Common Tasks: Writing, Analysis, Summaries and Data

Almost every workplace request to Claude falls into one of four families: writing something, analysing something, condensing something, or working with data. Each has its own prompt shape and its own failure mode. Knowing the pattern for each family lets you move fast without sacrificing quality.

Writing and drafting

Drafting covers emails, proposals, policies, announcements, job posts, speaking notes and social copy. The pattern is the five blocks from the first lesson, with particular weight on audience and tone. Three habits raise quality:

• Give Claude the key messages as bullets and let it handle prose; you keep control of substance while delegating sentence construction.
• Ask for a structure first ("outline the proposal in six headings") and approve it before asking for the full draft. This prevents a long rewrite later.
• Request alternatives for the parts that matter most, such as the opening line or the call to action.

Draft an internal announcement that we are moving to a four-day
on-site week from January. Key messages: decided after the staff
survey; Fridays remain remote; team leads will share rotas by
15 December; questions go to HR. Tone: warm, plain, no corporate
cliché. Under 200 words. Suggest two subject lines.

Analysis and reasoning

Analysis tasks ask Claude to compare, evaluate, find gaps or recommend: reviewing a contract clause, weighing two vendor proposals, stress-testing a plan. Attach the actual material; Claude cannot analyse a document it has not seen. Then specify the lens: "from a procurement perspective", "as a sceptical investor", "against our stated values".

Ask for the reasoning, not just the verdict. "Explain which option is better and why, then list the assumptions your answer depends on" produces an answer you can check and defend. For anything consequential, ask for the opposing case as well: "Now argue for the other option as strongly as you can."

Analysis is where Claude's confident tone is most dangerous, so the exam expects you to treat conclusions as input to your judgement rather than a replacement for it.

Summaries and condensing

Summarisation includes meeting notes, long email threads, reports, research papers and transcripts. The common mistake is asking for "a summary" with no indication of purpose. A summary for someone who missed the meeting differs from a summary of decisions for a project log, which differs again from a summary of risks for a steering committee. State the purpose, the reader and the length.

Useful shapes include:

• Executive summary: three to five sentences, decisions and implications only.
• Action list: who does what by when, extracted verbatim where possible.
• Structured digest: fixed headings such as Decisions, Open questions, Risks, Next steps.
• Layered summary: one sentence, then one paragraph, then detailed notes.

Summarise the attached transcript for people who missed the call.
Headings: Decisions, Open questions, Actions (owner and date).
Quote exact wording for any commitment about money or dates.
Flag anything that was ambiguous rather than guessing.

Instructing Claude to flag ambiguity instead of resolving it is important; summaries that quietly smooth over disagreements are a known risk.

Data and spreadsheets

Claude can read CSV and spreadsheet files, describe what is in them, calculate totals and trends, build charts, and produce new spreadsheets or formatted documents. In the Claude apps this is powered by a sandboxed analysis and file-creation capability (sometimes shown as a toggle in settings, and controlled by admins on business plans). When it is enabled, Claude writes and runs small pieces of code to do the arithmetic rather than estimating it in prose, which is far more reliable for anything beyond trivial sums.

Good practice for data tasks:

1. Describe the file: what each column means, any known quirks, which rows to ignore.
2. State the question precisely: "average deal size by region for closed-won deals in Q3", not "insights".
3. Ask Claude to show its working: which columns it used, how it filtered, any rows it excluded.
4. Ask for the output shape: a table in the chat, a downloadable spreadsheet, a chart.
5. Spot-check at least one figure against the source yourself (covered in detail in Domain 2).

Claude can also tidy messy data (splitting names, standardising dates), draft formulas for your own spreadsheet, and explain what an existing formula does. For large or sensitive datasets, check your organisation's data policy before uploading (Domain 6).

Matching the task to the prompt

Family	Must attach or state	Ask for	Watch for
Writing	Key messages, audience, tone example	Outline first, then draft, then alternatives	Generic tone, invented facts
Analysis	The actual document or data, the lens	Reasoning, assumptions, opposing case	Overconfident verdicts
Summary	Purpose, reader, length	Fixed structure, flagged ambiguity	Smoothed-over disagreement
Data	Column meanings, exact question	Working shown, specified output shape	Unverified figures

Exam focus

• Writing tasks improve most from clear key messages, a stated audience and an approved outline before the full draft.
• Analysis requires the source material to be attached and a stated lens; always ask for reasoning and assumptions, not just a verdict.
• A summary without a stated purpose, reader and length will be generic; ask Claude to flag ambiguity rather than resolve it.
• For data work, Claude performs calculations by running code when analysis is enabled; still spot-check figures against the source.
• Each task family has a characteristic failure mode (generic tone, overconfidence, smoothing, unverified numbers) that the exam expects you to anticipate.