per domain. Each diagnostic has 2 to 5 questions per domain. Domain accuracy describes those questions as much as the domain.
- Self-selected. People who found a practice site and chose to test themselves are not a random sample of candidates.
9Common questions
- What percentage of people pass the Claude Certified Associate exam?
- Anthropic does not publish a pass rate. On PrepClubs, 4% of 277 first attempts at the free CCAO-F practice diagnostic reached pass level (20 or more of 25 correct). That is a practice result on our own questions, not Anthropic's figure, and it describes people at the start of their preparation.
- What is the hardest domain on the Claude Certified Associate exam?
- In our data it is Troubleshooting and Optimization. First attempts answered 24.7% of its questions correctly, the lowest of the seven domains, and it was the weakest domain for 43.5% of candidates. Governance, Risk, and Responsible Use (34.2%) and Workflow Integration and Solution Design (39.1%) came next.
- What score do you need to pass a Claude certification?
- Anthropic's exam guides set the pass mark at a scaled 720 on a 100 to 1,000 scale. A scaled score is not a raw percentage and Anthropic does not publish the conversion, so PrepClubs aims at 78% correct on its practice exams as a margin above that line. On a 25-question diagnostic the first score at or above 78% is 20 correct.
- How long does the free Claude diagnostic take?
- The median first attempt at the CCAO-F diagnostic took 35.1 minutes for 25 questions; the middle half of candidates took between 24.9 and 44.2 minutes.
Check your own readiness
See where you land against these 277 candidates
The free CCAO-F diagnostic is the same 25 questions measured here. You get your score against the pass line and your weakest domain in about 35 minutes. The full-length exams cover every domain at exam length when you are ready to close the gap.
Suggested citation
“On a first attempt at a Claude Certified Associate (CCAO-F) practice diagnostic, 4% of candidates (11 of 277) reached a pass-level score, and the median candidate answered 44% correctly. The weakest official domain was Troubleshooting and Optimization, at 24.7% correct. PrepClubs Research, Report 2026-05.”
Download the figures as CSVMethod and standardsAsk us to check a figure
Reproduce any figure or table with attribution to PrepClubs Research and a link to this page. The underlying per-question records are not published: joined to an attempt they are re-identifiable at this sample size.
Other reports in this series
- PCR-2026: The State of Timed Assessment 2026
Every finding of the year in one document.
- 2026-01: How much of a timed aptitude test is never answered
Abandonment by position on speeded and generous-clock forms.
- 2026-02: What an extra thirty seconds on a question actually buys
The return on deliberation, measured inside the same question.
- 2026-03: Which aptitude tests are built as a difficulty ramp
Accuracy by position, with selection controlled out.
- 2026-04: What candidates actually score on aptitude practice tests
Percentile bands per form, and what a raw score is made of.
- 2026-06: How long it takes to get ready for a Claude certification exam
What repeat attempts show, and why a re-sit score is not a readiness signal.
- 2026-07: How people score on a first practice attempt, by test
Median score, time used and unanswered questions on the first practice attempt, for each test with at least 50 first attempts.