Role Prompting and Prompt Length: What Actually Helps
Does telling AI to act as an expert help? How long should a prompt be, and how many questions can you ask at once? A practical guide to what works.
Knowing that a good prompt needs context, a clear task and a defined output is one thing. Sitting in front of a blank box with a real job to do is another.
This guide gives you prompts you can adapt for everyday business tasks. Each one has placeholders in [square brackets] for you to replace, followed by a short note on why it works. If you want the thinking behind them first, read our guide to writing effective AI prompts.
A few ground rules before you start:
- Replace every placeholder. A prompt with the brackets left in will give a generic answer.
- Do not share sensitive information unless your company policy and the tool’s data settings allow it.
- Check the result. Treat each answer as a draft that you review before it reaches a customer or goes live.
The Basic Shape of a Business Prompt
Almost every prompt below follows the same pattern:
Task + Background + Audience + Format + Limits
For example: “Write [task] for [audience]. Here is the background: [background]. Format it as [format]. Keep it [limits].”
Once you recognise this, you can write your own prompts for jobs not covered here.
Write a first draft of a website page for [service] aimed at [audience]. Our main points are: [three to five real points]. Use plain language, short paragraphs and one clear call to action to [action]. Do not include any statistics, awards or client names.
Customer Communication
Reply to a customer enquiry
“You are helping me reply to a customer. Here is their message: [paste message, with personal details removed]. Here is the situation on our side: [what has happened and what you can offer]. Write a polite, clear reply of no more than 120 words. Do not promise a delivery date or outcome that I have not given you. End with one clear next step.”
Why it works: it gives the AI the facts, sets a length and, importantly, tells it what not to promise.
Handle a complaint
“A customer is unhappy about [issue]. We think [what went wrong, in plain terms]. We can offer [options you are genuinely able to offer]. Write a calm reply that acknowledges the problem, explains what we will do and avoids defensive language. Keep it under 150 words.”
Why it works: the AI cannot guess what you are willing to offer, so you state it. This keeps the reply honest and within your policy.
Marketing and Content
Draft a service page
“Write a first draft of a website page for [service] aimed at [audience, such as owner-managed businesses in a specific area]. Our main points are: [three to five real points]. Use plain language, short paragraphs and one clear call to action to [action]. Do not include any statistics, awards or client names. I will add those myself.”
Why it works: the last instruction prevents invented proof, which is a common problem in AI-written marketing copy.
Turn one idea into several formats
“Here is a blog post: [paste text]. Create: 1) three LinkedIn posts, each with a different angle, 2) a 60-word summary for a newsletter, 3) five possible email subject lines. Match the tone of the original. Keep each LinkedIn post under 120 words.”
Why it works: it uses a source you wrote, so the AI is reshaping your material rather than inventing new claims.
Improve a page you already have
“Here is the text from our [page]. Our audience is [audience] and the goal of the page is [goal]. List the three biggest weaknesses in the wording, then suggest a revised opening paragraph. Do not change the facts.”
Why it works: asking for weaknesses first makes the AI diagnose before it rewrites.
Plan a month of content
“We are a [type of business] serving [audience]. Our goals this month are [goals]. Suggest eight blog topics. For each, give the question it answers, who it is for and the action we want the reader to take. Do not suggest topics that are only about our company news.”
Why it works: it ties each idea to a reader and an outcome, so you get a plan rather than a list of headlines.
Research and Analysis
Summarise customer feedback
“Here are [number] customer comments: [paste, with names removed]. Group them into themes. For each theme, give a one-line description, how many comments mention it and one representative quote. Then list the three issues most worth fixing first and explain why.”
Why it works: the structure forces the AI to show its grouping, so you can check it against the source rather than trust a summary.
Compare options against your criteria
“I am choosing between [option A] and [option B] for [purpose]. My priorities, in order, are [priority 1], [priority 2] and [priority 3]. Compare them in a table against those priorities only. Say where you are unsure or where I should check the facts myself.”
Why it works: your criteria drive the comparison, and the final sentence invites the AI to flag uncertainty instead of hiding it.
Internal Work
Turn meeting notes into actions
“Here are my notes from a meeting: [paste notes]. Extract the decisions made, the actions agreed, who owns each action and any deadlines mentioned. If an owner or a deadline is missing, mark it as ‘not stated’ rather than guessing.”
Why it works: “not stated” stops the AI from inventing owners and dates.
Draft a proposal outline
“I need to propose [project] to [client or decision-maker]. Their main concern is [concern] and their budget is [if known]. Create an outline with sections for the problem, our approach, what is included, timings and next steps. Under each heading, list the points I should cover. Do not write the full proposal yet.”
Why it works: asking for an outline first lets you correct the structure before any text is written.
Tips for B2B Marketers
If you write for other businesses, add these to your prompts:
- The buyer’s role. “The reader is a finance director” produces a different result from “the reader is a marketing manager”.
- The stage of the decision. Someone comparing suppliers needs different content from someone who has not yet recognised the problem.
- The problem, not just the product. Describe the situation the buyer is trying to solve.
- Approved facts only. Give the AI the proof points you are happy to use, and tell it to use nothing else.
- A specific next step. “Book a call” and “download the checklist” lead to different copy.
How to Make These Prompts Your Own
The examples above are starting points. To get the most out of them:
- Run the prompt once and see what is missing. That tells you what context to add.
- Correct the answer with a follow-up. Our guide to improving AI responses with follow-up prompts shows how.
- Save the version that works. Keep a short document of your best prompts, with placeholders ready.
- Share them with your team. A set of tested prompts is more useful than everyone starting from scratch.
If a prompt keeps producing weak answers, check it against our list of common AI prompting mistakes.
Going Beyond Individual Prompts
Prompts are a good way to save time on tasks you do yourself. When the same task happens many times a day, such as answering common enquiries or qualifying leads, it can make sense to build AI into the process itself. Maximus Mediascape helps businesses explore AI automation, chatbots and connected workflows alongside its wider digital services.
Related guides: What Is an AI Prompt? A Beginner’s Guide, 10 AI Prompting Techniques for Better Responses Every Time, How to Improve an AI Response When the First Answer Misses.
