Interactive learning

Clear Context and Structured Questioning Techniques

Prompting in teaching practice

Author: Zsolt LengyeliTStudy Hungary Training and Research Centre
Reviewer: Mária HartyányiiTStudy Hungary Training and Research Centre

The building blocks of a request reach the AI model, which uses them to produce an organised answer. — A clear request helps produce a usable answer.

Learning objective

The quality of an AI answer depends, to a surprising extent, on the quality of the request. This material helps teachers give AI tools clear context and well-structured questions.

By the end, you will be able to

Name the seven building blocks of an effective educational prompt.
Turn a vague request into a structured prompt.
Clarify a task using examples, constraints and requests for step-by-step reasoning.
Improve results through deliberate iteration and adapt the output to learner level.

A good prompt starts with clear thinking

A prompt is the message we send to an artificial intelligence system – in practice, it is the only channel through which we can tell the model what we need. Today a prompt is not only text: we can attach images, audio files, PDFs, spreadsheets or other documents, and all of these become part of the input that the model “reads” before answering.

Prompting is not about memorising magic phrases. It is about thinking clearly and passing that clear thinking on to the model. A good prompt is not necessarily long, but it is unambiguous: it states the goal, provides the necessary background, describes the expected output and, when useful, shows an example. The teacher remains the “manager” of the process: the model can speed up the work enormously, but it cannot decide the aims, the priorities or the quality standards on our behalf.

Three types of prompts

It helps to be aware of what kind of request we are sending, because the model approaches each type differently:

01

Question

We expect an answer or an explanation.

“How does a residual-current device work?”

02

Instruction

We ask the model to create or change something.

“Write five multiple-choice questions on…”

03

Planning

We give an open task that requires a structure, plan or strategy.

“Plan a four-lesson unit on…”

State your expectation directly

Mixing these up weakens the result. “Could you write an e-mail to parents?” is phrased as a question, although we want an action. “Write an e-mail to parents about…” is clearer – and the clearer the expectation, the better the answer.

The seven building blocks of an educational prompt

When teachers use AI to prepare teaching materials, the following elements have the greatest influence on the relevance, precision and pedagogical usefulness of the response. Not every prompt needs all of them, but each missing element is something the model will have to guess.

Role, task, learner level and objective, context, format and evaluation criteria provide structured input for the AI answer. — The building blocks of a structured prompt: role, task, learner level and learning objective, context, output format and evaluation criteria

Example of a structured prompt

The same classroom task in seven short sentences: each sentence makes one building block explicit.

Role: Act as an experienced vocational teacher of electrical installation.
Task description: Create a worksheet about residual-current devices (RCDs).
Learner level: The learners are first-year apprentices, aged 15–16, with basic knowledge of circuits.
Learning objective: After the lesson they should be able to explain why an RCD trips and how to test it safely.
Context: We have 20 minutes and a demonstration board in the workshop.
Output format: Give 6 short questions and 2 scenario tasks, with an answer key, in a table.
Evaluation criteria: Check that every statement is technically correct and that safety rules are included.

From a vague request to a structured prompt

“What should I teach about RCDs?” – the model will answer, but it has to invent almost everything: the audience, the depth, the format and the purpose. The result is usually generic. When we add role, task and context in short, clear sentences, the model has much less to guess and the answer becomes directly usable.

What should I teach about RCDs?

Who is the audience? How much depth is needed? What format? What learning objective? The model has to guess all of these.

On the left, a vague request leads to a scattered result. On the right, a structured prompt leads to a focused answer. — A vague request leads to a generic, scattered answer; a structured prompt leads to a focused, usable result

Longer is not automatically better

Longer is not automatically better. It is not worth “talking the AI’s head off”: several short, focused sentences – or a task split into sub-tasks – work better than one long, convoluted paragraph. Compare these two instructions for the same task:

Convoluted request

Convoluted: “Look through the document thoroughly and pick out the most important data and information that could be relevant for a summary, then transform them into an easier-to-understand form, then create visual charts from them so that the data is even more illustrative, and finally shorten everything into clear bullet points.”

Clear request

Summarise the key data of the document.
Create charts from the summarised data.
List the most important information in bullet points.

One task, one clear sentence

Short, focused sentences and sub-tasks reduce the chance of misunderstanding.

Ask questions, show examples, add constraints

Start with questions

Effective prompting often begins before the actual instruction. What is the goal? What background information is missing? How will I decide whether the answer is good? A simple and powerful technique is to ask the model to ask us: “Before you start, ask me the questions you need to create a good worksheet.” This turns the prompt into a dialogue and reveals the gaps in our own request.

Show examples

A good example is often worth more than a long explanation. If we have a worksheet, quiz format or style we like, we can attach it and ask for “something similar”. One good and one bad example can increase precision dramatically, because the model can see the boundary of what we want.

Add constraints

Constraints narrow the space of possible answers: word limits, number of items, reading level, required terminology, “do not use technical abbreviations without explaining them”, or “use only the attached textbook chapter as a source”.

Ask for reasoning and neutral evaluation

Ask for reasoning, not only for the result

Language models need tokens to “think”. If we force a calculation or a multi-step decision into a one-word answer, errors become much more likely. For such tasks, ask the model to work step by step and to give the final answer at the end. Where precision matters – calculations, tables, data – ask it to use its tools, for example: “use code to calculate this” or “search the web and cite your sources”.

Ask neutral questions

The way a question is framed steers the answer. “What are the advantages and disadvantages of electric cars compared with conventional cars?” leads to a balanced answer. “Why are electric cars so much better than conventional cars?” leads to a one-sided list. AI systems tend to be very validating: they reflect and amplify the assumptions built into our questions. When we want an honest evaluation, we should phrase the question neutrally.

Let the AI write the prompt (meta-prompting)

We can also ask the model to help us formulate a better prompt: “Write a short, unambiguous prompt for ChatGPT that helps a beginner learn the basics of TypeScript. The prompt should ask for 1) five main topics, 2) one mini exercise for each, 3) a weekly study plan.” Reviewing the prompt the model produces is itself a good lesson in structured thinking.

Iteration: prompting is a dialogue

Prompting is rarely a single shot. A useful five-step routine – popularised, among others, by Google’s Prompting Essentials course – is: Task → Context → References → Evaluate → Iterate. Define the task, give the context, add references or examples, evaluate the answer against your criteria, and refine the prompt.

Task → Context → References → Evaluate → Iterate

The repeating cycle of writing, generating, evaluating and refining. — Write, generate, evaluate, refine – the iteration cycle of prompting

Model choice and prompt language

Two further practical points. First, the choice of model matters: a fast model is enough for everyday questions, while complex planning, analysis or calculations deserve a stronger “thinking” model. Second, the language of the prompt matters less than its clarity: current models work well in other languages (e.g. Hungarian), too, and a semantically muddled prompt is a much bigger problem than a few typos.

Adapting the output to the learners

One of the most valuable uses of structured prompting in vocational education is differentiation. By changing only the learner level and the objective in an otherwise identical prompt, the same topic can be prepared for different groups: a simplified explanation with everyday analogies, a standard worksheet, and an extension task for advanced learners.

For the same topic, change the learner level and the learning objective.

A teacher uses AI to prepare materials on the same topic at different levels for three learner groups. — The same structured prompt, varied only in learner level, produces differentiated materials

Learners needing support

A simplified explanation with everyday analogies.

Learners at the standard level

A standard worksheet adapted to the group’s prior knowledge.

Advanced learners

An extension task with more independence and a more complex objective.

From a one-line request to a structured prompt

Teacher exercise · 20–25 minutes · Individual work

Experience how the building blocks of a prompt change the relevance and usability of an AI answer. Use a generative AI tool.

Exercise steps

Choose a topic you will teach in the coming weeks. Write a one-line prompt, e.g. “Make a worksheet about welding safety.” Save the answer.
Open a new chat. Rewrite the request using the seven building blocks and save the new answer.
Compare the two answers: relevance to your learners, correct level, technical accuracy, and how much editing would still be needed.
Iterate once: add a constraint or an earlier worksheet you like as an example.
Ask: “What information did you have to guess in my first prompt?” Compare its list with your building blocks.

Prompt detectives

Classroom work with students · 30–35 minutes · Pair work and whole-class discussion

Learners formulate clear, context-rich questions and critically evaluate AI answers. Tools: a school-approved AI tool and a projector.

Show a vague prompt related to the learners’ trade and its AI answer on the projector, e.g. “How do I fix a car?”
In pairs, learners list what is missing: who is asking, what exactly the task is, what the situation is, and what kind of answer is needed.
Each pair rewrites the prompt using at least four building blocks: role, task, context and format.
Try the improved prompts. Add: “Ask me clarifying questions before you answer.”
Discuss which prompt worked best and why. Look for errors or missing safety information in the answers.

What will learners learn?

A precise question is a professional skill. If we do not ask about protective equipment, AI may not mention it. Fluent text is not the same as correct text.

Teacher preparation

Follow the school’s rules and the providers’ age requirements. If learner accounts are not permitted, demonstrate from your own account.

Remind learners never to enter personal data into prompts: names, addresses or photos of classmates.

Takeaways for teachers

Be clear about what you want before you type: goal, audience, format, criteria.
Use the building blocks as a checklist; every missing element is a guess for the model.
Prefer short, clear sentences and sub-tasks to long, convoluted paragraphs.
Show examples and add constraints; ask the model to ask you clarifying questions.
Ask neutral questions when you want an honest evaluation.
Iterate: evaluate the answer against your criteria and refine the prompt.
Stay in control: check facts, safety rules and pedagogical appropriateness before classroom use.

Effective AI use depends on thoughtful task design, not only on knowledge of individual tools. The passive user types, accepts and moves on; the active user asks, checks, adds and iterates – and gets exponentially more value from the same tool.

Check your knowledge!

Answer every question, then review your result and the explanations.

1What is a prompt?Only a text question, without attachments.The message sent to AI, which may include attached images and documents.All the data used to train the model.

2Which building block states what learners should be able to do after the lesson?The role.The output format.The learning objective.

3Which request best helps produce a usable worksheet?“Create 6 short questions and 2 scenario tasks with an answer key for first-year learners, to be completed in 20 minutes.”“Write something about RCDs.”“Create very detailed, very long and very good material.”

4Which question asks for a neutral evaluation?“Why are electric cars so much better?”“What are the advantages and disadvantages of electric cars compared with conventional cars?”“An electric car is always the better choice, isn’t it?”

5What should you do if the first answer is unsuitable?Send the same prompt again without changes.Choose a completely different topic every time.Give specific feedback and clarify a missing building block, constraint or example.

6How does a structured prompt help with differentiation?Changing the learner level and objective adapts the same topic to different groups.Every group must receive the same answer.We leave technical verification entirely to AI.

References and further practice

References

Google (2024–2026): Google Prompting Essentials. Online course, Coursera.
OpenAI: Prompt engineering guide. OpenAI Platform documentation.
Anthropic: Prompt engineering overview. Claude documentation.
Lengyel, Zs. (2026): Introduction to the concepts of artificial intelligence; Concrete applications for content generation. VITA training presentations.

Links

Credits and copyright

Author: Zsolt LengyeliTStudy Hungary Training and Research Centre
Reviewer: Mária HartyányiiTStudy Hungary Training and Research Centre
Project: 2025-1-LT01-KA220-VET-000362456
Learning material: M1-T1-MLC4

Attachments

Download the English learning materialm1-t1-mlc4-text-en.pdf Download the practical exercisesm1-t1-mlc4-exc-en.pdf

This work may be reused and adapted under the terms of the Creative Commons Attribution-ShareAlike 4.0 International License (CC BY-SA 4.0).

Funded by the European Union. Views and opinions expressed are those of the author(s) only and do not necessarily reflect those of the European Union or the National Agency.