CCAO-F domain - 14% of the exam

Prompting and Task Execution

Prompting and Task Execution is 14% of the CCAO-F exam. These are the objectives it covers, each with practice questions, with every answer explained.

The domain in numbers

34
Practice questions
4
Objectives
15
Easy
19
Medium

Where people go wrong

  • Regenerating instead of fixing the prompt

    The wrong answers regenerate the response, rerun it unchanged, or pick the best of several drafts. Regenerating reruns the same prompt, so Claude has no new information and each version shares the same faults: the quiz stays too hard, the thank-you email stays long and formal, and the newsletter stays stiff. The correct answer adds what the prompt never said, such as the class's level, a description of the readers, approved replies to follow or the voice wanted.

    See it in a sample question
  • Calling numbered steps in one prompt decomposition

    The wrong answers keep a large deliverable inside a single request, whether by resending the full report prompt or by numbering steps within one prompt and calling that decomposition. Both produce everything at once with no checkpoint, so a misread survey question in step one shapes the problems and action plan built on it. Decomposition sends each step separately and reviews it before the next, starting with an agreed outline for a long report.

    See it in a sample question
  • Borrowing the strategy of a different task type

    The wrong options apply the approach of another kind of task. Writing the briefing first suits drafting, but research needs cited findings whose sources are then opened. Reasoning and evidence suit analysis, but brainstorming needs a broad set of distinct ideas with no ranking yet. Many versions to choose from suit brainstorming, but a letter whose facts and tone are already settled needs one briefed draft refined through feedback.

    See it in a sample question

Objectives in this domain

Sample questions from this domain

Free samplePrompting and Task Executioneasy

A communications officer at a regional charity asks Claude to draft a thank-you email to donors who gave to a winter appeal. The draft is accurate but runs to 400 words and reads like a corporate report. The email must go out this afternoon, and most recipients are long-standing supporters who respond best to a short, personal note. What should the officer do next? Select TWO.

  • AReply with specific changes: cut it to about 150 words, use a warmer personal tone, and keep the appeal total. Correct
  • BRegenerate the response with the same prompt a few times and send whichever version reads best to her.
  • CTell Claude who the readers are: loyal supporters who prefer a brief, personal note to a formal update. Correct
  • DAsk Claude to score its own draft for warmth out of ten and keep revising until it rates itself highly.
  • ESwitch to the most capable model available and rerun the original prompt without changing any wording.
When a draft misses the mark, give Claude specific changes and the audience context it lacked instead of regenerating or escalating the model. Claude can only correct what it is told about. Naming the target length, tone and the facts to keep, and describing who will read the email, adds the information the first prompt lacked, so the next draft moves towards the goal in one step. Regenerating, self-scoring or changing model leaves that missing information out.

Why A is correct: This names exactly what to change (length and tone) and what to keep (the appeal total), so the next draft can fix the problems without losing the accurate content. A targeted correction is also the fastest route with an afternoon deadline.

Why B is wrong: Regenerating feels quick and offers choice, but an unchanged prompt gives Claude no new information about length or tone, so each version is likely to share the same faults. It relies on luck instead of direction.

Why C is correct: The draft missed because Claude did not know the audience. Describing the readers gives Claude the context it needs to choose an appropriate length, register and warmth, which fixes the cause of the corporate tone rather than one symptom.

Why D is wrong: A self-score sounds like a measurable quality check, but Claude's rating of its own work is not an independent signal of what donors will feel. The officer already knows what is wrong and should say so directly.

Why E is wrong: A stronger model can seem like the fix for a weak draft, but the problem is missing guidance about audience and length, not reasoning capacity. The same vague prompt is likely to produce the same long, formal email.

Free samplePrompting and Task Executionmedium

A marketing manager at a home-furnishings retailer must deliver a 12-page quarterly campaign review to the commercial director by Friday. She pasted all the campaign data into one chat with a single prompt asking for the complete report. The result covered every heading she listed, but each section was only two generic sentences with no analysis of the figures. What should she do next?

  • AAsk for an outline first, agree it, then have Claude draft one section at a time and review each before moving on. Correct
  • BPaste the same prompt again into a new chat and pick the strongest of several fresh versions of the report.
  • CAdd a line to the original prompt telling Claude to be detailed and thorough, then run the full report request once more.
  • DSwitch to the most capable model and resend the single prompt, since deeper reasoning will fill out every section.
When one prompt for a long, multi-part deliverable returns thin sections, break it into an agreed outline and section-by-section drafts reviewed in turn. A single prompt for a long report asks Claude to plan, analyse and write every part at once, so effort is spread thinly and errors in direction are only visible at the end. Agreeing an outline first and then drafting and reviewing one section at a time focuses each step and lets each reviewed section inform the next.

Why A is correct: Thin, generic sections across a long deliverable are the typical sign of one giant prompt. Working section by section gives each part Claude's full attention and lets her correct the direction early, which still fits a Friday deadline.

Why B is wrong: Regenerating feels quick when a deadline is close, but an unchanged prompt produces the same kind of result. Nothing about the request has changed, so the sections will stay thin.

Why C is wrong: Asking for more detail is a reasonable instinct, but the report is still one request covering twelve pages at once. She gets no chance to review or steer any section before the whole thing is produced.

Why D is wrong: A more capable model can help with hard reasoning, so this is tempting. The problem here is the structure of the request, and defaulting to the top model does not replace breaking a large deliverable into reviewed steps.

Free samplePrompting and Task Executionmedium

A communications officer at a local council is preparing a briefing for councillors on how other councils have run e-scooter trials. The council's policy reviewer will send back any briefing containing a factual claim that cannot be traced to a source. The officer has a week. Which approach best fits this research task?

  • AAsk Claude in a chat what other councils did, then ask it to add a list of sources at the end
  • BAsk Claude to draft the briefing in an engaging tone for councillors, then fill in facts later
  • CAsk Claude to confirm that each of its findings is accurate before the briefing goes to review
  • DUse research to gather cited findings, then open each cited source to confirm what it says Correct
For research tasks, prompt for findings with citations and then check the cited sources before relying on any claim. Research output is only as good as the sources behind it. Gathering findings with citations links each claim to where it came from, and opening those sources confirms that the citation really supports the claim. Neither a source list added afterwards nor a self-check by the model gives a reviewer something independent to trace.

Why A is wrong: Asking for sources sounds like it meets the reviewer's requirement. Sources added after the fact may not support the specific claims, and a listed source is not verified until someone opens it, so traceability is not actually established.

Why B is wrong: Starting with the audience and tone is right for a drafting task, and councillors are the readers. This is a research task first, though, and writing before the facts are gathered and sourced invites unsupported claims that the reviewer will reject.

Why C is wrong: A self-check feels like a cautious extra step. A second pass by the same model is not independent evidence and produces no source a reviewer can trace, so it does nothing to meet the policy reviewer's standard.

Why D is correct: Research tasks call for sources. Using research returns findings with citations, and opening each source confirms that it supports the claim, which gives the policy reviewer a traceable source for every fact. A week is ample time for this check.

Other domains in this exam

See also the CCAO-F cert hub, the study guide, and the cheat sheet.

Examworthy is not affiliated with or endorsed by Anthropic. Original, blueprint-aligned practice material only.