A new research study has produced a set of prompts so that we can prompt like pros! Of course, I have a whole chapter in my book AI Playbook on prompting! But I do love research to test methods and approaches, and I had to share this with you.
This episode is all about the moves that make your prompts work harder. A 2024 paper called Principled Instructions Are All You Need for Questioning tested lots of prompt patterns and pulled out 26 simple instructions that reliably boosted model performance. Later studies on “automatic prompt optimisation” more or less confirm the same thing: you do not need magic incantations, you need clear instructions, structure, and a bit of discipline.
So rather than unpacking the paper line by line, this episode turns those ideas into prompts you can actually paste into ChatGPT or your favourite model. Think of it as a short menu of patterns you can mix and match, depending on whether you want clarity, creativity, safety, or structure.
1. Clarity and audience prompts
Here are a few ways to get explanations that match where you are, not where the model assumes you are.
- “Your task is to explain [topic] using clear, plain English for a curious beginner. Avoid jargon unless you define it.”
- “Explain [topic] as if I am 11 years old, then give me a short, more technical version for an expert audience.”
- “Summarise [document or issue] for a time‑poor decision‑maker: three bullet points on what it is, why it matters, and what decision is needed.”
These prompts bring in audience, purpose, and style, instead of assuming the model will guess them.
2. Breaking work into steps
Complex tasks usually die in vague, one‑shot prompts. These patterns force the model to slow down and show its working.
- “Your task is to help me with [goal]. First, list the key steps we should take. Then ask me questions to fill any gaps. Only after that, propose a detailed plan.”
- “Think step by step. Restate the problem, list your assumptions, outline your approach in bullets, then carry it out, numbering each step.”
- “For this question, generate two different solution paths, compare their pros and cons, and then recommend one with a brief explanation.”
You can almost hear the difference in the output: it becomes more like a collaborative working session than a single guess.
3. Role and rules
Assigning a role and clear constraints is one of the strongest levers in the research.
- “You are a senior [role, e.g. ‘communications strategist in local government’]. Your task is to review my draft and suggest concrete improvements that make it clearer and more persuasive, without changing the underlying message.”
- “Act as a critical friend. First, tell me what is strong about this idea. Then list the main risks or weaknesses. Finish with a revised version that addresses the top three issues.”
- “From now on, treat everything I send as part of one project. Use UK English, keep the tone professional but warm, and always provide headings and bullet points where helpful.”
Here you are encoding role, tone, language and formatting as explicit rules, rather than hoping the system remembers.
4. Bias, safety and uncertainty
Modern guidance stresses asking the model to surface uncertainty and avoid stereotypes.
- “Ensure that your answer is unbiased and does not rely on stereotypes. If the question could lead to unfair assumptions, briefly flag them and suggest a fairer framing.”
- “If you are uncertain or information is missing, say so explicitly and tell me what extra context you would need to give a stronger answer.”
- “When discussing people or groups, focus on behaviours, systems, and evidence rather than personal characteristics or clichés.”
These prompts are especially useful in policy, education, HR, and any domain where a lazy stereotype can do real harm.
5. Editing and style‑matching
Finally, a few patterns for when you want the model to edit rather than overwrite your voice.
- “Improve the grammar, clarity, and flow of this text, but keep my tone, level of formality, and sentence length as similar as possible.”
- “Here is a writing sample that shows my style. Please learn this style, then draft a new piece on [topic] that feels like it was written by the same person.”
- “Suggest three alternative openings and three alternative closing paragraphs for this article, keeping the middle section unchanged.”
These are small, surgical instructions that tell the model exactly what to touch and what to leave alone.
About the Podcast Host
Joanne Sweeney is a specialist in public-sector marketing and digital communications. She’s the author of Public Sector Marketing Pro and the creator of the Professional Diploma in AI for Public Sector Marketing. With years of experience in training, consulting, and advocating for digital transformation, Joanne provides insight and guidance for government agencies navigating the digital landscape. She is also the host of the award-winning Public Sector Marketing Show podcast.
