Prompt engineering sounds technical. It is not. It is just the skill of telling AI what you want clearly enough that it gives you something useful back.
If you can write a client brief, you can write a good prompt. The principles are the same: context, specificity, and a clear description of the desired output.
The framework
Every effective agency prompt has four parts:
- Role. Tell the AI who it is. “You are a senior strategist at a digital marketing agency” produces different output to a generic request.
- Context. Give it the background. The client’s industry, their goals, the constraints, the audience. The more relevant context you provide, the better the output.
- Task. Be specific about what you want. Not “write me a proposal” but “write the executive summary for a proposal to a mid-size law firm looking to increase their organic search visibility by 40% over 12 months.”
- Format. Describe the output format. Word count, tone, structure, headings. “Write a 200-word executive summary with three bullet points highlighting our approach” is far more useful than “write something good.”
Prompts that work for agencies
Here are practical prompts for common agency tasks. Adapt them to your context.
Prospect research: “Research [company name] and their industry. Summarise their business model, recent news, main competitors, and likely marketing challenges. Format as a one-page brief I can read before a discovery call.”
Proposal executive summary: “Write a 250-word executive summary for a proposal from [your agency] to [client]. They need [service]. Their main challenge is [challenge]. Our approach is [approach]. Write in a confident but not aggressive tone. Use UK English.”
Meeting summary: “Here is a transcript from a client meeting. Summarise the key discussion points, decisions made, and action items. Format with clear headings and bullet points. Flag any unresolved questions.”
Content brief: “Create a content brief for a blog post targeting the keyword [keyword]. Include: suggested title, H2 structure, key points to cover, questions to answer, competing content to reference, internal linking opportunities, and target word count.”
Common mistakes
Being too vague. “Write me some social media posts” will give you generic output. “Write three LinkedIn posts for a B2B marketing agency promoting their new SEO audit service, targeting marketing directors at mid-size companies, using a professional but approachable tone” will give you something usable.
Not providing context. AI does not know your client, your agency, or your industry unless you tell it. The more context you provide upfront, the less back-and-forth you need.
Accepting the first output. AI is a starting point, not a finished product. The first response is a draft. Ask it to refine, adjust the tone, add specifics, or restructure. Iteration is part of the process.
Using AI for the wrong tasks. AI is excellent at structure, research, and first drafts. It is poor at original strategy, nuanced creative direction, and anything requiring genuine understanding of a specific client relationship. Knowing when not to use AI is just as important as knowing how to prompt it.
Building a prompt library
The most efficient agencies build a shared prompt library. Every time someone writes a prompt that produces consistently good results, they save it in a shared location with:
- The prompt itself
- What it is for
- Any variables to swap out (client name, industry, etc.)
- Example output
Within a few weeks, your team has a library of tested prompts for every common task. New team members get productive faster. Output quality becomes more consistent. Store these alongside your process documentation in an internal AI knowledge base for maximum impact.
This is part of Delivery Notes, a series on implementing AI inside your agency. Subscribe to the newsletter to get new articles weekly.
