What I value most isn’t a clever prompt. It’s being able to question an answer—and have my own thinking questioned, too.

This essay developed in conversation with ChatGPT (“Gip”). There’s a note about its role at the end.

I read a lot of articles about artificial intelligence. There’s certainly no shortage of advice: write a better prompt, assign the AI a role, give it more information, try this setting, avoid that one, etc., etc. Some of it is helpful. But after using ChatGPT extensively for years, I’ve found that what makes it most useful to me is harder to put into a list of tips.

I’m interested in what happens after the first answer. Can I question it? Can it question ME? Can we work through an idea without the whole exercise turning into an elaborate way of telling me what I wanted to hear in the first place?

I’ve used it for business administration, technical problems, entertainment production, travel planning, songwriting, this website, and a considerable assortment of things that simply interested or puzzled me. I’m not offering a formula that everyone needs to follow. This is what I’ve found valuable—and, just as importantly, where I think it can go wrong.

It started with a hamburger…

One morning, while watching a commercial for smash burgers, I found myself wondering why I would want someone to smash my hamburger. My mind went in a somewhat suspicious direction: perhaps somebody had figured out that a thinner patty could look substantial on a bun while containing less meat. Give the process an appealing name and perhaps you’ve also improved the profit margin. (That was my theory, anyway.)

ChatGPT gave me the conventional cooking explanation: pressing the meat against the hot griddle increases contact and encourages browning and crispy edges—the Maillard reaction. That explanation made sense. But it didn’t quite address what I was wondering about, which was whether the appeal of the technique and the economics of the portion could both be part of the story.

I pushed back about the geometry. If you flatten the same amount of meat, it spreads out. If you keep the diameter the same while making the patty thinner, there can be less meat. ChatGPT reconsidered that point, but I need to be clear about what the exchange did NOT establish: we hadn’t weighed anybody’s burgers or demonstrated why a restaurant chose to sell them that way.

The cooking explanation could still be correct, and my suspicion could still be wrong. Having a chatbot agree with my reasoning wouldn’t prove otherwise. In fact, if I kept pushing until it simply accommodated me, I’d have accomplished rather less than I might like to think!

What interested me was the chance to separate the questions: what does the cooking method do, what happens to the size of the patty, and what evidence would be needed to say anything about the business motive? Those aren’t all the same question. Sometimes I need help noticing that I’ve rolled several assumptions into one conclusion.

That is what I want from these conversations. Not an automatic endorsement, and not an automatic rebuttal, either. I want another explanation available alongside mine, with enough back-and-forth to work out where each one holds up.

Experience helps. It can also get in the way.

After more than five decades in professional entertainment and a family business, I don’t approach every situation as if I’ve never encountered people before. Neither does anyone else who has spent a working lifetime dealing with them. We develop a sense of how things tend to go, what people may leave unsaid, and why the official explanation isn’t always the entire explanation.

I think of some of that as ‘scar tissue’. You don’t necessarily remember every experience that taught you something, but the resulting reaction can be immediate: “I’ve seen this before.” That can be useful. It can also cause me to decide that a familiar-looking situation must have the same explanation as a previous one, when it doesn’t.

An AI system can draw on a great deal of recorded information, but it hasn’t lived a life. It hasn’t personally experienced the difference between how an organization says it operates and what happens when people actually have to work together. On the other hand, my having experienced something doesn’t mean my interpretation of it is the only reasonable one.

This is where I find the conversation worthwhile. I can explain why an answer seems incomplete in light of my experience, and I can ask what evidence would challenge my own explanation. If I’m extending a pattern farther than the facts justify, I want that pointed out. I may not especially ENJOY having it pointed out, but that’s a different matter.

Of course, the AI can miss the distinction, invent an explanation, or sound much more certain than it should. Disagreement isn’t automatically insight, any more than agreement is automatically validation. Both need examining. Sometimes the most useful outcome is simply recognizing that I don’t yet have enough information to be confident.

And sometimes I have to admit, “Fair enough. I was full of crap.” I’d prefer to discover that while thinking something through than after acting on it.

There’s context—and then there’s what it actually remembers.

People often talk about finding the perfect prompt. I understand why: the way a question is asked can make a difference. But for me, useful background matters at least as much. What am I trying to accomplish? What have I already tried? What responsibilities or limitations am I working around? What does a satisfactory result actually look like to ME?

Over time, conversations can build up that background. A technical problem may reveal something about how I process an explanation. Work on a creative project may clarify a preference that matters again later. When that context is available, a short “What do you think?” can mean considerably more than those four words would suggest to someone meeting me for the first time.

There is an important qualification here, though. Using ChatGPT for years does not mean every conversation is remembered in full, or that every mode or new task has access to the same history. OpenAI’s explanation of ChatGPT memory describes how memory works and its limitations. I can’t assume that something I explained months ago is available to the particular assistant I’m dealing with now.

So part of the work is supplying the relevant background, keeping useful notes, and correcting misunderstandings when they appear. If an answer sounds unlike me or overlooks something important, more eloquent wording won’t solve the underlying problem. The assistant may simply be missing information—or interpreting it incorrectly.

I also don’t want to confuse a useful conversation with an independent check of the facts. Asking the same system to reconsider can help, but sometimes what’s needed is an actual source, a measurement, or someone with relevant expertise. Two explanations agreeing with each other are not necessarily two independent pieces of evidence.

For that reason, I find ‘collaborator’ useful as a description of how I work with AI, provided we don’t take it too literally. It isn’t a person, and calling it “Gip” doesn’t make it one. I’m describing a process: bringing a question, considering an answer, challenging assumptions, and trying to improve the result. I remain responsible for what I do with it.

I don’t want ChatGPT to make my decisions for me. I also don’t want to dismiss a good explanation merely because it conflicts with mine. What I’m after is a better basis for deciding—not somebody, or something, that always lets me have the last word.

Perhaps that’s the most useful thing I can offer anyone experimenting with AI: give it the background it needs, correct it when it misunderstands you, and leave room for your own thinking to change as well. It’s less exciting than a secret setting, I suppose, but it has meant more to me than finding a particularly clever way to phrase a question.

And if, after all of that discussion and accumulated knowledge, we still manage to arrive at the wrong answer… well, there is one final benefit:

At least now there are two of you to blame.

This essay developed from an extended dialogue between Barry Knudsen and ChatGPT (“Gip”), which was used to challenge, organize and help draft the ideas presented here. Barry reviewed and takes responsibility for the final argument.

Originally published August 30, 2026. Revised September 8, 2026 for voice, clarity, and the qualifications above.

More personal writing → · About Barry