Prompting Claude for Drafting and Brainstorming

1.Drafting: give Claude your voice and your facts

Drafting is the task type where Claude most needs information it has no way of knowing. In Anthropic's AI Fluency course for nonprofits, a director writing an emergency grant proposal improves the draft in two ways. First, they upload past successful work so Claude can match the organisation's voice. Second, they add details only they have: track record, partnerships, staff expertise and actual community relationships. The course's writing exercise lists what a drafting prompt should state: what the funder cares about, the programme's key details, the organisation's voice and values, specific requirements such as word count, required elements and tone, and what makes the approach distinctive.

Format belongs in the prompt too. Anthropic's prompt engineering article advises telling Claude what to do rather than what not to do, for example asking for flowing prose paragraphs instead of saying "no markdown". When a tone is easier to show than describe, include one example and add more only if needed.

Sometimes the approach matters as well as the product. For a personalised fundraising email, guided chain of thought tells Claude what to work out before it writes:

Think before you write the email. First, think through what messaging might appeal to this donor given their donation history. Then, consider which aspects of the Care for Kids program would resonate with them. Finally, write the personalized donor email using your analysis.

When a draft makes factual claims, as a press release or grant proposal does, put the check into the prompt. Anthropic's example limits Claude to the supplied documents and then has it review every claim against them:

Draft a press release for our new cybersecurity product, AcmeSecurity Pro, using only information from these product briefs and market reports.
<documents>
{{DOCUMENTS}}
</documents>

After drafting, review each claim in your press release. For each claim, find a direct quote from the documents that supports it. If you can't find a supporting quote for a claim, remove that claim from the press release and mark where it was removed with empty [] brackets.

Revision feedback works the same way. The course advises being specific: instead of "make this more compelling," say exactly what to add, for example specific details about your community relationships.

Sources123

2.Brainstorming: ask for range, then judge separately

Anthropic's documentation says less about brainstorming than about the other three task types, so this section sticks to what it does support. The general rules still apply. A brainstorm about cheap marketing needs the budget, the audience and what has already been tried, because the prompt should carry as much context as possible. One Anthropic field guide to Claude Fable 5 says the most important step is giving Claude context about your starting point, then working with it as a thought partner.

The shape of the request is what changes. Brainstorming wants options to react to, not a single answer. The same guide describes asking for several design approaches and gives an example prompt: "Make me an HTML page with 4 wildly different design directions so I can react to them." Asking for clearly different options brings out what a single answer would hide. The author notes that Claude often finds high-value approaches they would have missed, and that brainstorming helps avoid setting a scope that is too narrow or too wide.

Generating ideas and judging them are different jobs. If one prompt asks for both, the judging can cut the list short before it has any range. Anthropic's support guidance says to break complex requests into substeps. Prompt chaining applies that across turns: each prompt handles one stage and its output feeds the next. For brainstorming, that means one prompt to generate options under your constraints and a separate prompt to evaluate each one for feasibility before you choose.

Sources452

3.When one request spans several task types

Real work rarely fits into one category. A business proposal on an emerging technology needs research, then analysis of what was found, then drafting. Spot which task type each part belongs to and give each part its own shape. The research step needs scope, context and the right capability. The analysis step needs named dimensions and grounding. The drafting step needs voice, facts only you know and concrete requirements.

Anthropic's support guidance describes this combination for research-heavy work. Extended thinking together with Research lets Claude plan its approach and then gather information thoroughly. The guide's examples include researching emerging technologies for a business proposal and analysing several scientific papers. When a request also includes a brainstorm, keep generating and evaluating as separate steps, as above.

Sources6

Exam traps

Each one states something that sounds right. Open it to see what is actually true.

1. 1.When a draft falls flat, the best follow-up is a general push such as "make this more compelling."Why is that wrong?

   Specific revisions work better. Tell Claude exactly what to add or change, especially details only you know.

   Covered in Drafting: give Claude your voice and your facts

2. 2.A good brainstorming prompt asks Claude for its single best idea and a final verdict in one step.Why is that wrong?

   Ask for several clearly different options, then evaluate them as a separate step. Breaking a complex request into substeps keeps generating and judging apart.

   Covered in Brainstorming: ask for range, then judge separately