This guide helps you choose the right AI tool for different situations, write clear prompts for AI, and use the conversation context effectively. The goal is to provide practical tips on how to get more value from AI more cost-effectively and with less effort: faster responses, better quality, and less unnecessary back-and-forth correction.

You can read more about AI and its use at the University of Oulu in the guide Artificial Intelligence and Its Use at the University of Oulu.

Key instructions in brief

• Choose the right tool for the task: not all AI solutions are suitable for everything, so it is worth selecting the tool based on the task.
• State clearly what you want: a good prompt explains the goal, the necessary background information, and the desired output.
• Keep requests and prompts concise: a clear and well-defined request usually produces a better answer and may also reduce costs in paid services.
• Start a new chat when the topic changes: a new conversation often helps produce more accurate and relevant answers.
• Proceed step by step: first ask for a draft, then refine it, and finish by polishing the result.

What should different AI systems be used for?

Choose the AI tool based on the task. The same tools are not suitable or optimal for every purpose, so the best results usually come from using the solution that fits each need.

What kinds of AI tools are generally available?

• General-purpose chat tools: for brainstorming, drafting, summarizing, translation, and general problem-solving.
• Tools specialized for coding: for programming, debugging, refactoring, and outlining technical solutions.
• Tools that use search or organizational information: for using your own documents, emails, meetings, and other internal materials.
• Agent tools: for planning and carrying out multi-step tasks, such as information retrieval, editing documents, coding, or automating workflows. Agents can often use other tools on the user’s behalf, so using them requires particular attention to permissions, data protection, and checking the final result.
• Image or other media generation tools: for visualization, drafts, concepts, and developing ideas for materials.

You can find up-to-date information on tools permitted at the University of Oulu in the List of AI Tools Maintained by ICT Services. At the same time, check whether the tool is suitable for the data you are processing (e.g., personal data, confidential materials, or other sensitive information). You can find more guidance on information classification in the guide Data Classification at the University of Oulu and Oamk.

How to write a good prompt?

A prompt means the input you give to AI. It can be a question, assignment, request, instruction, or background information you include. The AI’s response is built based on what you ask, how clearly you define the task, and what kind of context you provide.

A good prompt is clear and well-defined. Explain what you want, what the answer will be used for, who it is intended for, and what format you want the output in. The more precise the request, the more useful the answer usually is.

1. Goal: state as clearly as possible what you want the AI to do. Prefer a direct verb that describes the action you want, such as summarize, compare, suggest, write, or correct.
2. Context: add the necessary background information, limitations, and any other relevant information the AI needs to understand the task.
3. Audience and style: explain who the content is for and what tone it should use, such as professional, plain-language, or technical.
4. Response format: if needed, define the desired structure, length, or format of the answer, such as a list, table, email, summary, or step-by-step instruction.
5. Possible constraints: explain what the output should emphasize, what it should avoid, and what rules the answer should follow.
6. Examples if needed: if you want a specific structure, style, or response model, provide one or two examples that the AI can follow.
7. Iteration: in complex tasks, the best result often emerges step by step. For larger problems, you may get the best outcome by first asking for an outline of the issue and then asking the AI to focus more closely on specific details in one part at a time.

Example of a weak and a better prompt:

• Weak: “Write a summary of this.”
• Better: “Write a clear summary of about 150 words on the topic below for students on the course. Assume that the reader is encountering the topic for the first time, so focus on the essentials, use understandable general language, and avoid unnecessary specialist terminology. Finally, highlight the three most important things the student should remember.”

Aim to keep the prompt clear and concise. In many paid AI services, the price is based on the number of tokens used. A token is a small unit of text that the language models behind AI applications use when processing and generating text and content. The more unnecessary background information, repetition, or long answers accumulate, the more tokens are used. A clear and well-defined request often produces a better answer and helps avoid unnecessary costs in paid services.

Why should you open a new chat with AI more often?

AI often takes previous messages in the same conversation into account when generating responses. This is useful if the existing conversation supports the current task, but a history that is too long or unclear can reduce the quality of the answers. It is worth starting a new chat when the old conversation begins to steer the answer in the wrong direction.

This is related to the language model’s context window. It means the limited amount of text that the AI can take into account at one time. The same context window includes the user’s messages, the AI’s previous answers, and any other background information provided. As the conversation gets longer, older or irrelevant content may start to reduce the quality of answers or fall outside the model’s attention entirely. For this reason, it is worth starting a new chat when the topic changes, or the conversation history no longer supports the current task.

Open a new chat when:

• you move to a completely new topic or task
• the previous conversation has become long, and the answers are starting to get worse
• the AI starts repeating itself or incorrectly using old assumptions
• you want to start with a clean slate without the influence of the earlier conversation
• you want to try the same task in a new way or with new instructions

If you wish, before opening a new conversation you can ask for a summary of the current conversation, situation, decisions made, and possible next steps. This lets you transfer the essential context to a new conversation without unnecessary background information.

How to choose a language model suitable for the use case?

A language model, or LLM, means a large AI model trained on a large amount of text data that produces responses based on the prompt given by the user. A language model can, for example, answer questions, explain concepts, write and edit text, summarize content, or help with problem-solving. However, different language models may differ in areas such as speed, reasoning ability, context window size, specialization, and usage restrictions.

In many AI tools, users do not necessarily need to choose the language model themselves. If several language models are available, it is worth choosing the model based on the task, the required speed, and the material being processed. The most important factor is not the size or trendiness of the model, but whether it suits what you are doing and how much background information it needs to process.

When choosing a language model, it is best to consider what kind of task you are working on:

• A fast general-purpose model is well suited to everyday tasks such as brainstorming, drafting, creating lists, summarizing text, and producing initial suggestions for larger systems.
• A stronger reasoning model is a better choice when the task requires more detailed analysis, comparison of alternatives, planning, or multi-step problem-solving.
• A model that supports long context is useful when working with a long document, multiple sources, or an extensive conversation history whose content the language model needs to be able to use.

Also consider which service you use the language model in. The same language model may be available in different services or environments, but data protection, terms of use, and data processing practices may vary. Before entering, for example, your own files, meeting notes, internal information, personal data, or other confidential material into a service, make sure where such information may be processed. ICT Services maintain a List of AI Tools Maintained by ICT Services and their permitted use cases. You can read more about information classification and the restrictions related to different categories in the Data Classification at the University of Oulu and Oamk -guide.

If you do not know which model you are using, that is usually not a problem. It is more important to formulate the request clearly, provide the necessary context, and check the result before using it.

Example prompts

• Email drafting: “Write a polite and professional email to a partner. In the message, request a brief meeting next week to review the progress of the project. Suggest three different time options and keep the tone open.”
• Editing text: “Edit the following text to make it clearer and better structured while preserving the factual content. Shorten sentences that are too long, make the wording more precise, and provide the result directly as a final text version without any extra explanations.”
• Brainstorming: “Suggest five ways to solve this research or teaching challenge. For each option, briefly describe the idea, benefits, risks, and most suitable use case.”
• Comparison: “Compare the options below to support decision-making. Assess the differences, strengths, weaknesses, and risks. Present the result in a table and add a brief recommendation at the end.”
• Planning: “Create a step-by-step action plan for achieving this goal. Divide the plan into first steps, implementation phase, and longer-term development. Add the most important tasks and risks.”

Summary

You get the most value from AI when you choose the right tool for the task, write a clear prompt, provide only the relevant context, and check the result before using it. If needed, start a new conversation or choose a language model that is better suited to the task.

« Back