These twenty questions come from the free CCAO-F Form 1 on this site, picked so that every exam domain appears. The Associate exam is for people who use Claude at work rather than build on it, so the questions are about judgment in everyday use, not code.
Answer each one before you open it. Then read every explanation, not just the one for the right answer: knowing why the tempting option loses is what carries over to the questions you have not seen.
An associate asks Claude to "write something about our new returns policy" and gets bland, generic text back. Which revision to the request is most likely to improve the first draft?
- AState the audience, purpose, length, and paste the policy text the draft must reflect.
- BAdd an instruction to be specific and detailed and to avoid generic language.
- CSend the same request three times and combine the strongest paragraphs from the replies.
- DAsk Claude to explain how it would approach the task before it writes anything.
Show the answer and why each option wins or loses
ACorrect
Correct. The draft is bland because the request supplied no audience, no purpose, no length, and no source material. Adding those four things gives Claude something concrete to write from, which is what turns generic prose into a usable draft.
B
This does signal a quality bar, and adjectives like "specific" occasionally sharpen phrasing. But an instruction to be specific cannot supply facts Claude was never given; without the policy details the second draft is generic in more words.
C
Sampling several outputs is a legitimate technique when you want variety to choose from, and it can surface a better opening line. Here every generation starts from the same information-poor request, so all three drafts share the same defect.
D
Asking for an approach first is useful on genuinely ambiguous work and can expose a misunderstanding early. It costs a round trip here without addressing the cause: the request is underspecified, not misunderstood.
The principle
A weak first draft is usually a symptom of a weak brief. Supply context, audience, purpose, format, and source material before adding adjectives about quality.
74% of 228 answers on this site got it right on the first try.
At Brightpath Lending an associate needs 40 loan-decline letters condensed into one-line internal notes. The first ten notes come back factually accurate but inconsistent in tone and sentence structure, even after the associate added three more sentences of instruction. What should the associate do next?
- AAdd explicit written rules covering tone, verb tense, and the order of sentences.
- BBreak the batch into groups of ten and submit each group as a separate request.
- CPaste two finished notes written the way they want and ask for the rest to match.
- DAsk Claude to review its own notes and rewrite the ones that differ from the others.
Show the answer and why each option wins or loses
A
More rules do tighten output, and naming tense specifically would fix one visible symptom. The associate has already tried this once without success: format consistency is hard to describe in prose and easy to demonstrate, so another round of rules is the slower path.
B
Smaller batches genuinely help when quality degrades across a long output. Inconsistency appeared inside the first ten notes, so the batch size is not what is producing it.
CCorrect
Correct. Two worked examples fix tone, length, and structure at once, in a form that prose instructions struggle to convey. The desired pattern becomes something Claude can copy rather than infer.
D
Self-review can catch outliers and is a reasonable final pass. Without a stated standard, "consistent with the others" is ambiguous, and Claude may converge on the wrong house style.
The principle
When output quality fails on form rather than facts, showing beats telling. A couple of examples of the finished article usually outperform another paragraph of instructions.
29% of 185 answers on this site got it right on the first try.
An associate asks Claude in a single request to read six survey files, identify themes, rank the themes by frequency, and draft an executive summary. The summary reads fluently but the rankings look wrong. What is the best next step?
- AAsk for the executive summary again, requesting that the rankings be double-checked first.
- BRun theme extraction and ranking as separate steps and verify each before drafting.
- CProvide a template for the executive summary so the output follows a fixed structure.
- DMove the six files into a Project so they stay available across future conversations.
Show the answer and why each option wins or loses
A
Asking for a self-check sometimes catches arithmetic slips and is cheap to try. It still buries the ranking inside a request whose visible product is prose, so a wrong ranking can once again pass unexamined.
BCorrect
Correct. The later steps depend on the earlier ones, and only the final prose was inspectable. Separating extraction from ranking creates a checkpoint at the step that actually failed, before it is baked into a summary.
C
A template does improve the shape and consistency of the summary, which is worth doing eventually. It constrains presentation, not the analysis underneath, so the rankings would remain wrong in a tidier layout.
D
Projects are a sound way to keep reference material available for repeated work. Storage location is unrelated to why an intermediate analytical step went unchecked.
The principle
When steps depend on each other, decompose so each intermediate result can be inspected. One giant request hides the failure inside a polished final artefact.
70% of 135 answers on this site got it right on the first try.
Claude drafts a competitor comparison and states three market-share figures in a confident tone. The associate already has the vendor's published annual report open on screen. What should the associate do before circulating the draft?
- ACheck the three figures against the annual report and correct any that differ.
- BAsk Claude to rate how confident it is in each figure and flag the weak ones.
- CRe-run the same prompt twice and keep the version whose figures agree across runs.
- DAdd a note to the draft saying figures are AI-generated and may need confirmation.
Show the answer and why each option wins or loses
ACorrect
Correct. An authoritative source is already at hand, so each claim can be settled directly. Confidence in the wording carries no evidence; the report does.
B
This surfaces the model's own uncertainty signal, which can occasionally hint at weak spots, but a self-rating is not evidence. A stated figure can be wrong while the model reports high confidence.
C
Repeating the prompt tests only whether the output is stable, not whether it is true. A consistently repeated wrong number looks reassuring and remains wrong.
D
A disclosure label manages expectations and is sometimes appropriate, but here it hands the verification burden to readers when the source needed to resolve it is already open.
The principle
Fluency and confidence are properties of the writing, not of the facts. When an authoritative source is available, checking the specific claim against it is the only step that establishes accuracy.
86% of 99 answers on this site got it right on the first try.
An associate asks Claude for a summary of a 40-page policy document and receives a fluent one-page result. Which check most directly tests the summary for completeness rather than accuracy?
- AVerify that each statistic in the summary matches the number printed in the source document.
- BList the document's section headings and check each required topic appears in the summary.
- CRead the summary aloud to hear whether the argument holds together on its own terms.
- DAsk a colleague whether the summary's tone suits the leadership audience it is written for.
Show the answer and why each option wins or loses
A
This is a sound accuracy check and worth doing, but it tests whether what is present is right, not whether anything is missing.
BCorrect
Correct. Completeness is a coverage question: mapping the source's structure against the output reveals omitted topics, which fluent summaries hide well.
C
This tests readability and internal coherence, both useful before circulation, but a smooth summary can omit an entire section without sounding wrong.
D
Audience fit matters for the final draft, and a colleague's read is a real check, but tone tells you nothing about which content was dropped.
The principle
Accuracy and completeness fail differently: accuracy fails on what is stated, completeness on what is absent. Only a comparison against the source's full scope catches omissions.
57% of 97 answers on this site got it right on the first try.
At Brightwater Housing Trust, an associate uses Claude to turn a 30-page grant report into a two-page board summary. The summary reads well, and the board must act on it next week. Which two checks best confirm the summary is accurate and complete before it goes out?
Choose 2
- AConfirm the summary keeps the report's original section order.
- BVerify every figure and date in the summary against the report.
- CShorten the sentences so board members can read the summary quickly.
- DAsk Claude to confirm that nothing important was left out of it.
- ECheck that every decision the board must make appears in the pages.
Show the answer and why each option wins or loses
A
Preserving order can help readers follow a familiar document, but a reordered summary can still be fully accurate, and a same-order summary can still drop content.
BCorrect
Correct. Figures and dates are the claims most likely to be wrong and most consequential in a funding context, and the report settles each one directly.
C
Tightening prose improves readability for a busy board, but it addresses style rather than whether the content is right or whole.
D
This produces a confident-sounding assurance, and it can occasionally surface a gap, but the model is judging its own work with no independent reference.
ECorrect
Correct. The summary exists so the board can act, so completeness here means every required decision is present. That is the coverage test that matters for this purpose.
The principle
Completeness is judged against the output's purpose, not against a generic checklist. Define what the reader must be able to do, then verify the facts that decision rests on.
69% of 51 answers on this site got it right on the first try.
Meridian Housing Trust's grants team writes funding applications every month. Each one follows the same tone rules and reuses the same three background documents, and four staff members draft them. Today they paste those rules and documents into a new chat each time. Which change best fits how they work?
- ASet up a shared Project holding the reference material and house style, then work inside it.
- BDraft each application as an Artifact so reviewers can edit the wording in place before submission.
- CKeep starting new chats, pasting the tone rules and documents at the top of each one.
- DRun each application through research mode so drafts pull in current funding-sector sources automatically.
Show the answer and why each option wins or loses
ACorrect
Correct. The defining facts are repetition, shared standards, and multiple people: the same reference material is needed in every session by four drafters. A Project persists that material once so every conversation in it starts already grounded, which is exactly what pasting was simulating by hand.
B
An Artifact does give the team a clean, editable draft they can revise and hand to reviewers, and it may well be useful for the finished application. But it carries nothing between sessions — the background documents and tone rules would still have to be re-supplied every month, so it solves the wrong half of the problem.
C
This preserves the current output quality, since the model does receive the rules and documents. It just leaves the actual cost in place: four people repeating the same setup work monthly, with drift whenever someone pastes a stale version.
D
Research mode is the right tool when an answer depends on current external sources the team does not already hold. Here the inputs are fixed internal documents the team owns, so external retrieval adds noise and does nothing about the repeated setup.
The principle
Reach for a Project when the same context must be reused across sessions or across people. Recurring setup work pasted by hand is the clearest signal that context belongs in a persistent place instead of a conversation.
88% of 65 answers on this site got it right on the first try.
A communications associate must produce a one-page briefing on a regulation that changed last week, and the wording will be revised live with two colleagues. Which two product features fit this piece of work?
Choose 2
- AA separate conversation per colleague, so each set of revised wording stays isolated from the others.
- BResearch mode, to gather and cite up-to-date sources on what the new rule now requires.
- CA Project, so the finished briefing is stored for other teams to reuse next quarter.
- DOne long chat in which every revised version is pasted back in as fresh text.
- EAn Artifact, so the one-page draft sits in an editable panel the group reshapes as they talk.
Show the answer and why each option wins or loses
A
Isolating feedback does keep one reviewer's comments from anchoring another's. It also fragments a single document into three drifting copies, which is the opposite of what live joint revision needs.
BCorrect
Correct. The regulation changed last week, so the facts postdate what any model holds internally. Research mode is the surface built for pulling in and citing current external sources.
C
Storing the briefing for later reuse is genuinely worth doing, and a Project is where that belongs. It is not what this task needs, though: nothing here is reused across sessions, and it neither sources the facts nor supports live revision.
D
Pasting revisions back into the chat keeps everything in one thread and does produce a final version. Versions get buried in the transcript, and there is no single current document for three people to work on.
ECorrect
Correct. An editable, shareable deliverable that colleagues revise as the discussion proceeds is precisely the Artifact case — the document stays in one place while the conversation refines it.
The principle
Match the surface to what the work actually demands. Facts newer than the model point at research mode; a deliverable that gets edited and shared points at an Artifact, and the two combine cleanly in one session.
39% of 28 answers on this site got it right on the first try.
An associate must sort 400 short product blurbs into three tags by tomorrow morning, under a tight per-item budget. The rules are simple and no written reasoning is required. Which tier fits best?
- AOpus, because the sheer number of items makes maximum accuracy the dominant concern here.
- BHaiku, because the task is simple and repetitive, so speed and cost decide the choice.
- CSonnet, because a mid-tier model is the safe default for essentially every kind of workload.
- DOpus for the first hundred items, then Haiku once the tagging pattern has stabilised.
Show the answer and why each option wins or loses
A
High volume is a real reason to care about errors, since mistakes multiply. But volume plus a tight budget argues for the fastest, cheapest tier that clears the bar, and simple three-way tagging is well within Haiku's reach.
BCorrect
Correct. Haiku is the fastest and least expensive tier, built for high-volume, low-complexity work. Short blurbs, three fixed tags, and no reasoning output describe exactly that shape.
C
Sonnet is a sensible middle ground and would certainly produce correct tags. Treating it as a universal default skips the actual judgment: pick the cheapest tier that meets the requirement, and here a cheaper one does.
D
Sampling a stronger tier to establish a pattern is a legitimate calibration tactic. As a production plan it pays premium rates for work that never needed them, and the tags do not become easier after item one hundred.
The principle
Tier choice follows task complexity, not task size. High volume with simple rules and budget pressure points to the fastest, cheapest tier that meets the quality bar.
88% of 58 answers on this site got it right on the first try.
A department head asks an associate to use Claude to produce faster first drafts of grant reports. While mapping the process, the associate finds drafting takes two days but approval routing takes three weeks. What should the associate do next?
- ABuild the drafting assistant the department head asked for, since faster first drafts are the outcome she named.
- BShare the timing evidence with the department head and propose scoping the work around the routing delay.
- CAsk Claude to draft a replacement approval policy that removes two of the three review stages before the next cycle.
- DHold off on any Claude work until the department publishes a fully documented map of the process end to end.
Show the answer and why each option wins or loses
A
Delivers exactly what was asked and would produce usable drafts, but it improves a two-day step inside a three-week cycle, so the stakeholder sees almost no change in the outcome they care about.
BCorrect
Correct. Requirements analysis exists to test the stated request against observed evidence; surfacing the routing bottleneck lets the department head redirect the work while keeping the decision theirs.
C
Uses Claude for a real planning task, but the associate has no mandate to redesign an approval structure, and a drafted policy sidesteps the conversation with the person who owns it.
D
Documentation would help, but the associate already has enough evidence to raise the issue; waiting for a formal map stalls work that could start now.
The principle
Requirements work means confirming that the requested improvement actually moves the outcome. When evidence contradicts the stated request, bring the evidence back to the stakeholder rather than silently building or silently redesigning.
60% of 72 answers on this site got it right on the first try.
An associate is using Claude to turn a vague stakeholder request into a workable use case. Which two moves make the resulting requirements most reliable?
Choose 2
- AAsk Claude to select the single best use case so the team can avoid a long debate.
- BGive Claude real examples of the current inputs and outputs so it can name concrete gaps.
- CAsk Claude to estimate the dollar savings of the project and put that figure in the brief.
- DHave Claude write the requirements in technical language so engineering can build from them directly.
- EHave Claude list its assumptions and open questions, then check those with the stakeholder.
Show the answer and why each option wins or loses
A
Produces a fast decision and a clear recommendation, but selection is the stakeholder's call, and a single option hides the tradeoffs the requirements were supposed to expose.
BCorrect
Correct. Real artifacts ground the analysis in what the work actually looks like, so the gaps Claude names are the team's gaps rather than generic ones.
C
Gives executives a number they often want, but a savings figure invented without measured baselines becomes a commitment the associate cannot support later.
D
Helps a future handoff, but translating a vague request straight into build-ready specification skips the confirmation step and locks in unverified assumptions.
ECorrect
Correct. Making assumptions explicit turns them into questions a human can confirm or correct, which is the cheapest place to catch a misread requirement.
The principle
Claude is strongest at requirements work when it is fed real material and made to expose its own uncertainty. Confirmation with the stakeholder, not fluency of output, is what makes requirements dependable.
82% of 33 answers on this site got it right on the first try.
Northgate Libraries wants to shorten its six-week vendor selection cycle. An associate has Claude summarise last year's evaluations and finds most of the delay sits in reconciling scoring spreadsheets from five branches, not in reading proposals. How should the associate direct the next round of work?
- AHave Claude read and summarise each incoming proposal, since reading is the most effortful task.
- BAsk Claude to design a new procurement policy for the library system to adopt this cycle.
- CHave Claude generate a full replacement scoring rubric before the next selection cycle opens.
- DHave Claude consolidate the five branch scorecards into one comparison and flag disagreements.
Show the answer and why each option wins or loses
A
Saves genuine reading effort and would be welcomed by reviewers, but the associate's own analysis showed reading is not where the six weeks go.
B
Addresses process design, yet a new policy is a governance decision beyond the associate's remit and far larger than the reconciliation problem found.
C
A rubric could reduce future scoring variance, but it does not resolve the disagreements already sitting in this cycle's spreadsheets.
DCorrect
Correct. It targets the step the analysis identified, and flagging disagreement gives reviewers the specific conflicts to settle rather than a merged number that hides them.
The principle
Process optimisation should follow the evidence about where time is lost, not intuition about which task feels hardest. Let the analysis choose the target before choosing the intervention.
74% of 58 answers on this site got it right on the first try.
Meridian Housing Trust's grants team shares one Project for funder reports. Every chat starts the same way: someone pastes three paragraphs covering tone, required sections, and the funder's preferred terminology, then attaches the same 40-page style guide. Reports still come out inconsistent between staff. What should the team change?
- AAsk everyone to keep the three paragraphs in a shared document so the pasted wording stays consistent.
- BStart a fresh chat for every funder report so earlier drafts cannot influence the tone of the new output.
- CSet the standing guidance as Project instructions and load the reference document into Project knowledge.
- DPaste the guidance once at the top of one long chat and keep every funder report in that same conversation.
Show the answer and why each option wins or loses
A
This does reduce drift in the wording people copy, and a shared source of truth is a reasonable habit. It still leaves the guidance outside the Project, so it applies only when someone remembers to paste it, which is the actual failure here.
B
Fresh chats do keep an earlier draft's quirks from carrying into a new report, which is sometimes desirable. It does nothing about the missing durable configuration, so each new chat starts with less guidance rather than more.
CCorrect
Durable guidance that applies to every conversation belongs in the Project's configuration: instructions carry the tone and structure rules, and the reference document belongs in knowledge so it is available without being re-attached.
D
One long conversation does keep the guidance in view for a while, and continuity can help an editor compare funder reports. It ties the standard to a single chat that only one person is in, so colleagues working elsewhere get nothing.
The principle
Guidance that should apply to every conversation is configuration, not conversation. Put it in the Project once so it reaches everyone automatically rather than depending on someone remembering to paste it.
100% of 59 answers on this site got it right on the first try.
An associate is setting up a Project for weekly customer-support summaries. Which item belongs in the Project's knowledge rather than in its instructions?
- AThe requirement that each customer-support summary stays under 300 words.
- BThe current tier definitions that every new draft must correctly apply.
- CThe direction to write summaries in plain language for non-specialists.
- DThe rule that every weekly summary closes with a list of open escalations.
Show the answer and why each option wins or loses
A
This is a real constraint worth stating, and length limits are among the most useful things to set once. It is a behavioural rule rather than reference material, so it belongs in instructions.
BCorrect
Tier definitions are reference material the work must draw facts from, and they change on their own schedule. That is exactly what knowledge sources are for.
C
Audience and register are worth fixing so every summary reads consistently. Like other behavioural rules, it tells Claude how to write rather than supplying facts, so it belongs in instructions.
D
A required closing section is a sensible standard for a recurring deliverable. It describes output structure, which is instruction territory, not a document to consult.
The principle
Instructions say how to work; knowledge supplies what to work from. Sorting configuration by that question keeps instructions short and keeps reference material updatable on its own.
76% of 49 answers on this site got it right on the first try.
An operations lead at Northbrook Clinic Network sets up a Project for the quarterly compliance newsletter and connects both Google Drive and Gmail. The newsletter draws only on one shared Drive folder of approved policy notices, which a compliance officer maintains. Which configuration best fits this work?
- AConnect both Google Drive and Gmail, and rely on carefully worded prompts to keep the newsletter inside the policy folder.
- BKeep Gmail connected so colleagues' emailed corrections to the notices can be picked up as well.
- CLeave both connectors in place but also upload copies of the approved notices as Project knowledge for redundancy.
- DScope Drive access to the approved-notices folder and remove the Gmail connection from the Project.
Show the answer and why each option wins or loses
A
Prompt-level scoping does steer where Claude looks and is useful for one-off requests. It leaves broader access configured than the work requires, so a vague prompt or a new colleague can pull material from outside the approved set.
B
Email corrections are genuinely useful signal, and connecting Gmail would surface them. It also routes unapproved, unreviewed text into a compliance newsletter, which is precisely the material the approval process exists to filter out.
C
Uploaded copies do guarantee the notices are present even if the connector has trouble. They create a second copy that will drift out of step with the officer's folder, which is worse than having only the maintained source.
DCorrect
Access should match what the work actually needs. Scoping to the maintained folder keeps the newsletter on approved material and removes a connector that only adds unreviewed content.
The principle
Configure connector access to the narrowest source the work genuinely needs. Extra reach is not free capability; it is extra material that can end up in an output nobody intended.
72% of 54 answers on this site got it right on the first try.
A hiring team asks an associate to use Claude to screen applications and auto-reject candidates who fall below a scored threshold, with no reviewer ever seeing the rejected files. What should the associate propose?
- AHave Claude summarize each application against the posted criteria and send every file to a recruiter, who decides rejections
- BDecline the request and explain that evaluating job applicants with AI assistance is not a permissible use of the tool
- CRun the screen as requested, but keep a log of every rejected application and the score behind it so the result can be audited
- DRun the screen as requested and have Claude also draft the rejection notices so the wording candidates receive stays consistent
Show the answer and why each option wins or loses
ACorrect
Correct. It keeps the speed benefit of summarization while leaving the consequential decision with a human who sees every file. Rejection is the step whose cost lands on a candidate who never agreed to it, so that step stays with a person.
B
This does protect candidates from an unreviewed automated rejection, but it overcorrects. Summarizing and organizing applications for a human decision-maker is a normal, defensible use; refusing all AI assistance gives up real value the team is entitled to.
C
An audit log of the rejections and their scores makes the outcome reviewable after the fact, which is genuinely useful for oversight. It does not help the rejected candidate, though: the decision was still made and acted on without any human ever reading the file.
D
Consistent, respectful rejection wording is a real benefit and drafting notices is fine in itself. It does nothing about the underlying problem, which is that no human evaluated the applications before they were rejected.
The principle
Fix the part of the workflow where an unreviewed AI output determines someone's outcome, rather than banning the tool or documenting the harm afterwards.
68% of 60 answers on this site got it right on the first try.
A support associate at a health insurer receives a member asking whether one specific medication is covered under their plan. Which use of Claude is appropriate here?
- AAsk Claude to state whether the medication is covered under this member's plan so the answer goes out immediately
- BAsk Claude to draft a plain-language explanation of how coverage works, then verify this member's status in the plan system
- CAsk Claude to estimate how likely coverage is from how similar medications are usually treated, and send that to the member
- DAvoid using Claude for any member-facing question and answer this one by hand from the plan's own policy manual
Show the answer and why each option wins or loses
A
This is fast and directly answers what was asked. The model has no access to this member's actual plan record, so a confident coverage statement could be wrong in a way the member will act on financially.
BCorrect
Correct. It uses Claude for what it is reliable at, clear explanation of a general process, and takes the member-specific fact from the authoritative system of record before the reply goes out.
C
A likelihood estimate feels more careful than a flat yes, and it is useful framing internally. Sent to a member it still reads as guidance about their own coverage, and it is still not grounded in their record.
D
Working only from the manual guarantees accuracy of the source, but it discards drafting help that carries no risk at all. Explaining a general process with AI assistance is squarely within appropriate use.
The principle
Split the task: use AI for the general, checkable part and pull person-specific facts from the system that actually holds them.
49% of 61 answers on this site got it right on the first try.
A regional utility is piloting Claude across several teams. Which two of the proposed uses are appropriate as described?
Choose 2
- ADeciding which customers with unpaid balances have service disconnected, with the resulting list sent straight to dispatch
- BProducing the final safety-inspection sign-off for substations in place of a certified field inspector
- CDrafting outage-notification templates that a communications supervisor reviews before they are sent to customers
- DAnswering employee questions about the contents of their own individual disciplinary records held by HR
- ESummarizing recurring themes across a year of anonymized customer complaint text for a planning report
Show the answer and why each option wins or loses
A
Prioritizing accounts is a real analytical task, and the output would be usable as a shortlist. Sending it straight to dispatch removes the human judgment from an action that cuts off someone's power, which is exactly where review belongs.
B
The model could produce a well-structured inspection document quickly. A sign-off is an accountable professional attestation, not a document, and it cannot be transferred to a tool that did not inspect anything.
CCorrect
Correct. Template drafting is low-risk, reversible, and a named human approves the wording before any customer sees it.
D
This would give employees faster answers about their own information, which sounds helpful and consent-clean. Disciplinary records held by HR are sensitive personnel data, and routing them through a general drafting workflow exposes them without any need to.
ECorrect
Correct. Theme-finding across anonymized free text plays to the model's strengths, and the output informs planning rather than deciding anything about a specific person.
The principle
Appropriate uses tend to be reversible, human-reviewed, and aimed at aggregate insight; inappropriate ones place an irreversible outcome for an individual on an unreviewed output.
At Harbourline Logistics, an analyst asks Claude to summarize the quarterly carrier performance file and gets back something accurate but too general for the operations review. Two colleagues submit the same request and get the same flat result. Which cause should the analyst address first?
- AThe file is long, so a trimmed excerpt should be pasted in place of the whole document.
- BThe model tier is too light for analytical work, so the request should be rerun on a stronger one.
- CThe request named no audience and no decision, so nothing signalled which details mattered.
- DThe conversation includes no sample output, so an example should be attached before anything else.
Show the answer and why each option wins or loses
A
Trimming the input can help when a document genuinely exceeds what fits, but nothing here points to that: the output is accurate, so the material was read. Shrinking the input removes detail without telling Claude which detail to keep.
B
A stronger tier does help on genuinely hard reasoning, and it is a fast thing to try. It does not help here, because every tier will produce a general answer to a general request.
CCorrect
Correct. Three people wrote effectively the same vague ask and got the same vague result, which points at the request, not the material or the model. Naming who reads it and what they must decide tells Claude what to include and what to drop.
D
An example is a strong way to pin down shape and depth, and would likely improve the next draft. It is a second move, though: without knowing the audience, the example only teaches formatting, not relevance.
The principle
When the same weak result reproduces across people, the request is the variable to inspect, not the material or the model. Vague output most often traces to an unstated audience, decision, or success criterion.
66% of 58 answers on this site got it right on the first try.
An associate notices that Claude's outputs drift away from a required format after many turns in a long working conversation. What is the most reliable first fix?
- ARestate the format requirements in the current message, with a short example of the expected shape.
- BStart a fresh conversation so that no earlier turn can pull the output away from the required format.
- CAsk Claude to explain why it stopped following the format, then request another draft once that is clear.
- DSplit the work into several smaller messages so each request carries less for Claude to hold at once.
Show the answer and why each option wins or loses
ACorrect
Correct. Instructions given many turns ago compete with everything said since. Repeating the requirement where the work is happening, plus a concrete example of the shape, puts it back in front of the request being answered.
B
A fresh conversation does clear accumulated drift and sometimes is the right call. It also discards the context built up over the session, which is a heavy price for a problem a restated instruction usually solves.
C
Asking for an explanation produces a plausible-sounding account, and can occasionally surface a real conflict in the instructions. It costs a turn and does not itself correct the next output.
D
Smaller messages are good practice for complex work and reduce what any one request must juggle. Here they address workload rather than the actual issue, which is that the format requirement has fallen out of focus.
The principle
Instructions lose force as a conversation lengthens. Restating the requirement at the point of use, ideally with an example, is cheaper and more targeted than resetting the whole session.
53% of 47 answers on this site got it right on the first try.
Right-rates are first attempts on this site's free CCAO-F form, as of September 15, 2026.
Take all 60 free CCAO-F questionsWhere to go next
- The full free form. All 60 questions of Form 1, in study mode or timed exam mode: CCAO-F Form 1.
- What the exam covers. The domains and their weights: CCAO-F domains.
- Cost, length and passing score. CCAO-F exam facts.