(Image credit: Shutterstock)
I've been using ChatGPT since it was first rolled out in late fall of 2022. Like most people, when I first started querying it, I had no idea everything it was capable of — but after testing every new model extensively since then, I've become a power user. ChatGPT has had 34 models! And yet, getting useful answers has almost nothing to do with which model you're using and has everything to do with how you talk to it.I've watched people upgrade to the newest, most expensive version and still get mediocre results, while a friend on the free tier pulls genuinely useful work out of it. The difference is almost never the tool, but the instructions (prompts). This is important for those who might be upset about the recent suspension of Claude's Fable 5 or those of us still grieving ChatGPT-4o.So, when someone tells me ChatGPT gave them a bad answer, I ask one question: What exactly did you type?Nine times out of ten, that's where the problem is. After thousands of prompts and more late-night experiments than I'd like to admit, these are the five habits that consistently move me from generic answers to genuinely useful ones.1. Tell ChatGPT who it's supposed to beIf you've been reading my features for awhile, you may be familiar with the ‘3-word rule’ to get smarter answers. This particular rule helps remove any guesswork for the model. Most people assume the model knows what perspective to take. But even the smartest model works better when it doesn't have to guess. When it's left to do that, it averages — and average is exactly what a generic answer is.So instead of: "Help me create a presentation based on these meeting notes." Try using the prompt: "You're an experienced marketing expert who specializes in audience engagement. Help me create a presentation based on these notes and pertinent current trends." Get instant access to breaking news, the hottest reviews, great deals and helpful tips.The shift is immediate. Once you assign a role, the model stops writing for everyone and starts drawing on a narrower, more expert slice of what it knows.When you think of AI like a consultant, while alos precisely defining the job, the better the advice — and "be a sharp marketing expert" gets you somewhere "just help" never will.2. Give it context before you ask for anythingAlthough it may seem to be clairvoyant at times, especially if you have Memory enabled, ChatGPT isn't a mind reader. Although, people treat it like one constantly. In many situations, users spend more words setting up the situation than asking the actual question."I'm drafting a cover letter for a position that seems a bit out of my skills, but I'm hoping to switch careers after a layoff. It needs to be conversational, practical and easy to follow. Here's my draft."That one paragraph does a lot of quiet work. It tells the model the audience, the tone, and the goal — three things it would otherwise invent on its own, usually wrong. The context of switching careers after a layoff is essential otherwise the AI may have drafted the cover letter based on your current resume, which would have been awkward and perhaps even missed the mark.Without context, you get an answer built for the average reader. With it, you get one built for yours. The gap between those two is most of what people mistake for "the model isn't smart enough."3. Ask it to argue with youMost people use ChatGPT to confirm what they already believe. However, I've found that when you ask for it to slow down and ask questions that challenge your assumptions, the responses are much better.When I'm working through an idea, I'll ask: "What's the weakest part of this argument?"or "Give me three reasons this could fail."This habit has quietly saved me from publishing half-baked ideas more times than I can count. The model is a surprisingly good devil's advocate — but only when you explicitly hand it permission to disagree. Left alone, it defaults to agreeable, because agreeable is what most people reward it for.The trick is simply to ask for the criticism you'd rather not hear instead of the validation you secretly want.4. Refine — don't restartHere's where most people quit too early. The first answer isn't perfect, so they close the tab and start over. I do the opposite. My best results almost always come after a few rounds of back-and-forth and by using the branching option.Each time you nudge the AI with a piece of feedback that the model couldn't have known to apply on its own, you get a better response. Treat the first response as a rough draft you're editing together, not a verdict and the conversation starts compounding instead of resetting.5. Tell it what success actually looks likeAgain, the goal here is to avoid being overly generic. This prompt tyle is something most people don't do, but it's crucial. Rather than just describing the task, help the AI by describing the outcome. For example: "My goal is to achieve 200 more TikTok follwers by the end of the month. Based on my profile, what do I need to do?" or, this one, which I use all the time:"Create a list that helps an overwhelmed parent feel calmer in five minutes."Now the model has a target, not just an assignment. A huge share of disappointing prompts fail right here: the person explained what they wanted done but never said what a good version of it would feel like. Name the finish line, and the model can actually run toward it.My takeawayThe people getting the most from AI right now are the ones thoroughly prompting in more detail. You don't need to go overboard and write a novel or whitepaper into the chatbox each time, just a strong confidence of your goals.And that's genuinely good news. It means you don't need the newest release or the priciest plan to get dramatically more out of it. You really just need to create a habit of better instructions.Using ChatGPT has become the daily norm for many people. But getting the most out of this AI (or really any chatbot) comes down to the quality of your instructions. Quality in, quality out.Follow Amanda Caswell and stay ahead of the AI curve
