All guides
How-to

Better Prompts for ChatGPT, Claude and Gemini

Get better results from ChatGPT, Claude and Gemini with a simple prompt framework, practical writing examples and fixes for common chatbot mistakes.

Key takeaways

  • Give your chatbot a clear task, relevant context and a specific output format.
  • Define what a good answer looks like: audience, tone, length and factual boundaries.
  • Improve results through focused revisions, and verify important claims before using them.

A better prompt does not need fancy language or a secret formula. It needs enough information for the chatbot to understand what you want—and what you do not want.

The same core approach works across ChatGPT, Claude and Gemini. Whether you are writing an email, summarising a document or planning a project, start with a clear brief and refine the result.

What you need

  • Access to a chatbot.
  • A task with a clear purpose.
  • Relevant background information or source material.
  • A few minutes to review and revise the answer.

Use only information you are authorised to share. Remove sensitive personal details, confidential client information and credentials unless your organisation explicitly permits their use in the tool.

You do not need a paid plan to practise these techniques. Available features and usage limits vary, so build your prompt around what your current chatbot supports.

Step by step

1. Start with one clear task

Say exactly what you want the chatbot to do. Useful verbs include *draft*, *summarise*, *compare*, *explain* and *rewrite*.

A vague prompt leaves too many decisions open:

Help me with customer emails.

A clearer version defines the deliverable:

Draft an email telling a customer that their delivery will arrive two days later than expected.

For a complicated project, separate planning, drafting and editing. Asking for everything at once can make it harder to spot where the answer went off track.

2. Add the context that changes the answer

Include the audience, purpose and relevant circumstances. Focus on details that would change what a capable colleague produces.

For the delivery email, add:

The customer needs the order for an upcoming event. We want to acknowledge the inconvenience and explain the revised delivery timing. We cannot offer a refund or discount in this email.

Avoid dumping unrelated background into the conversation. More text is not automatically better context.

When supplying source material, separate it from your instructions with clear labels such as `Task:` and `Source material:`. Explain whether the chatbot should use only that material or may draw on general knowledge.

3. Specify the output format

Tell the chatbot what the finished answer should look like. This reduces the editing needed afterwards.

For example:

Return one subject line and an email body under 150 words. Use three short paragraphs. Do not include commentary before or after the draft.

Depending on your task, request:

  • A checklist for completing a process.
  • Bullet points for a meeting brief.
  • A numbered plan with dependencies.
  • A short explanation followed by an example.

Choose a format that matches how you will use the answer—not just one that looks tidy.

4. Define quality with concrete instructions

Words such as “professional” and “engaging” are open to interpretation. Translate them into observable writing choices.

Instead of:

Make it sound professional.

Try:

Use plain English, short sentences and a calm, respectful tone. Avoid corporate jargon, exaggerated apologies and sales language.

You can also supply a short example of the style you want. Explain which qualities to copy, rather than expecting the chatbot to infer them.

For factual tasks, add boundaries:

Do not invent dates, policies or promises. If essential information is missing, ask for it before drafting.

These instructions can reduce unsupported additions, but they do not guarantee accuracy.

5. Combine the pieces into a reusable prompt

Use this template when starting a new task:

Task: [What you want done]
Audience and purpose: [Who it is for and what it should achieve]
Context: [Relevant facts and source material]
Output: [Format and approximate length]
Style: [Tone and language preferences]
Boundaries: [What to avoid or leave unchanged]
Missing information: Ask up to three questions if essential details are missing. Otherwise, proceed and label any assumptions.

Not every request needs every field. A simple rewrite might need only the task, source text and tone.

Treat the template as a checklist, not a ritual. The goal is clarity, not a longer prompt.

6. Revise with specific feedback

Treat the first answer as a draft. Identify the gap between what you requested and what you received.

Weak feedback:

Try again. Make it better.

Useful feedback:

Keep the subject line and delivery information. Shorten the opening, remove the repeated apology and make the final sentence explain the next step.

Change one or two things at a time when possible. This makes it easier to see which instruction helped.

If the chatbot repeatedly misses the point, restate the task and constraints in a fresh conversation. Include any useful wording from the previous draft.

7. Check facts and fitness for purpose

Before using the answer, review it against your original brief:

  • Does it address the intended audience?
  • Is the format usable?
  • Are names, dates and figures correct?
  • Has it added unsupported claims or commitments?
  • Does the writing sound appropriate for your organisation?

For summaries, compare the answer with the source. For research, verify important claims using reliable sources rather than assuming a confident answer is correct.

You can ask the chatbot to flag possible problems, but its own review is not independent verification.

Common mistakes

Using role-play instead of a brief. “Act as an expert marketer” does not explain the audience, offer or goal. A role can help set perspective, but it cannot replace context.

Giving conflicting instructions. Asking for a comprehensive explanation in two sentences creates a trade-off. State which requirement matters most.

Assuming the chatbot has current information. Browsing and file access depend on the tool and configuration. Provide source material when it matters.

Overloading the prompt. Remove repeated rules and irrelevant background. Keep the task easy to identify.

Accepting polished writing as proof. Fluent prose can still contain errors, omissions or invented details.

Tools that help

  • ChatGPT: Useful for drafting, rewriting, analysis and everyday questions. Where available, file and image support can help you provide context.
  • Claude: Useful for long-context writing and reasoning tasks, including working through substantial source material.
  • [ProWritingAid](/tools/prowritingaid): Helps with editing, style checks and readability after you have a draft. Review its suggestions rather than accepting every change.

Start with one chatbot and a consistent process. Switching tools is less useful than learning to diagnose what your prompt is missing.

FAQ

Do ChatGPT, Claude and Gemini need different prompts?

The basic framework works across all three. Adapt instructions to available features and the responses you receive. There is no universal wording that guarantees the same result everywhere.

How long should a prompt be?

Long enough to communicate the task and essential constraints. Simple requests may need one sentence; complex writing projects may need a structured brief and source material. Remove anything that does not help.

Should I ask the chatbot to ask me questions first?

Yes, when missing information could materially change the answer. Limit the questions to essential gaps. For straightforward tasks, a clear initial brief is usually quicker.

Share

Comments & feedback(0)

Was this guide helpful? Ask a question, share a tip, or point out something we should fix.

Sign in to leave a comment or feedback on this guide.

Loading…