Prompt Writing Basics for ChatGPT, Claude and Gemini

You do not need secret "magic prompts" to get good results from AI assistants. The official guidance published by OpenAI, Anthropic and Google is strikingly similar: be clear and specific, give context, show what good output looks like, and refine through conversation. This guide turns that shared advice into a practical method that works across ChatGPT, Claude and Gemini, with examples you can adapt.

Why the same basics work everywhere

ChatGPT, Claude and Gemini are all large language models trained to follow instructions. They differ in strengths, interfaces and features, but they respond to the same fundamentals. Google's "Prompting guide 101" for Gemini in Workspace recommends four elements: persona, task, context and format. OpenAI's prompt engineering guidance stresses clear instructions, the desired outcome, audience, format and examples. Anthropic's documentation recommends being clear and direct, giving examples and separating parts of a prompt with tags. If you learn these basics once, you can use them in any assistant.

The four-part prompt

1. Role or perspective

Tell the assistant whose point of view to take. "You are an experienced accountant explaining to a first-time freelancer" produces different output from "You are a tax lawyer writing for other lawyers". The role sets vocabulary, depth and assumptions.

2. Task

State the action clearly with a verb: summarise, draft, compare, critique, translate, outline, rewrite. Ambiguous tasks ("help me with my presentation") produce generic answers.

3. Context

Provide the background the assistant cannot guess: who the audience is, what they already know, the goal, constraints, and any source material. This is usually the part people leave out, and it makes the biggest difference.

4. Format

Describe the output you want: length, structure (bullets, table, numbered steps), tone, reading level and anything to avoid.

Putting it together

Weak prompt: "Write about our new opening hours."

Better prompt: "You are the friendly owner of a small bakery in Amman. Write a short Instagram caption announcing that from 1 November we open at 7am instead of 8am on weekdays. Our customers are mainly office workers who grab breakfast on the way to work. Keep it under 50 words, warm but not cheesy, with one emoji at most and no hashtags."

The second prompt is longer, but it removes guesswork, so the first answer is much closer to usable.

Techniques that help in all three assistants

Give examples of what you want

If you have a style in mind, show it. Paste one or two examples of past posts or emails and say "match this tone and length". All three vendors' guidance recommends examples for consistent output.

Separate instructions from material

When you paste a document, make it obvious where your instructions end and the material begins. Anthropic suggests XML-style tags, for example putting the text between <report> and </report>. Simple headings such as "INSTRUCTIONS:" and "TEXT:" or triple quotes also work well in ChatGPT and Gemini.

Ask for structure

Request headings, numbered steps or a table when you need to scan the result. For decisions, ask for "pros, cons and a recommendation with reasons".

Break big tasks into steps

Instead of "write my business plan", work in stages: outline first, then each section, then a consistency review. You keep control and can correct direction early.

Ask it to ask you questions

For complex tasks, end with "Before you start, ask me up to five questions about anything that is unclear." This often surfaces missing context.

Iterate instead of restarting

Treat the first answer as a draft. Follow up with specific edits: "shorter", "more formal", "replace the second example with one about retail", "keep everything except the conclusion".

Prompts for checking facts and reasoning

Assistants can sound confident while being wrong. Prompts can reduce, but not eliminate, that risk:

  • "List any claims in your answer that I should verify, and how."
  • "If you are not sure, say so rather than guessing."
  • "Use only the document I provided; if the answer is not in it, say it is not there."
  • "Cite your sources with links" (in assistants with web search enabled), then open the links yourself.

Always verify statistics, quotes, legal or medical information, and anything you will publish.

Small differences worth knowing

The basics are shared, but each assistant has features that change how you work:

  • Saved instructions: ChatGPT offers custom instructions and projects; Claude offers projects and user preferences; Gemini offers "Gems", custom versions of Gemini with saved instructions, according to Google's plan page. Put your recurring context there instead of retyping it.
  • Files and sources: all three accept uploaded documents, with limits that vary by plan. Gemini Notebook (formerly NotebookLM, renamed by Google in July 2026) is built specifically around answering from your own sources.
  • Thinking or reasoning modes: paid tiers of each assistant offer deeper reasoning models. For hard analytical tasks, selecting them can matter more than prompt wording.
  • Research modes: deep research features run longer, multi-step web research and produce reports. Give them a precise question and scope.

Reusable prompt templates

Summarise a document

"Summarise the attached report for a busy manager who has not read it. Give five bullet points of key findings, then two risks and one recommended next step. Keep it under 200 words and quote figures exactly as they appear."

Improve your own writing

"Edit the text below for clarity and grammar. Keep my voice and meaning. Do not add new information. Show the edited version, then list the three most important changes you made and why."

Compare options

"Compare these three accounting tools for a two-person design studio: [names]. Use a table with price, ease of use, invoicing features and integrations. Mark anything you are unsure about, and tell me what to check on the vendors' websites."

Learn a topic

"Explain how compound interest works to someone with basic maths. Use one simple example with numbers, then give me three short questions to check my understanding. Wait for my answers before explaining the solutions."

Prompting in Arabic and other languages

All three assistants accept prompts in many languages, including Arabic. A few habits help when you work across languages:

  • Say which variety you want. Modern Standard Arabic and Jordanian or other dialects read very differently; specify the one your audience expects.
  • Specify the output language explicitly, especially if your prompt and source document are in different languages. For example: "The document is in English; answer in Arabic."
  • Keep key terms consistent. Provide a short glossary of brand names, product names and technical terms, and say which should not be translated.
  • Have a fluent reader check anything customer-facing, as tone and idioms are easy to get slightly wrong.

Common mistakes

  • Too vague: no audience, goal or format.
  • Too many tasks at once: five requests in one prompt often means one is ignored.
  • Hidden constraints: you know the word limit or banned topics; the assistant does not unless you say so.
  • Trusting without checking: fluent output is not the same as correct output.
  • Sharing what you should not: keep passwords, client secrets and personal data out of prompts unless your plan and policies allow it.

Bottom line

Good prompts are clear briefs: a role, a task, the context and the format you want, followed by focused iteration. That approach works in ChatGPT, Claude and Gemini alike, and it improves results more than any trick. Compare assistants in our best AI chatbots category, find specialist tools in AI writing, or browse the full AIMAPER directory.