Practical Prompting
In Part 1 and Part 2, we covered the basics what a prompt is, why it matters, and the three ideas that underpin good prompting (context, iteration, and specificity). If you haven't read those yet, it's worth a quick look before diving into this article. This article is about how to apply these principles in practice. We're going to take five real-world scenarios and show the difference between a weak prompt and a strong prompt. The goal is to give you a feel for the thinking behind good prompts so you can apply it to whatever you're working on.
Example 1: Marketing strategy for a small business
You run a small independent coffee shop in Brighton, UK. You want some marketing ideas but you've got a limited budget.
Weak prompt:
"Give me some marketing ideas for my coffee shop."
You'll likely get a generic list: "use social media, offer loyalty cards, partner with local businesses." It's not particularly useful. There's no sense of your situation, your audience, or your constraints.
Stronger prompt:
"I run a small independent coffee shop in Brighton, UK, near the university. Most of my customers are students and young professionals. I've got a monthly marketing budget of about £300 and I'm already on Instagram but my engagement is low. I don't have a website yet.
You are an expert marketing consultant who has specific knowledge on marketing strategies for small businesses. Can you suggest five practical marketing strategies I could start within the next month? For each one, explain what to do, roughly what it would cost, and what kind of results I might realistically expect. Keep the language simple and conversational. Use British English."
Why this works
You've told the AI who you are, who your customers are, what you've already tried, what your budget is, and what format you want the answer in. You've also asked the model to take on a specific persona - which helps focus it's attention. The model now has enough to give you something tailored and actionable.
Example 2: Corporate strategy thinking
You are a senior manager at a mid-sized manufacturing company and you want to use AI to help think through a strategic challenge.
Weak prompt:
"I work at a manufacturing company and want to create a growth strategy."
Again, this is too vague. The LLM doesn't know your industry, your size, your market position, or what "grow" means to you.
Stronger prompt:
"You are a professional management consultant with experience in the manufacturing sector.
I'm the operations director at a mid-sized UK manufacturing company. We make components for the automotive industry and employ around 200 people. Revenue has been flat for two years. Our main customers are in Europe, but we're seeing increased competition from lower-cost manufacturers in Asia. We've been thinking about three options: diversifying into a new sector (possibly aerospace), investing in automation to reduce costs, or expanding our sales team to target new customers in our existing market. Can you help me think through the pros and cons of each option? Consider factors like risk, capital required, time to see returns, and how each option fits with current trends in UK manufacturing.
Present this as a clear comparison, not a long essay. Use British English and use bullet points rather than long sentences."
Why this works
You've framed this as a strategic piece of work and constrained the context to your specific environment. Again the LLM has been asked to take on a specific role. You've given the AI three specific options to evaluate rather than asking it to generate ideas from scratch. And you've told it what factors to consider, which stops it from wandering off into irrelevant territory. You've also asked for a comparison format, which will make the output much easier to digest.
Example 3: Contesting an insurance decision
Writing a polite but firm email or letter contesting a decision by an insurer can be a difficult task to get right. Using AI to when you're frustrated can generate something that's less angry and more effective (although your anger may be completely justified and effective).
Weak prompt
"Write a letter to my insurance company about a rejected insurance claim on my burst water pipe."
Stronger prompt
"I need to write a polite but firm letter to my home insurance company contesting their decision to reject a claim. Here are the details: My insurer is Ava Insurance. My policy number is XX-12345678. I submitted a claim on 15 February 2026 for water damage to my kitchen ceiling caused by a burst pipe during the cold snap in January. The claim was rejected on 3 March 2026. Their reason was that they consider it 'gradual damage' rather than a sudden event, but the pipe burst suddenly during freezing temperatures and the damage happened overnight. I have a plumber's report confirming the pipe burst due to freezing conditions, and photographs showing the damage occurred in a single event, not gradually. The letter should be addressed to the complaints department. It should be polite and professional but clear that I disagree with their assessment and want the decision reviewed. Reference the plumber's report as supporting evidence. Ask for a written response within 14 days. State that if the matter isn't resolved, I intend to refer it to the Financial Ombudsman Service. Keep the tone measured and reasonable, not aggressive or threatening. Around 400 words. Use British English and don't use emotional language."
Why this works
You've given the AI lots of information here: the history of the issue, the evidence, the outcome you want, the tone, and even the escalation path. The result will be something that still needs review, but likely this will be more to hone. Without this level of detail, you'd spend more time editing the LLM's output than you saved by using it in the first place. This is the real power of good prompting, it gets you to your final desired output with fewer iterations.
Example 4: Appealing a parking ticket
Whether it's a council penalty charge or a private parking company, a well-written appeal can make a real difference.
Weak prompt
"Help me appeal a parking ticket."
Stronger prompt
"I received a Penalty Charge Notice (PCN) from Westminster City Council on 22 March 2026 for parking in a restricted zone on Marylebone High Street. The PCN number is WM-987654. I believe the ticket was issued unfairly for the following reasons: I parked at 14:10 and the restriction starts at 14:30. I have a message showing I arrived at 14:10 and a photo of the parking sign showing the restricted hours. I also left the area before 14:30, which I can demonstrate from a contactless payment I made at a nearby shop at 14:22. Can you write a formal but polite appeal letter to Westminster Council? It should clearly state the grounds for appeal, reference the evidence I have, and request that the PCN be cancelled. The tone should be respectful and factual, not emotional. Keep it concise, around 300 words. Use British English."
Why this works
By giving the AI the specific timeline, the evidence you hold, and the tone you want, the output will be well structured and factual. The LLM can also help you frame the argument in a way that aligns with how councils typically assess appeals, which is hard to do when you're annoyed about the ticket. Note in these last few examples we don't provide a role for the LLM to adopt. It's up to you if you'd like to do that, but in this case the situation is general enough that specifying a role is likely not that useful.
Example 5: An awkward conversation with a colleague
Your colleague's calendar is never up to date, which makes scheduling meetings difficult and you want to give them some unrequested feedback. This is probably my least favourite office situation and it's difficult to not come across passive-aggressive or heavy-handed.
Weak prompt
"Write an email asking someone to keep their calendar up to date because it's annoying when I want to schedule a meeting."
Stronger prompt
"I need to write a short, friendly email to a colleague asking them to keep their Outlook calendar up to date. The issue is that when their calendar isn't current, it makes scheduling team meetings much harder and wastes time for everyone involved. This has been a recurring problem with this particular colleague. The tone should be warm and collaborative, not critical or passive-aggressive. Frame it as a general team efficiency thing rather than singling them out. Maybe acknowledge that everyone's busy and calendars can slip, but emphasise that it really does help the whole team when availability is accurate. Keep it brief, around 100 to 150 words. It should feel like a friendly nudge, not a telling-off, even though it sort of is one. Use British English."
Why this works
A lot of effort in this last example is spent on tone guidance. The LLM has instructions on exactly what this should be and why. It's also been given the reason (scheduling is a pain) and the framing you want (team efficiency, not personal criticism). This nuance is a great example of good prompting because without it the LLM could default to either too passive or too direct a tone for your requirements.
Wrapping Up
Across these examples you'll notice some common themes:
- Who you are and what your situation is. The AI needs context about you to tailor its response.
- What role you want the AI to adopt? This may not always be necessary but it's useful for the LLM to help localise which persona it should adopt.
- What format you actually want the response in. A letter? A comparison? A list of ideas?
- Relevant details you want to include for context or to add into the final output. Dates, names, policy numbers, budget figures, evidence.
- The tone you want for the text and the intended audience. Professional but warm? Firm but polite? Friendly but clear?
- Format and length. Tell the LLM how long the response should be and how it should be structured. Without this, you'll often get something far too long.
Next time you use an AI tool, take an extra thirty seconds before submitting your prompt. Those few extra seconds will save you time refining and editing. Also it's worthwhile remembering that iteration is part of your process. If the first response isn't quite right, don't start from scratch. Tell the LLM what to change. "Make it shorter." "Less formal." "Add a section about budget." Treat it as a conversation, not a single shot.
Ultimately, the best way to get better at prompting is simply to practice and refine your prompting to include the specifics that are important to you. We're hoping this Explainer series will help you get started with some good fundamentals.
References
If you are keen to dive deeper into this topic here are some additional references: