Claude Certification Guide CCAO-F

Quick reference

Every trap from all 20 lessons — 120 of them — grouped by domain. This is revision material, not a substitute for the lessons: the traps make sense once you've read the reasoning behind them. Print-friendly with Ctrl/Cmd + P.

Output Evaluation and Validation

21% · 13 questions · 4 lessons
  • 2.1 Treating polished writing as evidence of accuracy
  • 2.1 Checking only what the output says, not what it left out
  • 2.1 Asking Claude whether its own answer is correct and accepting the yes
  • 2.1 Spot-checking the easy parts and skipping the numbers
  • 2.1 Assuming a summary of an uploaded file can't contain errors
  • 2.1 Giving every output the same level of review regardless of use
  • 2.2 Assuming a specific-looking citation or statistic must be real
  • 2.2 Trusting an answer because Claude stated it with confidence
  • 2.2 Reading each section on its own and missing contradictions between them
  • 2.2 Looking for bias only in offensive wording
  • 2.2 Fixing a hallucination by asking Claude to try again without a source
  • 2.2 Assuming bias is gone because the prompt said to be neutral
  • 2.3 Treating a cited answer as already verified
  • 2.3 Fact-checking by asking Claude, or another AI tool, the same question
  • 2.3 Skipping review because the output came from a trusted Project or template
  • 2.3 Letting Claude make the final call on legal, medical, HR, or financial matters
  • 2.3 Sending any reviewer rather than one qualified for the content
  • 2.3 Treating every output as needing expert sign-off
  • 2.4 Sending the same output to every audience unchanged
  • 2.4 Equating more detail with a better output
  • 2.4 Comparing drafts by gut feel instead of stated criteria
  • 2.4 Putting comparison data in paragraphs instead of a table
  • 2.4 Using an Artifact for a quick one-off answer
  • 2.4 Letting editing for tone change the facts

Workflow Integration and Solution Design

16% · 10 questions · 3 lessons
  • 4.1 Treating Claude's list of requirements as complete because it looks thorough
  • 4.1 Asking Claude to summarize requirements when the job is to find gaps and conflicts
  • 4.1 Letting Claude resolve a stakeholder conflict by picking a side
  • 4.1 Starting with 'where can we use AI?' instead of the business problem
  • 4.1 Choosing a use case because it is impressive rather than frequent and well-defined
  • 4.1 Leaving out the source material and asking Claude to infer the requirements
  • 4.2 Using a cited research report without checking the claims the decision depends on
  • 4.2 Asking for the whole plan in one prompt instead of building it in stages
  • 4.2 Accepting process-improvement suggestions that are generic rather than tied to your process
  • 4.2 Starting a new chat for each revision and losing the design's history
  • 4.2 Treating Claude's first design draft as the final solution
  • 4.2 Letting Claude make the go/no-go decision on a plan
  • 4.3 Redesigning the whole workflow before proving value on one step
  • 4.3 Removing the human review step because early outputs looked good
  • 4.3 Promising stakeholders that Claude will be accurate every time
  • 4.3 Underselling Claude by describing only its risks
  • 4.3 Trying to build a cross-system automation yourself with workarounds
  • 4.3 Escalating to a developer for something a Project or connector already handles

Governance, Risk, and Responsible Use

15% · 9 questions · 3 lessons
  • 6.1 Letting Claude make the final call on hiring, firing, promotion or discipline
  • 6.1 Treating Claude's answer as final legal, medical or financial advice
  • 6.1 Assuming a task is safe because the output will be reviewed 'at some point'
  • 6.1 Refusing to use Claude at all for any task that touches a sensitive area
  • 6.1 Believing a disclaimer in the output transfers accountability
  • 6.1 Judging appropriateness by how good the output looks rather than by the stakes
  • 6.2 Uploading restricted data and telling Claude not to retain it
  • 6.2 Assuming internal use means policy doesn't apply
  • 6.2 Using a personal account or unapproved AI tool for work data
  • 6.2 Skipping the task entirely when redaction would make it safe
  • 6.2 Thinking removing names alone always anonymizes the data
  • 6.2 Connecting a work data source without checking it is approved
  • 6.3 Assuming Claude's output is neutral because it comes from a machine
  • 6.3 Hiding AI assistance where the audience or policy expects disclosure
  • 6.3 Blaming Claude for an error in work you sent
  • 6.3 Accepting Claude's output without review once it has been reliable for a while
  • 6.3 Fixing a biased output once without changing the prompt or process
  • 6.3 Believing disclosure alone makes unreviewed content acceptable

Prompting and Task Execution

14% · 8 questions · 3 lessons
  • 1.1 Assuming Claude knows your company, audience or history without being told
  • 1.1 Making a prompt longer instead of making it more specific
  • 1.1 Using one prompting style for every kind of task
  • 1.1 Asking for brainstormed ideas and judging them as if they were vetted recommendations
  • 1.1 Leaving the output format and length for Claude to guess
  • 1.1 Asking for analysis without supplying the data to analyze
  • 1.2 Cramming a multi-stage task into one enormous prompt
  • 1.2 Moving to the next step without checking the previous step's output
  • 1.2 Splitting a simple task into steps it doesn't need
  • 1.2 Mixing gathering, analyzing and writing in the same step
  • 1.2 Letting each step run in a separate chat with none of the earlier results
  • 1.2 Treating a task that needs automation or system integration as just a longer chat
  • 1.3 Giving vague feedback like "make it better"
  • 1.3 Regenerating the same prompt and hoping for a better answer
  • 1.3 Changing several things at once so you can't tell what helped
  • 1.3 Forgetting to say what was good, so it gets lost in the rewrite
  • 1.3 Iterating endlessly on a small wording issue faster to fix by hand
  • 1.3 Not carrying a winning refinement back into the reusable prompt

Product and Model Selection

12% · 7 questions · 3 lessons
  • 3.1 Pasting the same background documents into a new chat every week instead of using a Project
  • 3.1 Using research mode for a quick question that a normal chat answers in seconds
  • 3.1 Keeping a long document you'll keep revising buried in chat replies instead of an Artifact
  • 3.1 Assuming a Project shares its knowledge with chats outside the Project
  • 3.1 Treating a research report's citations as verified just because they are there
  • 3.1 Picking the feature with the most power rather than the one that fits the task
  • 3.2 Using the most capable model for every task because it is the best
  • 3.2 Choosing the cheapest model for a high-stakes, complex analysis to save money
  • 3.2 Assuming a bigger model removes the need to check its output
  • 3.2 Switching models to fix a problem that is really an unclear prompt
  • 3.2 Judging a model on one impressive or one bad answer
  • 3.2 Memorizing version numbers instead of the trade-off each family represents
  • 3.3 Continuing one endless chat for weeks of unrelated work
  • 3.3 Starting a fresh chat without carrying over the decisions already made
  • 3.3 Assuming Claude remembers a previous chat when memory isn't on and nothing was persisted
  • 3.3 Re-pasting the same reference material into every chat instead of persisting it
  • 3.3 Repeating the instruction louder when Claude drifts in a long chat instead of resetting
  • 3.3 Letting sensitive details persist in memory or Project files without thinking about who sees them

Configuration and Knowledge Management

12% · 7 questions · 2 lessons
  • 5.1 Pasting entire reference documents into the Project instructions
  • 5.1 Uploading every loosely related file in case it helps
  • 5.1 Writing instructions that only describe the topic, not how to behave
  • 5.1 Leaving conflicting versions of the same document in Project knowledge
  • 5.1 Putting one-off task details into instructions that apply to every chat
  • 5.1 Sharing a Project with the team before testing it on real requests
  • 5.2 Connecting every account and folder available just in case
  • 5.2 Uploading a one-time copy of a document that changes every week
  • 5.2 Replacing a knowledge file without removing the outdated version
  • 5.2 Changing shared Project instructions without telling the people who use them
  • 5.2 Assuming a configuration stays correct forever once it works
  • 5.2 Treating a connector as a reason to skip checking what Claude retrieved

Troubleshooting and Optimization

10% · 6 questions · 2 lessons
  • 7.1 Adding emphasis or capital letters instead of finding the cause
  • 7.1 Blaming the model when the source file is outdated or wrong
  • 7.1 Switching to the most capable model as the first fix for every problem
  • 7.1 Continuing a very long, drifting conversation instead of starting fresh with a summary
  • 7.1 Rewriting everything at once so you can't tell what fixed it
  • 7.1 Declaring the problem solved after one good result
  • 7.2 Optimizing for speed by dropping the human review step
  • 7.2 Using the most capable model for every task in the workflow
  • 7.2 Re-explaining the same context in every new chat
  • 7.2 Ignoring feedback from the people who use the output
  • 7.2 Adding more steps and tools when a simpler workflow would do
  • 7.2 Measuring success by time saved alone