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.
You ask an AI tool for something and the answer comes back. It is not terrible, but it is not what you needed. It is too general, too long, in the wrong tone or it has missed the one point that mattered.
The common reaction is to delete it and start again. That is rarely the quickest fix.
An AI response is a first draft. The skill is knowing how to steer it from there, and that comes down to a few repeatable habits. This guide covers how to tailor an AI response, which follow-up prompts fix which problems, and how to make the good result repeatable.
If your first prompts are still hit and miss, start with our guide to what an AI prompt is and what makes a good one. This article picks up after the first answer arrives.
- GenericMake it specific to your business
- ToneName the tone to use
- LengthCut or expand to a set length
- FocusPoint it back at the real question
- GapsAdd the point it left out
Work Out What Kind of Miss It Is
“This isn’t right” is not enough information for you or for the AI. Before you reply, name the problem. Most poor responses fall into one of these groups:
- Too generic. It could have been written for anyone.
- Wrong tone. Too formal, too casual or too salesy.
- Wrong length. Padded with introductions and summaries, or too thin to use.
- Wrong focus. It answered a different question from the one you meant.
- Missing information. It left out something you assumed it would include.
- Doubtful accuracy. It states facts you cannot confirm.
Each group needs a different fix, which is why a vague “try again” so often produces another vague answer.
Follow-Up Prompts That Fix Specific Problems
The most useful follow-up tells the AI what to keep and what to change. Compare these:
Weak: “That’s not very good. Try again.”
Stronger: “The structure is right. Keep the three headings, but rewrite the opening paragraph so it speaks to a small business owner rather than a marketing manager, and cut the whole thing to 250 words.”
The second version keeps what works and points at what does not. Here are follow-ups you can adapt for the most common misses.
| The problem | A follow-up you can use |
|---|---|
| Too generic | “Make this specific to [your business, audience or situation]. Replace general statements with concrete details from what I have told you.” |
| Wrong tone | “Rewrite this in a [warm / direct / plain] tone, as if written by a person talking to a customer. Avoid sales language.” |
| Too long | “Cut this to [number] words. Remove the introduction and the closing summary and keep only the points that affect a decision.” |
| Too thin | “Expand the second and third points. Add a short example for each and explain why it matters.” |
| Wrong focus | “I was asking about [the actual topic]. Ignore [the part it covered] and answer only that.” |
| Missing information | “You have left out [the point]. Add it, and keep everything else the same.” |
| Unsure of accuracy | “List every factual claim in this response so I can check it. Mark any you are not certain about.” |
You do not need to use these word for word. The pattern matters: say what is wrong, say what you want instead and say what to leave alone.
Ask the AI to Question You First
Many poor responses come from missing information that you could easily have supplied. You can ask the AI to find the gaps before it writes anything:
“Before you write the answer, ask me up to five questions about anything you need to know to do this well. Wait for my replies.”
This works well for tasks where context matters, such as a proposal, a customer reply or a content plan. It turns a guess into a short conversation, and the final answer is usually closer to what you meant.
Be Clear About What a Good Answer Looks Like
If you have ever read an AI answer and thought “it just isn’t us”, the cause is often that the AI had nothing to match against. Two additions help:
- Describe the result. “I need something a busy owner can read in 30 seconds and act on.”
- Show an example. Paste a paragraph you like and say “match this tone and level of detail”.
An example does more than a paragraph of description. If you want to go further into this approach, our guide to AI prompting techniques explains how examples, structure and iteration work together.
Cutting Out Padding and Filler
A common complaint is the answer that opens with a paragraph of throat-clearing, repeats your question and finishes with a summary of what it just said. You can head this off in the prompt:
- State the length: “No more than 150 words.”
- State the format: “Three bullet points, one sentence each.”
- State what to leave out: “No introduction, no closing summary, no caveats unless they change the advice.”
- State the first thing you want to see: “Start with the recommendation.”
These instructions cost a sentence and often save several rounds of editing. They are also easier to apply than trying to fix a bloated answer afterwards.
When to Start a New Chat
Correcting the same conversation works well for small changes. It can go wrong when a chat becomes long, because earlier instructions can get diluted or ignored, and the AI may keep returning to an approach you have already rejected.
Start a new chat when:
- The AI repeats a mistake after you have corrected it twice.
- The conversation has covered several different tasks.
- You have changed what you want so many times that the original request no longer applies.
Do not start from nothing. Paste a short summary: the task, the audience, the format and the two or three corrections that mattered. You keep what you learned and drop the clutter.
Make the Good Result Repeatable
The most useful habit is the least obvious. When a prompt finally produces something good, save it.
Take the version that worked, including the audience, the format and the constraints you added along the way. Replace the parts that change each time with placeholders, such as [customer name] or [product]. You now have a template that gives you a good first answer next time, rather than a fifth one.
This is the practical answer to “how do I stop this happening again”. You are not trying to write a perfect prompt from scratch every time. You are building a small set of prompts that you know work. Our business prompt examples show what these templates can look like.
When the Prompt Is Not the Problem
Better follow-ups cannot fix everything. If a response is still poor after a few clear corrections, consider these:
- The AI lacks the facts. It cannot know your latest prices, your internal policies or recent events unless you provide them or it can search for them.
- The task is too big. Split it into smaller steps and check each one.
- The tool is a poor fit. Some tools handle some tasks better than others.
- The answer sounds right but is wrong. AI can state mistakes with confidence, so check important figures, dates, legal points and claims before you use them.
Our guide to common AI prompting mistakes helps you tell which of these you are dealing with.
A Simple Routine to Follow
When an answer misses, work through these steps:
- Name the type of miss.
- Say what to keep and what to change.
- Add any missing context or an example.
- Set the length, format and what to leave out.
- Check the facts that matter.
- Save the prompt if it works.
It takes a minute longer than typing “try again”, and it usually saves much more than that.
Using AI More Effectively in Your Business
Getting a good answer from a single chat is a useful skill. Many businesses also want AI to handle repeatable work, such as answering enquiries, qualifying leads or routing requests, without someone guiding it each time. Maximus Mediascape helps businesses explore AI, automation and chatbot solutions for exactly that.
Related guides: How to Write Effective AI Prompts for Better Results, Common AI Prompting Mistakes and How to Fix Them, 10 AI Prompting Techniques for Better Responses Every Time.
