Getting a useful result from a generative AI tool is a skill, and it comes down to how you ask and what information you provide to an AI tool. This page explains this skill, usually called prompting.

A prompt is simply the instruction or question you give the tool. Because these tools build their answer from the context you provide (see How does GenAI Work?), the golden rule is the same throughout this page: the more clearly and completely you describe what you want, the better the result.

1. Start with a clear, specific prompt

Vague prompts produce vague, generic answers, because the tool fills any gaps with whatever is most typical. Specific words also prevent misunderstandings (e.g. ask about "sampling bias" rather than just "bias," which could have many other meanings).

If your prompt includes a text to translate or summarise, keep a clear separation between your instruction and the text itself, so the tool doesn't confuse the two. A simple way to do this is to wrap the text in a marker such as quotation marks (“”) or triple dashes (--- ---).

2. Give it the context it needs

The tool only knows what it learned in training plus what you put in the prompt, so spell out anything that matters including the language, the source material and the context.

If you upload a file for the tool to work with, make sure it's well organized and clearly labelled (for example, what each column and unit in a spreadsheet means). Moreover, make sure that the model you are using is capable of interpreting the information you are providing (e.g. not all LLMs can read graphs in PDF documents).

3. Shape the length, tone, and format

Length: these tools are poor at counting, so an exact word count often won't land. Words like "concisely" or "extensively," or asking for a "top 10", tend to work better.

Tone: state the register you want (e.g. "in a formal academic tone," "in plain conversational language", or "neutral and objective"). Naming the reader also sets the tone, since the tool adjusts its language to the audience (e.g. "written for a first-year student").

Format: tell it how to structure the output (e.g. a bulleted list, a comparison table, headed sections, a step-by-step explanation. Being explicit here saves you reshaping the answer afterwards.

• Give an example. Showing the tool one or a few examples of what you want steers it better than describing the format. For instance: "The slogan for McDonald's is 'I'm Lovin' It'. What could a slogan for Wageningen University be?"
• Highlight what matters. Put your most important instructions at the very start and the very end of the prompt (both people and these tools pay most attention there), and use CAPITALS for words you want emphasised.
• Ask for step-by-step reasoning. For logic or multi-step problems, adding "explain your reasoning step by step" can improve the answer by making the tool work through it in stages.
• Ask it to check its own work. If the output isn't quite right, ask the tool to review it against your original request e.g. "check that your answer meets all the criteria above, and correct it if not."
• Refine in steps. Treat the first answer as a draft. A short back-and-forth ("shorter," "more formal," "add an example") almost always beats expecting a perfect answer first time.

A persona is a character or role you ask the tool to take on (e.g. a defined expertise, perspective, and tone) so that its answers come from a consistent point of view. Instead of a generic assistant, you get, say, a patient statistics tutor, a critical peer reviewer, or a debate opponent.

Because the tool builds its answers from the context you give it, describing a persona fills that context with the vocabulary, tone, and style associated with the role, which then shapes everything it says afterwards. It's worth being clear about what this does and doesn't do: a persona changes how the tool responds, but it doesn't actually give it real expertise or make its answers more accurate. A "professor" persona can still be confidently wrong, so keep checking the output as you normally would.

How to create one

Describe the persona at the start of your conversation, covering as many of these as are useful:

• Role and identity. Who the tool should be ("You are an experienced thesis supervisor").
• Background or expertise. The knowledge or field it should draw on. In many tools you can also upload files or documents (such as lecture notes or a paper) for it to work from.

• Tone and style — formal or casual, concise or detailed, encouraging or challenging.
• How it should behave — including what not to do ("ask me questions rather than giving the answer straight away").
• Audience (optional) — who the answer is for, if that matters.