# 🟱 Beginner Level — Session 4 # The art of speaking to AI (prompting) > **Applied AI program** — Instructor: Yann Isola > Public: general public, from 12 years old — no technical prerequisites > Recommended duration: 2 hours (adjustable 1h30–2h30) --- ## 📋 Session overview | Element | Detail | |---|---| | **Main objective** | Know how to write a good prompt and improve a response by iteration | | **Key message** | The quality of the question determines the quality of the answer | | **Material** | Video projector, access to an AI chatbot (ChatGPT, Le Chat de Mistral, Claude, Gemini
), interactive web page of the session | | **Student deliverables** | “Perfect prompt recipe” sheet, completed exercises, quiz | ### Educational objectives At the end of the session, each participant knows: 1. **Define** what a prompt is in your own words 2. **Identify** the 5 ingredients of a good prompt (context, role, task, format, constraints) 3. **Transform** a vague prompt into a precise prompt 4. **Iterate**: rephrase and clarify when the answer is not appropriate 5. **Cut** a large request into small steps (prompt chaining) --- ## ⏱ Unfolded minute by minute (2h) | Time | Activity | Format | |---|---|---| | 0:00–0:10 | Home + session 3 reminder + shock demo “same question, two prompts” | Plenary | | 0:10–0:25 | Part 1: What is a prompt? Garbage in, garbage out | Interactive presentation | | 0:25–0:50 | Part 2: The 5 ingredients of a good prompt | Presentation + live demos | | 0:50–1:00 | ☕ Break | — | | 1:00–1:15 | Part 3: Simple techniques (lists, tables, steps) + the power of the role | Live demos | | 1:15–1:30 | Part 4: Iterate + prompt chaining | Presentation + demo | | 1:30–1:50 | 🏆 Activity: the prompt competition | Team workshops | | 1:50–2:00 | Quiz + summary + teaser session 5 | Plenary | --- ## 🎬 Opening: the shock demo (10 min) **Objective:** create the “wow” effect from the first minute. Project a live chatbot. Type these two prompts one after the other: **Prompt A (wave):** > “Tell me about volcanoes. » **Prompt B (precise):** > “You are a passionate museum guide who speaks to 10-year-old children. Explains how a volcano erupts, in 5 numbered steps, with a fun comparison per step (for example with a bottle of soda). Maximum 10 lines. » Let the class compare the two answers. Ask the question: > “The AI is the same. What has changed? » Expected answer: **the question**. That’s what the session is all about. ⚠ **Volatile point:** interfaces and models evolve quickly. Test your demos **the day before** the session — exact answers vary from day to day and from tool to tool. This is normal, and it is even a good point to mention to students. --- ## 📖 Part 1 — What is a prompt? (15 mins) ### Simple definition A **prompt** (English word meaning “prompt” or “instruction”) is **the message that you send to the AI**: a question, an instruction, a request. That's all the AI ​​knows about what you want. **Key analogy to use:** Ordering at a restaurant. - “I want to eat. » → the waiter brings anything. - “I would like a margherita pizza, thin crust, without olives, please. » → you get exactly what you want. AI is like an ultra-competent server but **doesn't read your mind**.### “Garbage in, garbage out” English expression which means “waste in, waste out”. In other words: **if the question is fuzzy, the answer will be fuzzy.** The AI ​​doesn't guess what's in your head — it only knows the words you give it. **Points to hammer home:** - AI is not a soothsayer. She doesn't know your background, your age, or your goal — unless you tell her. - A bad result is not always the fault of the AI: it is often the fault of the prompt. - Good news: writing a good prompt can be learned in one session. It's today! ### Mini-interaction Ask: “Who has ever been disappointed by an AI response?” » Collect 2-3 anecdotes. Rephrase: “What if we rewrote the question together at the end of class? » (Save them for the final competition.) --- ## 🧑‍🍳 Part 2 — The 5 ingredients of a good prompt (25 min) Present the **recipe**: as in cooking, a good prompt combines ingredients. You don't need 5 every time, but the more you add, the more precise the answer. ### 1. 🌍 CONTEXT — “This is my situation” What the AI needs to know about you and your situation. > “I'm in 5th grade and I have a presentation on bees on Friday. » ### 2. 🎭 The ROLE — “You are
” The character that the AI should play. This changes the tone and level of the response enormously. > “You are a SVT (Life and Earth Sciences) teacher who loves bees. » **Must-see live demo:** Ask the same math question (“Explain fractions to me”) three times: 1. No role 2. “You are a math teacher for 6th graders” 3. “You are a pastry chef who explains fractions with cakes” The three answers will be radically different. Guaranteed effect. ### 3. 🎯 The TASK — “Do this” What you want, with a **precise action verb**: explains, summarizes, compares, corrects, translates, invents, list... > “Explain to me how bees make honey. » ### 4. 📐 The FORMAT — “Present it like this” The form of the response: bulleted list, table, numbered steps, short text, dialogue, poem, etc. > “In the form of 5 numbered steps. » ### 5. 🚧 CONSTRAINTS — “With these limits” The rules to respect: length, language level, tone, things to avoid
 > “In 3 sentences maximum”, “for a 10 year old child”, “with emojis”, “without complicated words”. ### The complete assembled prompt > “**[Context]** I am in 5th grade and I am preparing a presentation on bees. **[Role]** You are a passionate SVT teacher. **[Task]** Explain to me how bees make honey, **[Format]** in 5 numbered steps, **[Constraints]** with simple sentences and one emoji per step. » **Mnemonic:** Have the class find an acronym with C-R-T-F-C, or suggest the phrase: “**C**haque **R**query **T**finds **Easily **C**hemin”. The important thing is not the order, but the presence of the ingredients. --- ## ⚡ Part 3 — Simple techniques that change everything (15 min) ### Request a structure Show live the difference between: - “Give me gift ideas” → text pad - “Give me 5 gift ideas **in list form**, with the approximate price for each” → clear list - “
**in table form** with the columns: idea, price, for whom” → readable table**Magic words to display on the board:** - “Make me a **list of
**” - “Present this in a **table** with the columns
” - “Explained in **numbered steps**” - “**Summary in 3 sentences**” - “Give an **example** for each point” ### The power of the role (deepening) Explain why it works (simple version): the AI has read millions of texts. When we say to her “you are a 6th grade teacher”, she “activates” everything that resembles this context: simple vocabulary, friendly tone, adapted examples. Without a role, she takes on a medium, all-purpose tone. ### Constraints in practice Quick game: ask the class to propose crazy constraints for the question “Explain gravity”: in 3 sentences / for a 5-year-old child / with emojis / sports commentator style / in a poem. Test 2-3 live. Laughter guaranteed, and the lesson passes: **constraint sculpts the response**. --- ## 🔁 Part 4 — Iterate and chain (15 min) ### Iterate: the conversation, not the single question Essential message: **missing your first response is NORMAL.** Even professionals adjust their prompts. The AI ​​remembers the current conversation: there is no need to repeat everything, we clarify. Examples of reminders to display: - “It’s too long, summarized in 5 lines. » - “Too complicated, explain as if I were 10 years old. » - “Good, but give concrete examples. » - “Start again with a funnier tone. » **Analogy:** it's like sculpting. The first prompt gives a rough block, each restart refines the shape. ### Prompt chaining “Chaining of prompts” = **cut a large request into small steps**, one per message, like a chain of links. **Example: prepare a presentation on dolphins** 1. “Give me 5 plan ideas for a presentation on dolphins (5th grade level). » 2. “Develop plan #2 in 3 parts with 2 ideas per part. » 3. “Write the introduction, 5 sentences maximum, with a compelling question. » 4. “Propose 3 questions that the class could ask me, with short answers. » Why is it better than asking everything at once? - We **keep control** at each step (we can correct before continuing) - Each response is **more careful** because the request is simpler - We **understand** what we are building (instead of copying a ready-made tile) ⚠ **Volatile point:** some recent tools are increasingly handling complex requests in one go. Chaining nevertheless remains the most reliable and educational method to get started. --- ## 🏆 Flagship activity — The prompt competition (20 min) ### Principle Teams compete: **who will write the prompt that produces the best answer** for a given challenge? ### Organization 1. **Teams of 2-3** people. 2. Announce the **challenge** (choose according to the audience): - đŸ„‡ Challenge 1: “Obtain the best explanation of photosynthesis for an 8 year old child” - đŸ„ˆ Challenge 2: “Obtain the best birthday menu for 10 people with €50” - đŸ„‰ Challenge 3: “Obtain the best rules for an invented game that can be played in class without equipment” 3. Each team writes its prompt **on paper** (5 min) — this avoids tinkering live and forces reflection. 4. Each team dictates its prompt, you type it live on the video projector (where the teams use their stations). 5. The class **votes** for the best answer (not the best prompt!). 6.Debrief: analyze the winning prompt using the 5 ingredient grid. How many did he have? ### Evaluation grid (to be projected) | Ingredient | Here ? | |---|---| | 🌍 Background | ☐ | | 🎭 Role | ☐ | | 🎯 Clear task | ☐ | | 📐 Requested format | ☐ | | 🚧 Constraints | ☐ | ### Variant without screen If not enough positions: the trainer is the “human AI” and voluntarily responds literally. Prompt vague → answer deliberately off the mark. Very funny and very effective educationally. --- ## ❓ Frequently asked questions from students (and ready answers) **“Should you be polite to AI? Say please? »** It's not obligatory, the AI doesn't have feelings. But studies suggest that clearly and politely worded prompts sometimes yield slightly better responses — mostly because they are better structured. And it’s a good communication habit! **“Why doesn’t the AI ​​understand my question?” »** She “understands” the words, not your hidden intentions. Reread your prompt as if you were someone else: is it really clear? **“If I put all 5 ingredients, the answer is still perfect? "** No ! The AI ​​can make mistakes even with a perfect prompt (previous session reminder: hallucinations). A good prompt increases the chances, it does not guarantee them. We always check. **“Is a long prompt better? »** Not necessarily. A long but confusing prompt is worse than a short and clear prompt. The rule: **clear first, complete later**. **“Does it work the same on all tools? »** ⚠ The main principles (the 5 ingredients) work everywhere: ChatGPT, Le Chat, Claude, Gemini, Copilot
 The interface details change, the principles remain. --- ## ⚠ Points of vigilance for the trainer - **Test all your demos the day before** — the models change, as do the answers. - **Have backup screenshots** in case the network goes down or the tool is saturated. - **Remember the rules of use**: no personal data in the prompts (full name, address, passwords) — link with the session on private life. - **Do not present prompting as a magic formula**: it is a communication skill, not a secret code. - **Adapt the competition challenges** to the public (teenagers: games, presentations; adults: emails, organization, cooking). - **GDPR (General Data Protection Regulation)**: if you use shared accounts in class, do not enter any personal data of the participants. --- ## ✅ Summary to note (end of session) 1. A **prompt** = the instruction given to the AI. That's all she knows about our need. 2. **Fuzzy question → fuzzy answer** (“garbage in, garbage out”). 3. The recipe: **Context + Role + Task + Format + Constraints**. 4. Magic words: list, table, numbered steps, “in X sentences”, “for a child of X years”. 5. **Iterating is normal**: we specify, we reformulate, we refine. 6. For large requests: **one step at a time** (chaining). 7. A good prompt improves the answer but **we always check** the facts. **Teaser session 5:** “Now that you know how to talk to the AI
let’s find out what it can do other than text!” » --- *Professor document — Applied AI Program — Yann Isola — Beginner Level 🟱 — Session 4/10*