In short
To get real work done with an AI assistant, hand it work the way you would to a capable new colleague: give it the context, the goal and examples of good output, ask for its reasoning on complex tasks, check what comes back, and keep the prompts that work.
Most people’s first weeks with ChatGPT, Claude or Gemini follow a similar path. A few answers are impressive, many are generic, and some are confidently wrong. After a while the assistant settles into a narrow role, usually rewording emails or summarizing documents, and the bigger promise of saving hours a week fades into the background.
In most cases the difference between those two experiences lies in how the work is handed over. The same assistant that writes a bland paragraph from a one-line request will produce a usable first draft when it gets the context a human colleague would need. Prompt engineering is the name for learning to hand work over well, and for everyday use it has much more to do with clear briefing than with technical tricks.
This guide covers the main ideas, with a longer article on each one.
What prompt engineering means for everyday work
The term came from developers who build products on top of language models, where small changes to instructions can change results across thousands of requests. For someone using an assistant at work, it means something simpler: writing requests that give the model what it needs to do the job properly, and checking what comes back.
Both of the leading AI companies describe it in similar terms. Anthropic’s guidance for Claude suggests thinking of the assistant as “a brilliant but new employee who lacks context on your norms and workflows”, and proposes a simple test: show your prompt to a colleague with little context on the task, and if they would be confused, the model will be too (Anthropic). Once you see prompting as delegation, most of the advice follows naturally.
Brief the assistant properly
A request like “write a project update” leaves the assistant to guess the audience, the length, the tone, what happened in the project and what the reader needs to decide. It fills those gaps with the most generic options available, and the result reads like it.
A good brief covers what a new colleague would ask about before starting: what the task is for, who will read the result, what material to work from, what a good result looks like and any constraints on length, format or tone. Explaining why also helps. Anthropic’s documentation gives the example of an instruction never to use ellipses, which works better when the prompt adds that the text will be read aloud by a speech engine that cannot pronounce them, because the model can then apply the reason to similar cases.
We go through a full briefing structure, with before and after examples, in Brief AI Like a Colleague, Not a Search Box.
Show it what good looks like
Some requirements take a paragraph to describe and one example to show. A house style, the structure of a weekly report, the way your team labels support tickets: a short example often does more than a paragraph of instructions.
Giving the model a few examples of input and output is known as few-shot prompting. It is one of the most dependable ways to control format and tone, and it is especially useful for repetitive tasks where you want every result to follow the same pattern. The examples need to be chosen with some care, though, because the model will copy their quirks along with their strengths. Show, Don’t Tell: Examples That Shape the Answer covers how many to use and how to pick them.
Ask for reasoning when the task needs it
For problems with several steps, such as working out a budget, comparing options against criteria or analyzing a set of figures, asking the model to work through the problem before answering can improve accuracy. This approach, known as chain-of-thought prompting, was popularized by Google researchers in 2022.
Newer models have changed how much you need to ask for it. Many current assistants include a reasoning mode that thinks through problems on its own, and OpenAI’s guidance for its reasoning models advises against adding unnecessary step-by-step instructions to them (AI News, on OpenAI’s guidance). Knowing which situation you are in saves time and gives better answers. Ask It to Think Step by Step, and When That Helps explains the difference.
Check before you trust
Language models produce fluent text whether or not it is correct, and they can state invented facts, figures and sources with the same confidence as real ones. A good brief lowers the risk without removing it.
A few habits cover most of the danger. Check every figure, name, date and quote that matters against a source you trust. Ask the assistant to separate what came from your material from what it added. Be most careful in the areas where you know least, because those are the areas where you will be least able to spot a mistake. The more a document matters, the more of it you should verify yourself.
Build your own library
Once a prompt works well for a recurring task, save it. Over a few weeks you will collect briefs for the reports, emails, summaries and analyses you do regularly, and each one can be reused and improved instead of being written from scratch. Many assistants now let you save instructions as projects, custom assistants or templates, which makes this easier.
A prompt library also makes it easier to share what works with your team, and to spot when a change in the tool changes the results.
Download the briefing checklist
The seven things a good AI brief covers, on one page to keep next to your screen.
Free. We will also send you our newsletter about new courses and guides. Unsubscribe in one click.
Learn it properly
The free Prompt Engineering course on Train Your Skills covers these ideas in six short lessons, from how assistants read a prompt to building your own prompt library, with a test at the end of each lesson. You can start the course here or browse all the free courses.
Frequently asked questions
What is prompt engineering?
Prompt engineering is the practice of writing instructions that get reliable, useful results from an AI model. For everyday work it mostly means briefing the assistant clearly, giving it the right material and checking the output.
Is prompt engineering still worth learning now that models are smarter?
Yes, although the emphasis has shifted. Newer models need fewer tricks and handle vague requests better, but they still cannot know your audience, your standards or your data unless you tell them. The briefing part of prompting has become more important as the technical part has become less so.
Do the same prompts work in ChatGPT, Claude and Gemini?
A well-written brief works across all of them, because clear context and instructions help any model. Small differences appear in formatting preferences and in how each one handles long documents, so it is worth testing an important prompt in the tool you actually use.
Should I write prompts in English if English is not my first language?
Most leading assistants work well in many languages, and writing in the language you know best usually gives a clearer brief. If the output must be in English, say so in the prompt. For specialized technical work, results in English can be slightly more reliable, so test both if accuracy is critical.
Is it safe to paste work documents into an AI assistant?
That depends on your company’s policy and on the plan you use. Business and enterprise plans from the main providers usually exclude your data from model training, while free consumer plans may not. Check your company’s rules before pasting anything confidential.
Sources
- Anthropic, “Be clear and direct”, Claude documentation. platform.claude.com
- AI News, “OpenAI shares best practices for prompting o-series reasoning models”. ainews.com