I stopped giving ChatGPT, Claude, and Gemini detailed prompts for a week, and one struggled more than I expected

One piece of prompting advice you've likely heard at least a few times before is to give AI chatbots as much context as possible. Don't leave much room for assumptions or interpretation, and be as clear about your requirements as you can to ensure the final result is exactly what you're looking for. It's advice I've followed myself, and when you think about it, it makes complete sense. How can you reasonably expect an AI chatbot to give you exactly what you want if you don't tell it what that is? However, what we tend to forget is that AI models have come a long, long way since the early days of ChatGPT. That got me wondering: do we really need to be so specific with our prompts anymore? AI labs are increasingly pushing the narrative that their models can reason through complex problems, understand what we're trying to accomplish, and make decisions on their own. So, shouldn't they also be getting better at figuring out what we want without us having to spell out every little detail? I decided to put that theory to the test. For a week, I ditched my usual detailed prompts and gave ChatGPT, Claude, and Gemini the same deliberately vague instructions across five everyday tasks. As it turns out, one chatbot needed a lot more hand-holding than the other two. It all started with a piece of advice from Claude Code's creator What happens when you stop telling AI what to do? I've always thought the best prompting advice can come from the people building the AI tools we use every day. So, when I came across a conversation featuring Boris Cherny, the creator of Claude Code, and Cat Wu, Anthropic's Head of Product for Claude Code, one particular piece of advice caught my attention. Cherny explained how prompting has changed as AI models have gotten smarter. In the past, getting the best results often meant carefully crafting your prompts, providing detailed instructions, and making sure the AI had all the context it could possibly need. This is where all the advice stemmed from! According to Cherny, newer models are increasingly capable of figuring things out on their own, so there's less need to micromanage them with overly detailed prompts. Of course, there's a difference between giving an AI unnecessary instructions and leaving out important information altogether. Cherny's advice wasn't necessarily to stop providing context, but rather to trust newer models to figure out how to accomplish a task instead of spelling out every step for them. Still, it got me thinking. What if I took that advice a step further and deliberately stopped giving AI chatbots all the information they'd normally need? Would they ask the right follow-up questions, make reasonable assumptions, or simply give me generic answers? I kept the tasks simple and the prompts even simpler I gave three chatbots the bare minimum To keep this experiment as fair as possible, I disabled memory features and used the anonymous or temporary chat modes available in each chatbot. That way, none of them could rely on information from our previous conversations to figure out what I wanted. I then picked five everyday tasks that people commonly turn to AI for, each requiring a different set of skills: Choosing between an e-ink tablet and an iPad: Could the chatbots help me make a purchasing decision without knowing my budget, preferences, or intended use? Planning a three-day trip to Tokyo: Could they put together a practical itinerary without knowing anything about my interests, budget, or travel preferences? Explaining recursion: Could they break down a programming concept without knowing my experience level or how I prefer to learn? Brainstorming YouTube channel ideas: Could they come up with creative, useful ideas without knowing my interests, target audience, or the kind of content I wanted to create? Building a business website: Could they create something usable without knowing what kind of business I had in mind or what features the website needed? I gave ChatGPT, Claude, and Gemini the exact same prompt for each task, without adding any extra instructions upfront. I was particularly interested in whether they'd ask follow-up questions, make reasonable assumptions, or simply work with whatever little information I'd provided. I also wanted to keep my own preferences from influencing the results too much. So, rather than relying entirely on my judgment, I had a separate AI evaluate the responses without knowing which chatbot had generated them. I labeled the responses Exhibit A, B, and C, and asked it to compare how well each one handled the task. Interestingly, my judgment and the anonymous evaluation were pretty much on the same page, especially when it came to which chatbot struggled the most. Gemini struggled to keep up with ChatGPT and Claude Gemini kept answering when I wanted it to help While you can see snippets of the responses above and get a general idea of how each chatbot performed, I found that Gemini consistently struggled the most. That's not to say its responses were outright bad or unuseable. In fact, it understood what I was asking in all five tests. The problem was that, compared to ChatGPT and Claude, Gemini rarely went beyond the bare minimum. For instance, when I asked all three chatbots to explain recursion, ChatGPT went as far as providing a step-by-step visual walkthrough, a Python code example, and even a multiple-choice question to test my understanding. Claude also did a great job of breaking down the concept, particularly when explaining how recursive calls return values. Gemini, on the other hand, stuck to simple analogies and explanations without including a single code example. While its explanation was easy enough to follow, it felt like something I'd need to follow up on to actually understand the concept. I noticed a similar pattern when I asked the chatbots to help me choose between an e-ink tablet and an iPad. ChatGPT provided an interactive questionnaire to narrow down my needs, while Claude offered a concise but practical comparison. Gemini explained the differences between the two devices, but didn't do nearly as much to help me arrive at a decision. Even when planning a three-day trip to Tokyo, Gemini fell behind. ChatGPT put together a detailed itinerary with a map, estimated costs, and ways to personalize the trip, while Claude's itinerary was thoughtfully organized and easy to navigate. Gemini's itinerary was reasonable, but an interactive feature it offered reportedly failed to load during testing. Ultimately, that was my biggest issue with Gemini throughout this experiment. It generally understood what I was asking, but it often stopped at answering the question rather than helping me accomplish what I'd actually set out to do. Claude was the only one that knew when to ask questions ChatGPT preferred to answer first and ask questions later While there's not much I have to say about Gemini's approach beyond the fact that its responses were bare-bones and disappointing as a whole, ChatGPT and Claude were a different story. Both did a pretty good job of making up for the lack of detail in my prompts, but what I found interesting was how differently they went about it. ChatGPT almost always took the initiative and tried to give me as complete an answer as possible, even when I hadn't provided nearly enough information. It would then offer ways to personalize or refine the response further. Instead of simply asking me to provide more details, ChatGPT would often generate an interactive questionnaire with multiple-choice options I could select. For instance, when I asked whether I should buy an e-ink tablet or an iPad, it followed up by asking about my budget and what I primarily intended to use the device for. All I had to do was select the options that applied to me and hit a button, and ChatGPT would automatically submit another prompt based on my answers to generate a more personalized recommendation. This might sound like a small detail, but I think it made a massive difference to the overall experience. While Claude and Gemini would also occasionally ask about my preferences, I generally had to type out my answers and send another message myself. ChatGPT essentially eliminated that extra friction. I could give it a vague prompt, get a useful response, and then refine it with just a few clicks instead of having to come up with another prompt myself. The only exception, though, was when I asked all three chatbots to build me a business website. ChatGPT immediately jumped into action and generated a downloadable HTML website, while Gemini provided a fairly generic template. Claude, on the other hand, asked me for more information about the business before writing any code. Once I provided those details, Claude built a website for a fictional coffee shop called Brew & Bloom, and the end result felt much more aligned with what I'd asked for. ChatGPT's willingness to take the initiative had been an advantage in most of my other tests, but this time, it worked against it. By making assumptions instead of asking questions upfront, it ended up creating something that was functional but far less personalized! Ultimately, I think this was one of the most interesting takeaways from the entire experiment. Sometimes, having an AI chatbot figure things out on its own is exactly what you want. Other times, knowing when not to make assumptions can make all the difference. The overall result of the experiment was clear, though Gemini still seems to struggle when you leave out details from your prompts, whereas ChatGPT and Claude are much better at filling in the gaps, whether that's by making reasonable assumptions, anticipating what you might need next, or asking for more information when necessary. That's not to say detailed prompts are a thing of the past. Providing more context will almost always help when you have specific requirements in mind. After this experiment though, I'm convinced that you don't always need to spend ages crafting the perfect prompt to get a useful response. At least, not with every chatbot.