Curriculum › Prompting and Task Execution · 14% of the exam

Writing prompts for business tasks

What you’ll be able to do

  • Turn a vague business request into a prompt with a clear goal, audience, context and finish line
  • Choose a prompting approach that fits the task type: analysis, research, drafting or brainstorming
  • Spot the missing ingredient in a weak prompt before you send it

What you need to know

Brief Claude like a capable new colleague

Claude starts each conversation knowing a great deal about the world and nothing about your organization, unless you have set up a Project, connected a source or enabled memory. A useful mental model: brief it the way you would brief a smart new hire on their first morning. They can write well and reason well, but they don't know your customers, your house style, last quarter's decision or who will read the result.

Most effective business prompts cover the same handful of ingredients:

  • Goal — what the output is for and what decision or action it supports ("so the leadership team can choose between two vendors").
  • Audience — who will read it and what they already know (a board, new hires, frustrated customers, a technical team).
  • Context and source material — the facts, documents, notes or data Claude should work from. Paste or upload them rather than describing them.
  • Constraints — tone, length, things to avoid, policies to respect, deadlines or budget limits.
  • Output format — a one-page memo, a table with named columns, five bullet points, an email under 150 words.

Not every prompt needs all five. A quick rewording request needs one line. But when output disappoints, the cause is usually one of these ingredients missing, not a lack of words.

Before
Write an email about the office move.
After
Draft an email to all 80 staff announcing our office move on the date in the attached plan. Readers care most about parking and desk booking. Friendly, plain tone, under 200 words, with a short "what you need to do" list at the end.

Specific beats long

A precise sentence ("flag any clause that lets the vendor raise prices mid-contract") does more than a paragraph of general encouragement ("be thorough and careful and make sure you catch everything important"). Long prompts full of repeated emphasis or vague adjectives give Claude more to interpret without telling it anything new. When you are tempted to add more words, ask instead: what exactly would a great answer contain that the last one didn't?

Examples help too. If you want a particular style, paste one or two short samples of what good looks like, such as a past newsletter item or a reply your team liked, and say what to copy from them (tone and structure, not the content).

Adapt the approach to the task type

The blueprint names four task types, and each benefits from a slightly different prompt:

  • Analysis — supply the data or document, name the question, and say what to compare or look for. Ask Claude to separate what the data shows from its interpretation, and to say when the data can't answer the question. "Here are three months of survey results. Which themes appear most often in negative responses, and roughly how often? Quote two examples of each."
  • Research — define the question, the scope (time period, region, types of sources) and what you need back. For multi-source questions, the research feature or web search returns cited findings; either way, plan to check key claims against the sources. Ask for uncertainty to be stated rather than smoothed over.
  • Drafting — give audience, purpose, tone, length and the facts to include. Claude writes best from real inputs (your notes, the policy text, the meeting summary) rather than inventing specifics. Expect to edit: a draft is a starting point you own.
  • Brainstorming — ask for quantity and variety first, and relax constraints so ideas can range widely. "Give me 15 names for the onboarding program, from playful to formal." Judge and filter afterwards, possibly in a separate step. Brainstormed ideas are raw material, not recommendations.

Business and technical tasks follow the same rules

The objective says "business and technical tasks." For an associate, technical tasks mean things like writing a spreadsheet formula, cleaning up a messy list, explaining an error message or drafting a requirements summary for the engineering team. The same ingredients apply: say what you have (paste the actual data or error), what you want, and the format. If the request turns into building software, integrating systems or designing an automated agent, that is the point to involve a Claude developer or architect rather than stretching a chat prompt.

Key concept

A good prompt closes the gap between what you know and what Claude knows: goal, audience, source material, constraints and format, adapted to whether you are analyzing, researching, drafting or brainstorming.

Exam questions describe a disappointing result and ask for the most appropriate fix. The right answer usually adds the specific missing ingredient (the audience, the data, the format) rather than adding emphasis, switching tools or giving up.

Practice scenario

ScenarioA marketing lead asks Claude, "Give me ideas for our spring campaign." The answer is a generic list that could apply to any company. She plans to rewrite the prompt as "Give me really creative, original, high-impact ideas for our spring campaign, and think hard about it."
Work it through, then open this

Her rewrite adds intensity but no information. Claude still doesn’t know the product, the audience, the budget, the channels or what the campaign must achieve, so the ideas will stay generic. A better prompt names the product line, the target customer, the goal (for example, sign-ups for a spring promotion), the channels and budget range, and asks for a varied set of ideas, say 12 ranging from low-cost to ambitious. Because this is brainstorming, she should welcome range and then shortlist in a follow-up step, rather than asking for the single “best” idea up front.

Hands-on exercise — Rebuild a weak prompt from its ingredients

Beginner · 15 min

What you’ll learn

  • Which prompt ingredients change the output most
  • How the same topic needs different prompts for different task types
  • Why supplying source material beats describing it
  1. In a new chat, send a one-line request you’d genuinely make at work, such as “Summarize our leave policy.” Then send a second version that adds the goal, audience, the actual policy text (pasted or uploaded) and a format.

    • Why: Seeing the two answers side by side shows how much of output quality comes from what you supply.
    • You should see: The first answer is generic or invented; the second is grounded in your text and shaped for the reader.
  2. Take one topic, such as “reducing meeting overload,” and write four prompts: one each for analysis (paste calendar data or survey results), research, drafting (an email to staff) and brainstorming (a list of options).

    • Why: It makes the differences between task types concrete.
    • You should see: Four noticeably different outputs, each more useful because the prompt matched the task.
  3. Take a prompt you’ve used before and remove every sentence that adds emphasis without information (“be very thorough,” “this is important”). Replace them with one specific instruction about what a good answer contains.

    • Why: It trains the habit of specific over long.
    • You should see: A shorter prompt that gets an equal or better answer.

Exam traps

Assuming Claude knows your company, audience or history without being told

Unless a Project, connector or memory supplies it, Claude only knows what is in the conversation. Missing context produces generic output.

Making a prompt longer instead of making it more specific

Repeated emphasis and vague adjectives add words, not direction. Add the precise detail that’s missing instead.

Using one prompting style for every kind of task

Analysis needs data and a question, research needs scope and sources, drafting needs audience and tone, brainstorming needs room for range. One template fits none of them well.

Asking for brainstormed ideas and judging them as if they were vetted recommendations

Brainstorming produces raw options. Evaluate, check feasibility and shortlist as a separate step before anyone acts on them.

Leaving the output format and length for Claude to guess

If you need a table, a 150-word email or three bullets for a slide, say so. Otherwise you’ll spend a follow-up turn reshaping it.

Asking for analysis without supplying the data to analyze

Without the actual figures or document, Claude can only speak generally, or worse, produce plausible-looking numbers that aren’t yours.

Sources

Quick check

A customer success manager wants Claude to draft replies to complaints in the company's voice. Her prompt describes the voice as "warm but professional, not too casual, empathetic yet efficient." The drafts are fine but don't sound like her team. What is the most effective addition to the prompt?