Mastering AI tools at work isn’t about spending 10,000 hours perfecting your craft. According to Professor Ethan Mollick from Wharton, you only need about 10 hours of focused practice to become competent with AI tools. If you can dedicate that time and apply the right techniques, you’ll quickly move from beginner to someone who’s extracting real value from these powerful tools.
But here’s the thing: most people approach AI prompting all wrong. They get caught up in “prompt engineering” – trying to craft the perfect one-time prompt that delivers exactly what they want. Not only is this overkill, it’s also unnecessarily complicated. There’s a better way: conversational prompting.
The Two Types of Prompting
Before diving into techniques, let’s clarify the distinction between two approaches:
- Conversational Prompting: Interactive, back-and-forth exchanges with AI
- Engineered Prompting: Carefully crafted, standalone prompts designed to run without human intervention
The key question to ask yourself is: “Will this run without me?” If the answer is no, conversational prompting is your best bet. If you need something that runs automatically and consistently, then engineered prompting makes sense.
Today, we’re focusing on conversational prompting – and the techniques below will transform how you work with AI assistants like ChatGPT, Claude, Gemini, or any other model you prefer. These techniques are just a starting point and can be readily incorporated into a more thorough conversational prompting workflow.
1. You Don’t Need to One-Shot It
Have a conversation. Iterate. Improve.
The biggest mistake people make is trying to get the perfect result on their first attempt. Instead, think of your initial prompt as a starting point for a conversation.
For example, if you’re looking for presentation examples, don’t stress about including every detail upfront. Start with your core request, then add context as you think of it:
- Initial prompt: “I need two examples for a presentation”
- Follow-up: “Actually, these need to be about ad tech”
- Next iteration: “Make these brief examples, not detailed case studies”
Brain dump what comes to mind initially, then refine as you go. This approach is far more efficient than spending hours crafting a perfect initial prompt.
2. Context Is King, Queen, and Crown
The AI only knows what you tell it.
While AI models have vast knowledge, they have no implicit understanding of who you are or what you’re trying to accomplish. You need to provide direction on which parts of their knowledge are relevant to your specific situation.
Instead of asking to “analyze trends in a dataset,” be specific:
- What type of trends? (CPM trends)
- What timeline? (Last quarter)
- For what purpose? (Budget planning for next campaign)
The more context you provide, the better your results will be. Think about all the relevant information the AI needs to give you exactly what you want.
3. Write Like Slack, Not Google
Be conversational, not robotic.
AI chat interfaces are optimized for conversational data – they’re trained on human-to-human interactions. So talk to them like you would talk to a colleague, not like you’re entering a Google search query.
Instead of: “best bike wirecutter”
Try: “I’m looking for a good commuter bike recommendation. My budget is around $800, I’ll be riding about 5 miles each way to work, and I want something reliable that won’t require tons of maintenance. I’d prefer something from a reputable review source.”
Pro tip: Try using Perplexity instead of Google as your default search engine for a week. It bridges the gap between traditional search and conversational AI, helping you get comfortable with this more natural prompting style.
4. Be Crystal Clear About Format
Tell the AI exactly how you want your answer structured.
AI assistants excel when you specify not just what you want, but how you want it delivered.
Instead of: “Give me five examples”
Try: “Give me five examples in a table format, with one column showing a bad example and another showing a good example side by side”
Other formatting examples:
- “Summarize this article in five bullet points”
- “Explain this topic for beginners in three short paragraphs”
- “Create a numbered list with brief descriptions for each item”
The more specific you are about format, the more useful your results will be.
5. Ask for Multiple Options
Let the AI do the heavy lifting of generating options.
One of AI’s greatest strengths is its infinite patience. Instead of generating one option and then asking for alternatives, request multiple options upfront and pick your favorite.
“Give me 5 different headline options for this blog post” is much more efficient than going back and forth five times asking for alternatives.
This approach works especially well for creative tasks like:
- Writing headlines or titles
- Brainstorming ideas
- Creating different versions of content
- Generating multiple approaches to a problem
6. Use Language Creatively
Push the AI in the direction you want, even if it sounds abstract.
AI models are exceptionally good at working with language, so don’t hesitate to use descriptive, even unusual language to guide them toward your desired outcome.
Try phrases like:
- “Make this 100 times more specific”
- “Make this weirder”
- “Transform this to sound more professional”
- “Reformat this for a technical audience”
- “Synthesize these ideas into something completely new”
You can also ask for combinations: “Give me 20 ideas, then I’ll tell you which 5 are my favorites, and you can create 10 more that are similar to my preferred ones.”
7. Break Down Complex Tasks
Decompose big requests into smaller, manageable pieces.
Just like with human collaboration, trying to accomplish too much at once leads to poor results. If you find the AI getting confused or providing irrelevant responses, break your task into smaller components.
Instead of: “Create all the content for this slide including the title, subtitle, example, and help me find a relevant GIF”
Try this sequence:
- “Give me a title for a slide about X”
- “Now create a subtitle that complements this title”
- “What keywords would help me find a relevant GIF for this topic?”
- “Create a concrete example that illustrates this concept”
This approach makes it easier to iterate and improve each component individually.
8. Show, Don’t Just Tell
Examples are worth a thousand words of explanation.
Providing examples of what you’re looking for will dramatically improve your results. This works both intuitively (like with humans) and mathematically (it activates relevant parts of the AI model’s knowledge).
The difference between “zero-shot” prompting (no examples) and “one-shot” prompting (one example) is substantial. Even a single example will guide the AI much more effectively than detailed descriptions alone.
Example in action: Instead of just asking for a subtitle, provide examples of subtitles from other slides you’ve created. This gives the AI implicit understanding of your style, tone, and structure preferences without requiring detailed explanations.
When You Hit a Wall: A Troubleshooting Guide
Even with these techniques, you’ll occasionally get stuck. Here’s your three-step recovery plan:
1. Add More Context
Think about what additional information the AI might need. What relevant details haven’t you shared yet? Don’t worry about perfect formatting – just dump whatever seems relevant to your request.
2. Start a New Chat
Sometimes conversations accumulate too many irrelevant details or errors. Even when you correct mistakes, the AI can get biased toward thinking about the wrong things (like when someone tells you not to think of a purple elephant, and that’s exactly what you visualize). Starting fresh often solves this.
3. Try Another Model
Different AI models have different strengths:
- Grok: More creative and unconventional
- Claude: Better written, more narrative flow
- ChatGPT: Well-rounded for most tasks
- Gemini: Strong analytical capabilities
Most models offer generous free tiers, so experiment to find what works best for your specific needs.
Bonus tip: For important projects, try running the same prompt through multiple models simultaneously and compare results.
Your Conversational Prompting Toolkit
Remember these eight core techniques:
- Have conversations – iterate and improve rather than one-shotting
- Provide rich context – the AI only knows what you tell it
- Write conversationally – like Slack, not Google
- Specify format clearly – tell the AI exactly how you want results structured
- Request multiple options – let AI generate choices for you to select from
- Use creative language – push AI in your desired direction with descriptive prompts
- Break down complex tasks – decompose big requests into manageable pieces
- Provide examples – show the AI what good looks like
The goal isn’t perfection on the first try – it’s having productive conversations that get you exactly what you need. With just 10 hours of practice using these techniques, you’ll find yourself working with AI as naturally as you collaborate with your best colleagues.
Start with one or two of these techniques, get comfortable, then gradually incorporate the others. Before you know it, conversational prompting will become second nature, and you’ll wonder how you ever worked without it.
This is a written version of a video I posted, if you’d like to watch the video: