Prompting for Written Work

Generic AI Writing Is a Briefing Problem, Not a Model Problem.

When people complain that AI-generated writing sounds generic, they almost always blame the model. The real culprit is the brief. A model producing generic output has been given a generic brief. Feed it a specific, well-constructed brief and the output will be specific, well-suited, and often close to publish-ready.

Written work — emails, reports, proposals — is where intermediate prompting technique pays off most directly. These are tasks with clear audience, purpose, and format requirements. They benefit most from explicit briefing. And they are the tasks most people do most often, which means getting this right has compounding value across your entire workweek.

The Five Elements of a Writing Brief

Every AI writing brief benefits from five elements. You do not need all five every time — but knowing them ensures you include the ones that matter for each task:

• Audience: Who is reading this? Their role, their relationship to you, what they already know, what they care about.
• Purpose: What is this piece trying to do? Inform, persuade, request, acknowledge, reassure, move to action?
• Constraints: Length, format, tone, what to include or avoid, any company voice requirements.
• Context: What happened before this? What does the reader already know? What is the relationship history?
• Output spec: What does good look like? A draft ready to send, three options, a rough structure?

Emails: The CARD Framework

For email specifically, four elements are most commonly missing from weak briefs: Context, Ask, Relationship, and Detail level. Providing all four transforms email prompting from hit-or-miss into reliable.

Weak email brief: "Write a follow-up email after a job interview."

Strong email brief: "Write a follow-up thank-you email after a second-round interview for a Senior Product Manager role at a fintech startup. I met with the VP of Product and two engineers. The conversation was positive — they emphasized culture fit and asked about my experience with Agile in regulated industries. Tone: warm, professional, genuine. Length: under 150 words. No hollow phrases like 'It was a pleasure meeting with you' — be specific about something from the conversation."

The second brief specifies context (what kind of interview, what round), ask (a thank-you, not a status check), relationship (first impression, still building it), and detail level (specific references, not generic warmth). The output will be noticeably different — and better.

Reports: Structure Before Draft

For longer documents like reports, resist the urge to ask for the full draft in one prompt. Generate the structure first, review it, adjust it to match your actual needs, and then ask for sections to be drafted. This two-step approach catches structural problems before they become wasted paragraphs.

Step 1: "I need to write a monthly marketing performance report for the executive team at a B2B software company. The key metrics are website traffic, lead volume, pipeline contribution, and content performance. The audience is data-literate but time-constrained — they skim. Propose a structure for the report including section order and 1-sentence description of what each section covers."

Step 2: "Good. Now draft the executive summary section. Key numbers this month: website traffic +12% MoM, leads down 8% due to seasonal adjustment, pipeline contribution on track at $2.1M. Tone: confident but not defensive about the lead dip. Length: 100–120 words."

Proposals: The Reader-First Model

Business proposals fail most often because they lead with what the writer wants to say rather than what the reader needs to know. When briefing AI on a proposal, explicitly instruct it to structure from the reader's perspective — what problem the reader has, why it matters to them, and what they need to feel confident deciding yes.

Weak proposal brief: "Write a proposal for a website redesign project."

Strong proposal brief: "Write a project proposal for a website redesign for a 20-person accounting firm whose current site is not generating any inbound leads. The decision-maker is the managing partner — non-technical, focused on ROI. Structure it from the reader's perspective: start with the business problem they have, not our capabilities. The project is 6 weeks, $15,000. End with a clear decision prompt, not a soft close."

What Not to Leave Vague

Three things consistently go wrong when left unspecified in writing briefs:

• Tone: "Professional" means very different things. Specify the register — formal, conversational, warm but direct, authoritative, collaborative.
• Length: Without a length constraint, models default to whatever seems thorough. Specify a word count or a constraint like "could be read in 60 seconds."
• Clichés to avoid: AI writing defaults to certain stock phrases. Name the specific ones you do not want: "I hope this email finds you well," "As per my previous email," "Please do not hesitate to contact me."

Editing the Output

Even a well-briefed piece usually needs one editing pass for voice. The AI will produce something structurally sound and tonally appropriate — but it will not sound exactly like you. A five-minute editing pass to adjust sentence rhythm, add specific details only you know, and remove anything that rings false is normal and expected. The goal is a strong draft that you refine, not a final output that requires zero touch. Setting that expectation correctly makes the entire workflow feel productive rather than frustrating.