After hundreds of conversations, requests and outputs with ChatGPT, I’ve identified five things that I do that others don’t that makes my interactions more productive.

1. Don’t start with a prompt. Start with a conversation.

What? But Beth, everyone talks about starting with a prompt!

Yeah, I know. But I am not a good prompt writer. And if I waited for my brain to compose the perfect prompt, I’d stare at a blank screen all day.

Most of the time I jump in with something like, “Ok, today’s challenge is….” And I just describe what’s going on, such as

• I need help creating great examples of AI applications for people in the audience during my keynote tomorrow.
• I gotta figure out what all this jargon means.
• I need help classifying purchases on my credit card statements so I can do my taxes.

Then I work with the tool to figure out how to proceed.

2. Poke and prod and challenge.

Sometimes ChatGPT pours out elegant prose, and I’m temporarily blinded by the quality. But before I consider the task done, I take a good look. I ask questions such as:

• Do you really think this is true, or are you writing what you think I want to hear?
• Are you sure about that?
• Is this the best you can do?
• Do you think this sounds like me?

Here’s a typical interaction:

3. Direct and redirect, directly.

AI chatbots can’t read your mind, and it’s easy for them to go off track. I’m not shy about stopping faulty thinking or expressing my displeasure (as evidenced by this list of the top five arguments I’ve had with ChatGPT). Some of my favorite responses:

• Dude, really???
• Are you kidding me with this?
• Nope.
• You really missed the boat on this.
• Wrong.

4. Keep at it unless all hope is gone.

My threads with ChatGPT are reallllyyyy long. Most of the time I start a conversation and keep returning to it so we get better outputs and deeper understanding.

Sometimes, though, it’s evident that the chatbot and I are not on the same planet. So I might reformulate the prompt in a new thread (or even move to another chatbot platform) to clear both our heads.

5. Paste the final draft back into the conversation.

When I was in grad school for journalism at the University of Alabama, I got an internship for the parent company of Southern Living magazine. I’d write a press release, and my boss would take it and say, “Great! Now this shows me exactly what I don’t want.” And then she’d write her own version. It was hurtful, but we ended up with better copy.

Since I can’t really hurt ChatGPT’s feelings, I continue this technique. I ask ChatGPT to write things for me, but rarely does the output really ring true. So I’ll let it do its thing, then I’ll go back and paste what I finally created to show it what I was hoping for.

That’s actually what I’m doing with this blog post. I wrote every word, but I talked to ChatGPT for more than an hour and asked for 5+ versions of a potential blog post until I had clarity to write my own.

Bonus: Know when to give up.

Sometimes I end my conversation with, “I give up. You’re just not getting this. I’ll just do it myself.”

If it takes you three hours of conversations with a chatbot to do something that would take you 30 minutes to finish on your own, you’re doing something wrong. Sometimes it’s more efficient to just throw in the tech towel and go back to the way we did it in 2021.

What are your favorite AI techniques?

How do you get the best out of our AI responses? Share them below so we can all keep learning.