Claude has become one of the AI tools I use most often, whether I’m brainstorming, planning something, vibe coding, researching a topic or simply trying to make sense of a complicated question. But using an AI chatbot every day doesn't necessarily mean you're using it well.
In fact, some of the habits that feel like they should make Claude better can have the opposite effect.
That’s what caught my attention when I came across a list of 27 Claude tips from AI consultant Ruben Hassid, who says he's spent more than 1,800 hours using Anthropic's chatbot.
Some of his advice goes against the way many of us have been taught to use AI. Keep everything organized in Projects? Maybe not. Keep correcting Claude until it understands what you want? That could actually make things worse. Spend 10 minutes crafting the perfect prompt? There may be a better approach.
Not every tip will make sense for every Claude user, and some of Hassid's recommendations are based on his own experience rather than hard-and-fast rules. But several are worth trying if you use Claude regularly.
Here are five Claude habits worth breaking — and what to try instead.
1. Stop correcting Claude over and over
We've all had this conversation with an AI chatbot. You ask Claude for something. It gets it wrong and then you go back and forth trying to get the response you're actually looking for — all while wasting your tokens in the meantime.
Even when you explain the problem in even more detail or add another layer of clarification, before long, you're six messages deep trying to rescue an answer that wasn't particularly good to begin with.
Hassid recommends doing something much simpler: go back and edit the original prompt. This is a feature that a lot of users overlook but it's one of the best ways to avoid correcting every wrong answer, which in turn, keeps adding to the conversation.
That's because every correction you add becomes part of the conversation Claude is working from. Instead of giving the model one clear instruction, you can end up with a growing trail of bad outputs, corrections and exceptions.
Editing the original prompt gives Claude another shot with cleaner instructions.
For example, imagine you asked: Plan a three-day trip to Boston.
Claude gives you an itinerary packed with museums, historical sites and early mornings. But what you really wanted was a relaxed weekend centered around food and neighborhoods.
You could spend several messages correcting the itinerary.
Or you could edit the original: Plan a relaxed three-day trip to Boston centered around great food and interesting neighborhoods. Include no more than one museum, avoid early mornings and leave plenty of unplanned time.
Same request. Much better context.
Try it: If Claude completely misses the mark, resist the urge to argue with it. Edit your previous prompt and regenerate the answer instead.
2. Stop keeping every conversation alive forever
This one is incredibly easy to do. Once Claude understands what you're working on, abandoning that conversation can feel wasteful. So you keep going...and going!
Eventually the chat contains previous questions, rejected ideas, unrelated tangents, corrections and instructions that stopped being relevant 30 messages ago.
Hassid recommends starting fresh more often, arguing that very long conversations can eventually make Claude less effective.
There isn't a magic number of messages where Claude suddenly becomes worse, and Claude's context window is designed to handle a significant amount of information. But context isn't the same thing as useful context.
A model having access to something doesn't necessarily mean you want that information influencing its next response.
Imagine using the same Claude conversation to plan a vacation for weeks. You've discussed five hotels, changed the dates twice, ruled out three destinations and completely changed your budget.
Claude may technically have all that information available, but you're now asking it to distinguish between what you wanted two weeks ago and what you want today. Sometimes the clean slate is the advantage.
Try it: When a conversation starts feeling cluttered, ask Claude to summarize the important information and decisions you've made. Paste that summary into a fresh chat and continue from there.
3. Stop obsessing over the perfect prompt
The internet has spent the past few years convincing us that getting great AI results requires elaborate prompts. A few minutes on X or Instagram and you'll be swamped with plenty of prompts.
But Hassid makes a compelling case for doing something considerably less polished: brain dumping.
Instead of spending several minutes turning an idea into the perfect prompt, simply tell Claude everything you're thinking.
This works particularly well with voice input because we naturally provide more context when we're talking. We mention what we're trying to accomplish, what we've already tried, what we don't like and the half-formed thoughts we might remove if we were carefully typing a prompt.
For example, instead of: Create a weekly meal plan for a family of four.
Try the prompt: I need dinners for the next five days. I don't want anything that takes more than about 30 minutes, we're trying to spend less on takeout, one person hates fish, I already have chicken and pasta in the house and Wednesday is going to be really busy, so that meal needs to be ridiculously easy. Help me figure this out.
The second prompt isn't perfect, but it's packed with useful information.
Try it: Open Claude and dictate your request as though you're explaining the problem to a friend. Don't organize it first. Let Claude organize it for you.
4. Stop telling Claude only what you want
Quite often AI users will prompt an assistant with "Make this better." "Give me something more interesting." "Make this easier." "Be more creative."
And while they will probably give you answers, these instructions are incredibly subjective. Claude has to guess what "better" or "interesting" means to you.
Hassid recommends giving the model negative examples — showing Claude what you don't want. Say you're trying to come up with ideas for a 10-year-old's birthday party.
Claude suggests a trampoline park, bowling, laser tag and an arcade.
Instead of responding: Give me more creative ideas.
Try the prompt: I don't want trampoline parks, bowling, arcades, laser tag or other typical birthday-party venues. I'm looking for something the kids probably haven't done at another birthday party this year.
Suddenly, "creative" has boundaries and it's far more useful to Claude. The same technique works for recommendations, travel plans, recipes, gift ideas, schedules and almost anything else where personal taste matters.
Try it: When Claude gives you suggestions you dislike, tell it specifically what those options have in common and ask it to avoid that entire category.
5. Stop putting everything into Projects
Projects are one of Claude's most useful features. They let you keep related chats, instructions and reference material together so you don't have to repeatedly explain what you're working on.
But that doesn't mean every task belongs inside one. Hassid argues that fresh conversations can sometimes be better for creative work because accumulated context can steer Claude toward ideas and patterns you've already explored.
I'd treat this as a rule of thumb not something universal for Claude. Projects can be enormously useful when the model genuinely needs background information.
The question is whether that background is helping with the task you're doing right now.
If you're using Claude to help plan a major home renovation, for example, keeping your measurements, budget, preferences and previous decisions in one Project makes sense.
But if you suddenly want Claude to give you completely unexpected ideas for what to do with an empty room, you might get more variety by asking in a fresh chat without all those previous assumptions.
Otherwise, you risk asking Claude to surprise you while simultaneously surrounding it with everything you've already considered.
Try it: Use Projects when Claude needs context. Try a fresh chat when you want novelty. You can even give the same prompt to both and compare the results.
The takeaway
Sometimes the best thing you can give Claude is a clearer problem. So the next time Claude isn't giving you what you want, don't automatically add another paragraph of instructions. Try editing the prompt, start a new chat or tellit what you don't want.
You may just find Claude understands you better when you give it less to untangle.
Follow Tom's Guide on Google News and add us as a preferred source to get our up-to-date news, analysis, and reviews in your feeds. Subscribe to Tom's Guide on YouTube and follow us on TikTok.
More from Tom's Guide
