7 Prompting Habits That Actually Help You Get Better AI Answers

A lot of AI prompting advice is wrong, or at least misguided.

You have probably seen it:

Start every prompt with an introductory preamble:

“You are a world-class expert...”
“I'm new to AI...”
“Act as a genius strategist...”
“Think step by step...”

A lot of the time, better results come from something much less theatrical: being clear, giving the right context, and not making the model guess what you want.

In a March 2026 Daily Tech News Show segment, the hosts suggested two possible prompting approaches: asking the chatbot how a beginner should use it within the prompt, and telling it to ask clarifying questions before answering. That second approach can sometimes be useful, but it is not always necessary. Many weak AI answers are not really the model “failing.” They are the result of vague input.

Here are the prompting habits that actually tend to help.

1. Stop writing prompts like you are casting a spell

Most of the time, you do not need prompt theater.

Try it yourself and look at the results. You usually do not need to preface prompts with an introductory role assignment or persona just to influence the response. You also do not need to invent a fake persona for yourself, as was suggested in the podcast segment. If what you want is a strong answer, the better move is to provide any relevant context and then state the task clearly.

That is cleaner, faster, and often leads to better results.

2. Do not default to asking it to ask questions first

This was another somewhat mystifying suggestion.

If your request has several moving parts, you can say:

“Before answering, ask me up to five clarifying questions.”

Sometimes that helps. But it can also work against the model's ability to answer efficiently by splitting the task unnecessarily instead of keeping the instruction focused.

It is often better to stay with the matter at hand and make your original request clearer from the start.

3. Give the job, the context, and the goal

A good prompt usually does three things:

• states the task
• gives the relevant context
• optionally defines what a good result should look like

That is not flowery, but it is much more likely to produce something useful.

4. Start simple before adding examples

A lot of people want to jump straight into elaborate prompting.

Often, the better move is to try zero-shot first. It may sound technical, but it simply means giving a clean, direct instruction and seeing how well the model responds. Usually, you only need examples if the task is more complex or if you need the output to match a very specific structure, tone, or pattern.

Instead of writing a prompt-turned-thesis packed with examples, multiple personas, and a style guide, try a clearer and more pointed version of the request first. If that is not enough, then add a strong example that closely matches what you want.

5. Stop forcing “think step by step”

This is a somewhat outdated habit, especially for reasoning models, as reflected in OpenAI's own best-practices guidance (linked below).

If you actually want the reasoning included, what usually matters more is simply asking for it directly.

Try it yourself. The results are often better than what older prompting habits would suggest.

6. If format matters, say what you want

Sometimes your prompt produces an answer with formatting you do not want.

The remedy is simple: ask for the format you do want, whether that is a numbered list, bullet points, a short blurb, or a seven-paragraph response.

Leaning on both OpenAI and Anthropic documentation, a useful general rule is to state what you want rather than focus on what the model should avoid. Experimentation still matters, of course. Like any rule, there are exceptions. What matters most is the result.

In general, it is better to give the model a target than an avoidance rule.

7. Treat the chatbot more like a collaborator than a vending machine

One of the best ways to use an LLM is not to fire off one prompt and hope for the best, but to use it iteratively when the first answer is not quite right.

That usually means asking useful follow-up questions or giving specific refinement instructions.

The best advice here is simple: read what was produced, then make a specific request for changes, clarification, expansion, or omission.

The fact that publicly available LLMs allow for this kind of iterative development is one of their greatest strengths. You can gradually shape the result into exactly what you want.

What actually matters most

There is no one-size-fits-all solution when it comes to prompting and performance, but some approaches are clearly better than others.

The broad takeaway is to abandon overly flowery or overly complicated prompt-writing habits in favor of simplicity and directness:

• be clear
• give context
• ask for clarification only when needed
• start simple
• specify format and constraints
• add examples only when they are actually necessary

That may be less exciting than a lot of online prompting advice, but for people who care more about results than ritual, these habits should become standard.

Source note

Examples and discussion of clarifying-question prompting were drawn from Daily Tech News Show, "Your Hate for Thin Phones Saved the S-Pen - DTNS 5231," March 23, 2026.

The broader framing also aligns with published guidance from OpenAI and Anthropic on prompt engineering and reasoning best practices, especially around simplicity, directness, output constraints, formatting control, and when to use examples.