Most disappointing AI answers have the same cause: a vague request. Ask “write me a marketing email” and you will get something generic, because the assistant has to guess who you are, who the email is for and what you want it to achieve. The fix is not a secret formula. It is the same thing that makes a good brief for a human colleague.
Updated September 2026: we added links to OpenAI’s and Google’s prompting guidance and to the research showing that models learn from examples in the prompt.
Here are seven habits that work across ChatGPT, Claude, Gemini and most other assistants. The AI companies’ own advice points the same way: Google’s tips for Gemini organize instructions around persona, task, context and format.
1. Say who it is for and why
Context changes everything. Compare these two requests:
- “Explain inflation.”
- “Explain inflation to a 15-year-old who is saving for a car, in about 150 words, with one everyday example.”
The second prompt tells the assistant the audience, the purpose, the length and the style. It will almost always produce a more useful answer. OpenAI’s prompting advice puts it plainly: be "specific, descriptive and as detailed as possible about the desired context, outcome, length, format, style."
2. Define what a good result looks like
Before you type, ask yourself how you will judge the answer. Do you want three options or one recommendation? A table or a paragraph? A draft you will edit, or something ready to send? Put that in the prompt.
3. Give it the material
Assistants are much better at working with information you provide than at recalling details from memory. If you want a summary of a report, paste or upload the report. If you want an email in your voice, include a couple of emails you have written before. This also cuts down on made-up details, one of the most common failures of AI assistants.
4. Show an example
If you have a specific format in mind, show it. One example of the output you want is often worth a paragraph of description. The 2020 paper introducing GPT-3 showed that a model could pick up a new task from a few examples written into the prompt, with no retraining. You can also show what you do not want: “Avoid the tone of this sample, which feels too salesy.”
5. Break big jobs into steps
For complex tasks, ask for a plan first, review it, then ask for each part. Writing a report? Start with an outline. Analyzing a spreadsheet? Ask what questions the data can answer before asking for conclusions. You stay in control, and errors are easier to spot early.
6. Ask it to check itself
A simple follow-up such as “What might be wrong or missing in this answer?” or “Which of these claims should I verify?” often surfaces weaknesses. It is not a guarantee of accuracy, but it is a quick way to find the parts that deserve a second look.
7. Iterate instead of starting over
The first answer is a draft. Tell the assistant specifically what to change: “Shorter, drop the second point, make the opening more direct.” Conversations build on context, so a few rounds of focused feedback usually beat rewriting the prompt from scratch.
A template you can reuse
Prompt template I am [who you are] working on [task]. The audience is [who will read or use it]. My goal is [outcome]. Here is the relevant material: [paste or attach]. Please produce [format and length]. Keep the tone [tone]. Before you start, ask me up to three questions if anything important is unclear.
That last line is underrated. Inviting the assistant to ask questions often reveals assumptions you did not realize you were making.
What not to worry about
You do not need special keywords, elaborate role-play setups or pages of instructions for everyday tasks. Modern assistants respond well to plain, direct language. Save the detailed instructions for work you will repeat often, and consider saving them as a reusable project or custom assistant.
Brief it like a colleague
Treat the AI like a capable new colleague who knows nothing about your situation. Give it context, a clear goal and the right material, then review its work with a critical eye. That approach will serve you well no matter which assistant you use.
Sources
- Best practices for prompt engineering with the OpenAI API, OpenAI Help Center
- Tips for creating custom Gems, Gemini Apps Help
- Language Models are Few-Shot Learners, arXiv, 2020
