Practical AI for Everyone

“Garbage in = garbage out“. How you chat with AI matters because it shapes the answers you get. Confused by all the jargon around “prompt engineering” just trying to figure out how to get more useful answers?

Don’t worry – I’ve got your back. I’ll share my learnings with you after my “trials and tribulations” of 3 years of intensive prompting experience. I’ll show you how to prompt ChatGPT & Co. with 7 best practices and the simple “CTF” framework. Generally, note that AI outputs are not always reliable, so always review them critically.

Table of Contents

Effective Prompting is Powerful (& Easy!)

Using AI chatbots is easier than you think and doesn’t need special tech skills. The secret? It’s about being good at telling “someone” clearly what you need, like how you’d explain something to a new, eager “intern” or someone “just out of college”.

This skill of asking the right way is what helps you get great results for any use case (e.g. creative brainstorming, drafting outlines, generating stories etc.) It’s straightforward once you get the hang of it (“practice makes perfect”).

Don’t let tech jargon scare you. With a couple of simple tips and the hands-on blueprint (“CTF” method), you’ll prompt like a pro in no time…

The “7 Sins & Wins” of Prompting

First things first: Let’s set the scene with some general Do’s and Don’ts of prompting. Think of these as the 7 basic principles to “interview an AI” smartly:

1) Be clear, specific and… polite

Pretend you’re assigning a task to someone who’s keen but needs precise instructions. Also, be kind to the AI: It’s not just a sign of decency, but research shows “saying please and thank you” etc. also improves the quality of its outputs. For instance, instead of saying “write about AI technology,” try “please compose a 100-word paragraph on the latest AI developments this year.”

2) Engage in “roleplaying”

Research shows GenAI models give even more relevant answers if you put them into the right “perspective”. For example, instead of asking ChatGPT to “write a LinkedIn post about […]” tell it “You’re an expert content marketeer, please write a LinkedIn post about […]“. This also works for writing styles (e.g., “[…] write it like Shakespeare“).

3) Provide enough context

A little framing goes a long way. For example, if you’re asking for the continuation of a story, give it some relevant background first. It’s like catching up with co-workers in a meeting and making sure you’re all aligned on the current status, before discussing the next project stage. The more puzzle pieces there are already around the “hole”, the easier it is to complete it, right?

Bonus Tip: Let ChatGPT ask for missing context (add this to your prompt)

“[…] Whenever my input isn’t enough for you to perform a task effectively, ask clarifying questions to refine your response. However, when there’s enough context for meaningful progress, adopt an ’80:20′ approach – prioritizing actionable results over perfection. If you ask clarification questions, do it in a bundled way. […]”

4) Let it learn from examples

Like teaching a child, examples help AI grasp your requirements and generate better outputs. If you’re looking for a specific format or tone or problem-solving approach, provide samples to guide it. Put the examples e.g. at the end of your prompt like an appendix and refer to it. This is called “multi-shot prompting” (with multiple examples). (With one example, “one-shot prompting”. If you don’t give any examples, unsurprisingly “zero-shot prompting”.)

5) Break things down

If a concept is complex, break it down. One way to achieve this is to divide your instructions into smaller tasks. A surprisingly simple yet effective “trick” is to nudge AI towards “system 2” thinking by telling it to “think this through step-by-step“. Adding this to your prompt changes the way the AI approaches the question and improves quality. You can also use “reasoning models” that do that already per default (e.g. OpenAI’s “GPT-5 Thinking”). This is also called “chain of thought” prompting.

6) Don’t forget to review

Just like with humans, misunderstandings happen… Collaborate smartly with AI: let it do what it’s good at, you do what you’re good at. The “first mile” (providing enough input) and “last mile” (final touch) of any task are typically on you – with some “back and forth” in-between. If its response isn’t what you expect, consider how your prompt can be tweaked: Edit your message, play with follow-up prompts etc. Rinse and repeat.

You can even ask the AI itself to review and improve your prompt. Research (and experience) show that AI is surprisingly “skilled” at crafting prompts. Check out the following “meta-prompt” template (and share your experience in the comments).

Bonus: Try the “prompt of all prompts”

“Please help me (aka “the user”) craft the best prompt for [ChatGPT].
This is how: Provide 2 sections in each of your answers:

a) revised prompt (building on my input) with a succinct score (X/10) indicating its current “fit for purpose” with very concise feedback (regarding the prompt engineering criteria below) and
b) questions to clarify / refine what the prompt should do.

Iterate until I say the prompt is OK. Prompts should follow prompt engineering best practices, i.e., (loosely) structure prompts according to the following criteria and roughly in this order:

1) The user provides enough context details, e.g., the role/persona which ChatGPT should adopt.
2) The user breaks down the task/steps.
3) The user specifies the expected output (regarding formats, style, length etc.).
4) The user can optionally add example/s of an ideal solution as inspiration (“x-shot prompting”).
5) Generally, prompts shall be concise (but not simplistic), clear, specific and polite.

All needed user inputs are delineated with “[]”. When the prompt’s quality is good enough (e.g., reaches score of >8/10) give the user a hint. Let’s begin with my initial inputs: [enter your prompt goal / draft].”

7) Adopt an “interviewer’s mindset”

Prompting isn’t just technical – it’s a conversational skill, like interviews. Start with open-ended prompts to explore ideas, like “What strategies could improve team collaboration?” Then, use probing prompts such as “What if we introduced flexible work schedules?” to go deeper. Close with focused, closed-ended prompts like “Would measure X likely improve productivity?” to validate or refine certain ideas.

With these tips, you’re on your way to writing better prompts. You know what’s even better? You don’t have to memorize them. The following framework (“CTF”) incorporates these principles by default.

Use the Simple “CTF”-Framework

CTF means you’re providing proper Context, Task and Format – 3 building blocks which, if you put them together, let you hit the nail on the head:

Context: Again, this is the “backstory” of your request (incl. the mentioned tweaks like giving it a role to play etc.). Give the AI enough “briefing” so it gets the bigger picture of what you’re asking for and why.

Task: This is what you’re asking the AI to do. Remember? It’s where you’re crystal clear about what you need done when you’re delegating something. Laying out the goal, maybe even step by step, ensures less confusion about what you’re expecting.

Format: How would you like your answer served? For example, define how a report should be structured to fit your needs. Maybe you like bullet points or a full-on essay etc. Specifying the right tone helps you get your info in your preferred style.

Pretty intuitive and easy to remember, right? Worth mentioning is there are many prompting frameworks (e.g. “RASCEF”: Role, Action, Steps, Context, Examples, Format). But so far this one (“CTF”) convinced me the most with its balance of comprehensiveness and simplicity. But now let’s put the theory into practice…

Examples: The Good, The Bad & The …

Let’s have a look at an exemplary situation and compare the results of ChatGPT prompts A) without and B) with the CTF formula:

Prompt A (without “CTF” i.e. just a generic question):

“Write a paragraph why ‘GreenPlanet Corp.’ should adopt sustainable technology.“

What ChatGPT says to this (hint: Captain Obvious)?

Prompt B (with CTF framework):

“[Context] You are a green technology advisor. You were hired to assess the situation of ‘GreenPlanet Corp.’, a pioneering sustainability-focused car manufacturing company situated in San Francisco.
[Task] Compose a detailed and convincing proposal. Explain the significance of adopting renewable energy sources and highlight how green production technology can contribute to sustainability and cost savings. Approach the problem with the ‘problem-solution-benefit’ framework.
[Format] Please write max. 200 words, use a professional tone and articulate yourself in an easy-to-understand language. Use short sentences and format your response in a well-structured way with headings and bullet points.”

What does ChatGPT say now?

Judge (and try) yourself which version you’d consider more useful. While I used ChatGPT here, these principles are of course relevant for all alternatives (Copilot, Claude, Gemini etc.) as well.

Wrap-up: Welcome to the “AI Whisperer” Club

After this “tour de force” you can call yourself a “prompt engineer” with a handy prompting cheat sheet at your disposal. Kudos. If I were to summarize this all in 1 sentence:

“A problem well-stated is half-solved.” – Charles Kettering

We covered all basics, but this is just the start. There is more to cover in future posts (e.g. the specifics of prompting APIs or AI agents etc.). But rest assured, getting this far already puts you ahead of 90% of prompts out there.

I’m excited to hear your experiences, successes or mishaps with prompting. What is your go-to framework, maybe a different one? Share your thoughts in the comments (or get in touch) – and spread the word if it helped you.

Cheers,
John