Prompting Claude for Drafting and Brainstorming Tasks

1.Drafting: give Claude your voice and the facts it can't know

When Claude writes something for you, such as a proposal, an email or a press release, the usual failure isn't bad grammar. It's generic content. Claude can only describe your organisation in general terms unless you give it the specifics. Anthropic Academy's writing walkthrough follows James, who runs a small environmental justice organisation and must write an emergency grant proposal in 48 hours. He gets usable drafts by uploading past successful work so Claude can pick up his organisation's voice, and by adding the details Claude has no way of knowing: his track record, partnerships, staff expertise and actual community relationships.

The exercise that follows the walkthrough breaks a drafting prompt into ingredients:

Tone and style are hard to describe and easy to show. Anthropic's prompt-engineering blog recommends a single example when you need a particular tone or style, adding more only if the output still misses. It also advises framing formatting as what to do rather than what to avoid. 'Write in smoothly flowing prose paragraphs' works better than 'don't use markdown.'

Revision is where drafting prompts most often go wrong. Vague feedback such as 'make it better' or 'more compelling' gives Claude nothing concrete to change. The walkthrough's advice is to name the specific change instead: add these details about our community relationships, or fix this deficit-based framing of beneficiaries. When you review a draft, check every data point, ask whether the language sounds like your organisation or like anyone's, and note what's missing that only you know. Each of those gaps becomes the next, more specific instruction.

Sources12

2.Keeping a draft within your source documents

Some drafts must not say anything the source material doesn't support: a press release built from product briefs, or a summary for a regulator. Two instructions from Anthropic's hallucination guide handle this. The first is external knowledge restriction: tell Claude to use only the documents you provide, not its general knowledge. The second is citation checking after drafting: have Claude find a supporting quote for each claim and remove any claim it can't support. The guide's press-release example combines both:

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.

The empty brackets are worth noticing. They don't silently drop the unsupported claim, they mark where it was. You can then decide whether to find evidence for it or leave it out. This is the drafting counterpart of the 'I don't know' permission used in analysis: the prompt gives Claude a visible, acceptable way to say the sources don't support something.

Sources3

3.Brainstorming: ask for breadth, state the limits, judge separately

The sources behind this lesson have no dedicated brainstorming guide, so this section applies Anthropic's general prompting principles to idea generation. It doesn't cite any brainstorming-specific technique. Three of those principles carry most of the weight.

First, ask for breadth explicitly. Anthropic notes that if you want 'above and beyond' output, you should request it directly rather than expect Claude to infer it from a vague prompt. Its own example turns a bare request into one that asks for range and ambition:

Create an analytics dashboard. Include as many relevant features and interactions as possible. Go beyond the basics to create a fully-featured implementation.

Second, state the constraints and the reason for them. Breadth without limits produces ideas you can't use. The blog's specificity checklist asks for clear constraints, relevant context about the audience and goal, and any requirements or restrictions such as budget limits. If the real situation is a small budget, a particular audience or a deadline, put it in the prompt. Explain why it matters, too, because Claude generalises from the reason as well as the rule.

Third, keep generating separate from judging. When the order of steps matters, Anthropic advises writing them as numbered steps: for example, 1) list options, 2) assess each one's feasibility against the constraints, 3) recommend. When each stage deserves full attention, prompt chaining goes further and gives each stage its own prompt, feeding one stage's output into the next. The blog notes that chaining trades speed for accuracy by making each step simpler.

Sources42

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' or 'improve the tone.'Why is that wrong?

   Revision feedback works when it names the concrete change and supplies the missing details. General pushes give Claude nothing specific to change.

   Covered in Drafting: give Claude your voice and the facts it can't know

2. 2.Asking Claude to 'brainstorm ideas' is enough for it to return a broad, ambitious set, because that's what brainstorming means.Why is that wrong?

   If you want expansive output, say so explicitly, and state the constraints the ideas must meet. Otherwise Claude has to guess the scope from a vague request.

   Covered in Brainstorming: ask for breadth, state the limits, judge separately