Are You Still Using Claude the Old Way?

You’ve been asking Claude questions, writing prompts, and hoping for better answers. A few changes in how you give instructions could make a bigger difference than you think.

You open Claude, type a question, and wait for the answer. A few seconds later, you get something that sounds impressive but isn’t quite what you wanted.

Maybe the email is too formal. Maybe the research summary misses the most important detail. Maybe the answer is so long that you spend more time editing it than you would have spent writing it yourself.

So you try again. You ask Claude to make it shorter, sound more natural, or include the information it left out. Eventually, after several attempts, you get something useful.

If that sounds familiar, the problem might not be Claude. It might be how you’re asking.

AI assistants have changed considerably over the past few years, but many people still use them much the same way they did when conversational AI first became popular. They type a simple request, expect the system to understand everything they have in mind, and then spend the next several minutes correcting the result.

Anthropic, the company behind Claude, has published guidance on getting better responses from its AI models. Some of the advice is surprisingly ordinary. You don’t need complicated commands or programming experience. You need to give Claude the kind of information you would give another person who was helping you with a task.

Here are five changes worth trying.

1. Stop Making Claude Guess What You Want

Imagine asking a coworker to write an important email without explaining who it’s for, what happened, or what you want the person to do after reading it.

You probably wouldn’t expect a great first draft.

Yet people regularly give Claude instructions such as “Write a professional email” or “Make this better” and assume it will figure out the rest.

Claude can produce an answer to those requests, but it has to make decisions on your behalf. Sometimes those decisions are exactly what you wanted. Other times, they’re completely off.

A better approach is to explain the result you’re looking for.

Instead of typing, “Write an email to my manager,” try something more specific:

“Write a short email to my manager explaining that the project is finished, two issues were corrected, and the updated file is ready for review. Keep it friendly and professional. Use three short paragraphs.”

The difference isn’t that the second instruction sounds more technical. It’s that Claude now knows what information matters, who will read it, and how the response should look.

Anthropic recommends giving clear, direct instructions rather than assuming Claude will understand unstated expectations.

Before sending your next prompt, ask yourself a simple question: Could another person complete this task correctly using only the information I’ve provided?

If the answer is no, Claude might need more information, too.

2. Explain Why You’re Asking

There’s a difference between asking Claude to summarize a document and explaining that the summary will be used by someone who has only two minutes to review it.

The first request tells Claude what to do. The second tells it what matters.

Say you’re reviewing a long report about a company’s expenses.

You could ask, “Summarize this report.”

Or you could explain, “I’m preparing for a meeting with a small-business owner. Summarize the main expenses, identify anything that increased significantly, and explain the findings in plain English. The owner doesn’t have a financial background.”

Now Claude has a clearer picture of the audience and the reason behind the request.

That context can influence what the assistant emphasizes, how it explains unfamiliar terms, and how much detail it includes.

The same idea works outside business.

Someone planning meals for a family might explain how many people they’re feeding, how much they can spend, and how much cooking time they have. A student might mention their grade level and which part of a subject they find confusing.

You don’t need to tell Claude your entire life story. Just give it the details that could change the answer.

3. Show Claude What a Good Answer Looks Like

Sometimes you know exactly how you want something written, but describing it is difficult.

You might want an article that sounds like a newspaper story rather than a school assignment. Or a customer message that feels warm without sounding overly familiar.

You can spend several paragraphs explaining the style. Another option is to show Claude a short example.

Imagine you’ve written a customer announcement that sounds just right. The next time you need something similar, provide that announcement and say:

“Use this as a reference for the tone and length. Write a new announcement about our updated business hours. Keep the wording original.”

This gives Claude something concrete to work from.

Anthropic’s guidance recommends using relevant examples when you need a particular format, tone, or structure. Examples are especially useful when your preferences are difficult to explain in a few words.

But don’t assume that more examples always mean better results.

Start with one good example. If Claude still struggles to match the format or style, add another that shows what you’re looking for.

And remember that an example is a guide, not something to copy word for word. Review the final result to make sure it reflects your own information and isn’t reproducing someone else’s writing.

4. Tell Claude What to Do, Not Just What to Avoid

Think about the last time you tried to correct an AI-generated response.

Maybe you typed, “Don’t make it too long.”

Then Claude shortened it, but left out something important.

So you added, “Don’t remove the details.”

Then it gave you another version that was still too formal.

It’s easy to end up with a long collection of instructions about everything you don’t want.

Anthropic suggests that people can often get better control over a response by describing the desired format directly.

For example, instead of saying, “Don’t use bullet points,” you could write, “Explain this in natural paragraphs, using simple sentences and a conversational tone.”

Instead of “Don’t make it complicated,” try “Explain this to someone who has never used the software before. Use familiar words and give one practical example.”

The goal is to give Claude a clear direction.

That doesn’t mean negative instructions are useless. If something must be excluded, say so. But it also helps to explain what should take its place.

This is particularly important when writing articles, reports, presentations, and customer communications. A few words about the desired result can prevent several rounds of corrections.

5. Give Claude Permission to Say It Doesn’t Know

Here’s one habit that deserves more attention.

People often ask AI questions as though a confident answer is the only acceptable outcome.

They ask for statistics, research findings, company information, or explanations of events and expect the assistant to provide an answer immediately.

But Claude can make mistakes. Like other AI assistants, it can sometimes produce information that sounds convincing even when the details are inaccurate.

That’s especially important when you’re using AI for research.

Imagine asking Claude to identify businesses that recently opened in your city. It might provide a useful starting point, but you still need to check whether those businesses exist, are currently operating, and actually opened when the assistant claims they did.

One way to make your expectations clearer is to add an instruction such as:

“Use reliable sources for factual claims. If you cannot verify something, clearly identify it as unconfirmed instead of guessing.”

Anthropic recommends explicitly allowing Claude to express uncertainty when the available information isn’t enough to support an answer.

That instruction won’t eliminate mistakes, and it doesn’t replace checking the sources yourself. But it gives the assistant permission to acknowledge gaps rather than treating every question as something that requires a definite answer.

For anyone using AI to prepare research, business reports, or information for publication, that distinction matters.

The Best Prompt Isn’t Always the Longest

There’s a common assumption that getting better results from AI requires learning increasingly complicated prompting techniques.

Sometimes advanced methods are useful. If you’re working with large collections of documents, complicated instructions, or software development projects, additional structure can help Claude keep track of the task.

But ordinary users don’t need to turn every question into a technical assignment.

The bigger improvement may come from something much simpler: being clear about what you need, explaining why it matters, showing an example when necessary, and allowing the assistant to acknowledge what it doesn’t know.

You can even test the difference yourself.

Choose a task you’ve already given Claude. Try it again with a clearer description of the audience, the expected result, and the information that must be included.

Then compare the responses.

Does the new answer require fewer corrections? Does it sound more like what you wanted? Did Claude include the details that matter?

Those questions are more useful than counting how many words you managed to fit into the prompt.

Claude is capable of helping with much more than answering simple questions. But giving it a better assignment is still your responsibility.

And sometimes the most useful improvement isn’t learning a new AI trick.

It’s learning how to ask for exactly what you need.