AI Native Careers

Writing prompts for everyday work tasks

· By AI Native Careers editors

A prompt is just the request you type into an AI assistant. You don't need special syntax or tricks to write a good one. The published guidance from the companies that build these models, Anthropic and OpenAI among them, comes down to a few habits that also make you clearer when you brief a person.

Be clear and direct

Say exactly what you want. Vague requests get generic answers. Anthropic's prompting guide offers a useful test: show your prompt to a colleague who knows little about the task and ask them to follow it. If they would be confused, the model probably will be too.

Compare:

Summarise this.

with:

Summarise these meeting notes for my manager, who missed the meeting. Use five bullet points or fewer. Put decisions first, then open questions, then action items with an owner for each.

The second request says who it is for, what to include, how long it should be and how to order it.

Explain why, not just what

Context helps. Anthropic's guide notes that explaining the reason behind an instruction gives the model more information about your goal and leads to more targeted answers. "Keep it under 100 words because it goes in a mobile notification" works better than "keep it short", because the model can make sensible choices you didn't spell out.

Useful context to include:

  • Who will read it and what they already know.
  • What it is for: a decision, an update, a first draft for you to edit.
  • What good looks like, or what to avoid.
  • The source material, pasted in, if the task depends on it. OpenAI's guide likewise recommends including the relevant information in the prompt rather than relying on what the model happens to know.

Show an example

If you want a particular format or tone, show it. Both Anthropic's and OpenAI's guides describe giving a few examples of the input and the output you want, which they call few-shot prompting, as a reliable way to steer format, tone and structure. Anthropic suggests three to five varied examples for best results.

For everyday work, even one example helps: paste a past status update you were happy with and ask for this week's in the same style.

Separate the parts of a longer request

When a prompt mixes instructions, background and material to work on, mark where each part starts and ends. Headings, bullet lists, or simple labels such as "Instructions:", "Background:" and "Notes to summarise:" all work. Both guides recommend structuring longer prompts this way so the model doesn't confuse your instructions with the material.

Ask for a specific output

Name the format: a table with named columns, a numbered list, a two-paragraph email, three options to choose from. If you will paste the result somewhere, say where, so it arrives in a usable shape.

Let it say it doesn't know

Models can produce confident text that is wrong. Anthropic's guide on reducing these errors suggests explicitly allowing the model to say it doesn't know, and asking it to back each claim with a quote from the material you gave it. The same page is clear that these techniques reduce errors but don't remove them, so you still check anything that matters.

Iterate instead of starting over

If the first answer misses, say what to change: "shorter", "less formal", "you left out the budget question". Treat it like editing a draft with a colleague. When you find a prompt that works for a recurring task, save it and reuse it.

A template to adapt

I need [the output] for [the reader], who [what they know or need].

It's for [the purpose].

Here is the material: [paste it].

Format: [length, structure].

Here is an example of the style I want: [example].

If something isn't in the material, say so rather than guessing.

Sources