Every AI tool you use — ChatGPT, Claude, Gemini, Perplexity — lives or dies by the quality of what you put in. Prompt engineering isn't about memorising formulas. It's about learning to communicate clearly with a system that takes you completely literally.
Same tool.
Same question.
Wildly different results.
Two youth workers. One tool. The difference between a generic 5-point list and a full 90-minute workshop plan with icebreakers, activities, and facilitator notes. The difference is entirely the prompt.
When most people start using AI, they treat it like a search engine — type a few keywords and hope for the best. The results are mediocre, they conclude the tool isn't that useful, and they move on. This is the most common and most avoidable mistake in AI adoption.
AI language models are not search engines. They don't rank existing content — they generate new responses based entirely on the context you provide. The richer, more specific, and better-structured your input, the dramatically better your output. This relationship is more direct than almost any other tool you use: the quality of your thinking going in is reflected almost exactly in the quality of what comes back.
In practice this means that two youth workers asking ChatGPT the same question — "help me plan a session on digital citizenship" — can get results so different they feel like different tools entirely. One gets a generic 5-point list. The other gets a structured 90-minute workshop plan with icebreakers, differentiated activities for different learning styles, reflection prompts, and facilitator notes. The difference is not the tool. It is entirely the prompt.
"Prompt engineering is the new literacy. Just as learning to write a clear brief transformed what you could get from a freelancer, learning to write a clear prompt transforms what you can get from AI."
The good news: this skill is learnable in hours, not months. It follows consistent principles that transfer across every AI tool. You don't need to be technical. You need to be clear, specific, and willing to iterate. Everything in this section shows you exactly how.
Every strong prompt contains some combination of these five elements. You don't always need all five — but knowing each one and when to use it will transform your results immediately.
Tell the AI who it is. The role instruction is the fastest way to shift the register, expertise level, and perspective of any response. "You are a youth worker" produces completely different output to "You are an academic researcher" or "You are writing for a 14-year-old". The model draws on different parts of its training data depending on the role you assign.
Be specific: "an experienced Erasmus+ project coordinator" outperforms "a professional". Roles work especially well for writing tasks, where voice and audience awareness matter enormously.
Give the AI your situation. AI models have no knowledge of your organisation, your participants, your country's context, or your constraints — unless you tell them. Context is the ingredient most beginners leave out, and its absence is responsible for most generic, unhelpful AI output.
Think of it as the briefing you'd give a skilled freelancer on their first day: who are the people involved, what's the setting, what has happened before, what do they already know?
Be precise about what you want. This sounds obvious but is where most prompts fail. "Help me with my project report" is not a task — it's a gesture. "Write a 400-word executive summary of the attached project report, focusing on outcomes for young people and impact evidence" is a task.
Use action verbs: write, summarise, compare, identify, rewrite, translate, structure, explain, critique.
Tell the AI how to structure the output. Without format instructions, AI defaults to whatever structure it thinks is appropriate — which is often not what you need. Specify: length, structure, tone, and any layout requirements.
Format instructions are especially important for outputs you plan to use directly — a section heading structure you didn't expect means editing time you didn't budget for.
Tell the AI what NOT to do. Constraints are the most underused element in prompting. They prevent the model from making assumptions that waste your time. Negative instructions are just as powerful as positive ones, and often more efficient.
Copy any of these directly, or use them as starting points. Every prompt follows the five-ingredient structure from above. Replace text in [BRACKETS] with your own details.
You are an experienced non-formal education facilitator. Design a 90-minute workshop for [NUMBER] participants aged [AGE RANGE] on the topic of [TOPIC]. The group has [PRIOR KNOWLEDGE LEVEL] familiarity with this topic.
Structure it as: Objectives (3 bullets), Energiser (10 min), Main activity (50 min) broken into stages, Reflection (20 min), Closing (10 min). Include facilitation notes and any materials needed. Use non-formal, participatory methods throughout. Avoid lecture-style delivery.
Generate 5 icebreaker activities for a group of [NUMBER] participants from [NUMBER] different countries, ages [AGE RANGE]. The group has just met. Each activity should take 5–10 minutes, require no materials, work with mixed language levels, and be culturally neutral.
For each: name, instructions, purpose (what it builds), and one facilitator tip.
You are an experienced Erasmus+ grant writer. Write a 200-word theory of change paragraph for a project called "[PROJECT NAME]" that [BRIEF DESCRIPTION OF WHAT IT DOES]. The target group is [TARGET GROUP]. The expected change is [WHAT YOU WANT TO CHANGE].
Use the structure: current problem → our intervention → immediate outcomes → longer-term impact. Write in active voice, avoid jargon, and be specific about evidence of need. This will be read by a European funder.
You are a communications specialist. Rewrite the following project description for a general audience — someone with no knowledge of youth work or EU funding programmes. Maximum 150 words. Start with the problem, explain the solution clearly, and end with the impact. Avoid all acronyms.
[PASTE YOUR ORIGINAL DESCRIPTION HERE]
You are a professional and direct communicator. Help me write an email to [RECIPIENT TYPE] regarding [SITUATION]. The message needs to: [MAIN POINT 1], [MAIN POINT 2]. The tone should be firm but respectful.
Maximum 150 words. No unnecessary softening or excessive apology. End with a clear next step or request.
Rewrite the following participant information text so that it is clear and accessible to someone whose first language is not English, reading at approximately B1 level. Use short sentences. Replace idioms with literal equivalents. Define any specialist terms the first time they appear. Keep all factual content exactly the same.
[PASTE TEXT HERE]
I am writing a funding application for a project addressing [ISSUE] among young people in [REGION/CONTEXT]. Summarise the current evidence base for this issue: key statistics, trends, and research findings from the last 5 years.
Present as 4–5 bullet points suitable for a grant application. Note any figures that will need independent verification before use.
You are a policy analyst summarising for a non-specialist youth work audience. Explain the key points of [POLICY/DOCUMENT NAME] in plain language. Focus on: what it means for youth work organisations, any funding or participation opportunities, key deadlines or requirements, and what action is recommended.
Maximum 300 words. Use subheadings.
You are a social media manager for a youth organisation. Create a 5-post Instagram series about [TOPIC] aimed at young people aged [AGE RANGE]. For each post: caption (max 150 characters), 3 hashtag suggestions, and a brief description of what the accompanying visual should show.
Tone: [TONE]. Do not use corporate language or clichés like "empowering" or "transformative".
Write a 200-word newsletter section announcing [WHAT YOU'RE ANNOUNCING] to our network of youth work professionals. Context: [BRIEF CONTEXT]. Tone: warm and direct, not overly formal.
Include: what it is, why it matters, what readers should do next. End with a single clear call to action.
You are a non-formal education evaluator. Generate 8 reflection questions for participants at the end of [TYPE OF ACTIVITY/TRAINING]. The questions should cover: what they learned, how they will apply it, what surprised them, and what they would change.
Mix closed questions (for quick surveys) and open questions (for deeper reflection). Suitable for mixed language levels. Avoid leading questions.
You are a grant writer summarising project outcomes. Using the following data and participant feedback, write a 250-word impact narrative for our funder report. Focus on: what changed for participants, evidence of that change, and one or two specific stories or quotes that illustrate the impact. Write in past tense, active voice.
[PASTE DATA AND FEEDBACK HERE]
Explain [COMPLEX TOPIC] to a group of 15-year-olds who have no background in the subject. Use one clear analogy, two real-world examples from everyday life, and avoid all technical terminology. Maximum 200 words.
End with one question that would spark a good group discussion.
I have the following activity designed for adults: [DESCRIBE ACTIVITY]. Adapt it for young people aged [AGE RANGE] with [ANY SPECIFIC CHARACTERISTICS]. Keep the core learning objective the same.
Adjust: the language used in instructions, the time required, the materials, and any facilitation approach. Note any parts that need particular sensitivity with this age group.
These are the patterns that hold most people back. They're all fixable once you know what to look for — and every one of them has a reliable solution.
"Young people" spans a 10-year age range and dozens of different contexts. The model cannot calibrate tone, vocabulary, or content without this information.
Always include age range, background, and what they already know. This single change dramatically improves tone and pitch accuracy.
The model handles depth much better than breadth. Asking for everything produces shallow results across all items.
Each follow-up builds on the previous output. The model remembers the whole conversation — use that context.
Manual editing is slower and produces worse results than iterating with the model. You're doing the model's job.
Each refinement takes seconds. Three rounds of iteration is faster than one round of manual editing and produces stronger output.
AI defaults to whatever structure feels "complete" to it — which is often not what you need for the context you're working in.
Length, structure, headings, tone. Include all four in every prompt you write for anything you'll use directly.
AI output is a first draft, not a final product. Statistics, citations, and specific claims need verification.
The efficiency gain comes from starting further ahead — not from skipping the review. You're editing, not writing from scratch.
These techniques separate confident users from power users. None require technical knowledge — just a willingness to experiment with how you structure your thinking before you type.
The best prompt isn't the cleverest one. It's the one that would make a skilled human colleague understand exactly what you need.
Youthwork.AI · Prompt Engineering · Part 02