How to Talk to AI: A Simple Guide to Getting Better Results

ChatGPT, Claude, Gemini, Grok - they are all remarkable in their own right, but I’m not sure we’re squeezing the absolute most out of their wide-ranging capabilities.

I'm talking about our prompts or the questions we ask - we can do better.

And when I refer to we I mean me. Just yesterday I wrote a piece of code for a client that didn’t work. I didn’t have the time, energy, or motivation to deal with it so I slapped it into Claude and asked:

“Can you fix this, please?” 🤦‍♂️

So how could I have done better with my lazy 'fix this' moment?

Step 1: Start with a clear goal

Keeping in the “fix my code theme from above” the first part I failed to do is provide Claude with a goal. Instead of saying "Fix this code," I would have had better results with something like:

"I need help fixing this code. What I’m trying to do is to analyze year-on-year trends however it’s not working as intended." 

Being specific about you are doing/have done and what you need from the AI. I’m priming the AI to dive deeper than just “code help” and removing all unnecessary parts enabling it to focus on the analytical mechanics of the code bringing it into the center of the domain in which we’re working.

Step 2: Outline the response you want

The second part I messed up was not explaining how I wanted the help. Instead of letting Claude guess, I should have said something like:

"Please analyze the code and provide: the specific issues you find, a corrected version of the code, and explanations for what each fix addresses. Also point out any potential aspects I may have missed."

Being clear about the parts I want to see helps get answers that will naturally align to the world outside of AI, rather than just a fixed piece of code we don't understand.

Step 3: Add guardrails

Here's where my lazy prompt really failed me.

When you're working with code, you usually have a mental list of all those little things that could go wrong - messy data, weird formatting, numbers that look right but are seriously wrong. I didn't mention any of these to Claude.

What I should have said:

"The code needs to handle missing months in the data, work with different date formats, and the calculations must match our existing quarterly reporting numbers. Also, please flag any potential performance issues with large datasets."

These warnings help prevent solutions that might technically work but would cause problems down the line

Step 4: Context Dump

Finally, I should have provided any relevant background information even if it’s only remotely related. If you think it could help, add it:

"The code worked fine with our US data but started failing when we added international sales numbers. The output feeds into executive dashboards, so accuracy is crucial. We're particularly interested in spotting unusual seasonal patterns."

This context helps the AI understand not just the code, but its place in the bigger picture.

The takeaway

So what's the real difference between the approaches?

With even a slightly structured prompt we're not just getting better answers - we're getting something we can use and trust.

And if you boil down all four steps, they have one thing in common - they reduce ambiguity.

I know you'll be tired, rushed, or just not in the mood to write perfect prompts - I do it 20+ times a day myself.

But when the stakes are high or you need quality help, taking an extra minute to frame your request properly can save hours of frustration.

The result? An AI-powered super-you.

TOGETHER WITH THE FYXER AI

Hire Faster with Fyxer AI

Fyxer, your AI Executive Assistant, streamlines your recruitment process:

• Neatly organized emails.

• Replies to candidates tailored to your tone of voice.

• Crisp notes from every interview and meeting.

7-Day Free Trial

WANT TO DIVE DEEPER?

While we'll explore more advanced prompting techniques in future issues, here are some solid resources if you're keen to get ahead:

• Anthropic/Claude Prompt Library

• OpenAI/ChatGPT Prompt Best Practices

We've barely scratched the surface of what's possible, but hey - everyone has to start somewhere. Even me.