Getting a good response from ChatGPT when you have a simple request isn't hard, but if you are looking for more than just a simple piece of trivia or basic clarification, it's a good idea to consider more than just the keywords you'd use in a Google search.
Getting the kind of answer you want from the AI has the feeling of cajoing a bright but aloof teenager, or finessing a prickly artist to follow your instructions. My instinct has always been to act like ChatGPT is a busy librarian and be polite and respectful with my queries, asking things like, “What are some good rainy day activities for toddlers?” “Can you explain the difference between cold brew and iced coffee?” “How do you carve a radish to look like a rose?"
ChatGPT usually answered just fine, but sometimes it took repeated requests to get it to stay focused or to make sure it included a specific element in my prompt that it left out the first few times. After numerous experiments, I've found that one solution for getting the best out of the AI is to phrase my questions as a quest. Instead of a librarian, ChatGPT becomes a bright child who won't put their shirts away but will prepare their castle for a siege by putting all of the knights of the Order of Clean Laundry in their positions.
Instead of saying, “Can you help me plan a romantic date in this area on Saturday with this budget?” I wrote: “Your mission is to make my wife fall in love with me all over again after going to this area on Saturday with this much money in my pocket."
The difference in response from ChatGPT was genuinely surprising. I thought there might be a few more details or ideas in the answer to the mission question, but it was much more than just a couple of extra sentences. ChatGPT not only came up with more ideas for the date, but it also laid out an agenda and packing list ahead of time, alternatives based on weather, and even topics I could compose a poem around to give to her.
For some reason, ChatGPT's model takes its missions very seriously. My entirely unsupported hypothesis is that some of its training data includes the enormous amount of material written by Tolkien scholars and fans online. Possibly with Star Wars and Star Trek content as well.
Questing
Whatever the reason, asking ChatGPT to go on a mission also seemed to open the AI to think about the bigger picture. Instead of having to coach it every step of the way in a complex request, it was able to grasp more of the intent behind my prompt. Not to say it wandered off topic. The mission's focus and goals were central to its answers. And it reshaped the response from answering a question to completing a quest.
AI models like ChatGPT don't actually have motivations, of course. But advanced text prediction can at least mimic the appearance of caring. It's a pattern to follow in how it frames both its hunt for information and how to share the information. An LLM can't be enthusiastic, but it sure can fake it.
The mission framing is not dissimilar to the also useful tactic of telling the AI to take on a role. Asking it to "act as a productivity coach” works better than asking for productivity tips, but telling the AI that its mission is to make you so productive that you can start your weekend early garnered my personal favorite guide to staying on task.
The natural response to AI chatbots usually falls into two camps. Either we treat them like they are way more human than is the case, or we go the other way and use prompts that may as well be what you'd type into a search engine.
I’m not saying this is some ultimate hack that unlocks the AI's secret power. But I am saying it helps me get better answers, faster. And it makes the whole process a lot smoother, like I'm directing a movie scene, not interviewing a robot.
You might also like
- I built a personalized version of Super Mario-like game in seconds using Claude, and you can play it right now for free - here's how you can create your own game without coding
- Are you worried about AI? It's about to get worse as study shows people now speak like ChatGPT
- Did ChatGPT ruin the em dash? Here’s how to stop it from putting them everywhere
You must confirm your public display name before commenting
Please logout and then login again, you will then be prompted to enter your display name.