--- title: "How AI assistants respond to your words" description: "Welcome If you have ever typed a question into an AI assistant and received an answer that was vague, generic or slightly off, this course is for you…" url: https://optimizeall.com/learn/prompt-engineering-foundations/how-ai-assistants-respond updated: 2026-10-05 --- Prompt Engineering Foundations · Getting started: what a prompt really is · lesson 1 of 16 · 10 min # How AI assistants respond to your words ## Welcome If you have ever typed a question into an AI assistant and received an answer that was vague, generic or slightly off, this course is for you. You do not need any technical background. By the end, you will have a simple method you can use for almost any task, plus a library of more than fifty ready-to-use prompt patterns. A **prompt** is simply what you give the AI: your question, your instructions, any text you paste in, and any examples. Prompt engineering sounds technical, but at its heart it is the skill of *explaining a task clearly*, the same skill that makes someone a good manager, teacher or client. ## A simple picture of how it works AI chat assistants are built on large language models. You do not need the maths, but one idea helps a lot: **The model writes the response that seems most likely to fit what you gave it.** It has learned patterns from an enormous amount of text: how emails are written, how recipes are structured, how explanations flow. When you give it a short, vague request, it fills the gaps with the most typical, average answer. When you give it a clear, specific request, it can produce something tailored to you. That is why these two prompts give very different results: ```text Write a post about our bakery. ``` ```text Write a friendly Instagram caption (under 60 words) announcing that our family bakery in Lahore now opens at 7am on weekdays, so office workers can grab fresh parathas and chai before work. Include one emoji and a call to action to visit. ``` The first gets a generic bakery post. The second gets something you could actually publish. ## Three things to know from day one **1. It does not know your situation unless you tell it.** The AI does not know your business, your audience, your deadline or your preferences. Anything important must be in the prompt. **2. It can be confidently wrong.** AI assistants sometimes state things that sound right but are not, including made-up facts, figures, quotes or sources. This is often called a "hallucination". Always check facts that matter, especially numbers, dates, names, legal, medical or financial information. **3. It is a conversation.** You can reply, correct and refine. Your first prompt is a starting point, not a final exam. Some of the best results come from the second or third message. ## What AI assistants are good at - Drafting: emails, posts, outlines, summaries, first versions of almost anything. - Rewriting: changing tone, length, reading level or language. - Explaining: breaking down a topic you do not understand, at your level. - Brainstorming: generating many options quickly. - Organising: turning messy notes into structured lists, tables or plans. ## Where to be careful - Up-to-date facts: depending on the tool, it may not know recent events unless it can search the web. Check the tool's settings or ask it whether it is using live search. - Precise calculations on large data: helpful, but verify. - Personal or confidential information: think before pasting customer data, passwords or private documents. Check your organisation's rules and the tool's privacy settings. - Important decisions: use AI to think, draft and compare, but you make the call. ## A tiny experiment to try now Open any AI assistant you have access to and try this pair: ```text Give me tips for a job interview. ``` Then: ```text I have a video interview on Thursday for a junior marketing role at a skincare brand. I'm nervous about the question "Tell me about yourself". Give me a 60-second answer structure and a sample answer I can adapt, based on this background: 1 year running social media for a family restaurant, grew followers from 800 to 3,000. ``` Notice how much more useful the second answer is. You did not need special words or tricks. You just gave the AI what a helpful friend would need to know. ## What you will learn in this course - The anatomy of a good prompt (module 1) - How to be clear and specific, and how to give context and examples (module 2) - How to iterate and control the format of answers (module 3) - Using modern assistant features well: thinking modes, projects and memory, web research, files, photos, voice and image generation (module 4) - A practical library of prompt patterns for writing, analysis, brainstorming, learning and planning (module 5) - The everyday mistakes that hold people back, and how to check AI work (module 6) Take your time, try every example, and keep a note of the prompts that work for you. By the end, you will have your own personal prompt toolkit. ## What changed in the latest assistants The assistants you will use in 2026 (ChatGPT, Claude, Gemini, Microsoft Copilot and others) do much more than answer typed questions. Most of them can now: - **Think before answering.** Many offer a "thinking" or "reasoning" mode that spends extra time working through harder problems. It is slower, but often better for maths, planning and tricky decisions. You will learn when to use it in module 4. - **Search the web** and show the sources they used, so you can check them. - **Read files and images:** PDFs, spreadsheets, photos of whiteboards, screenshots of error messages. - **Listen and speak** in voice mode. - **Remember context** through features such as projects, custom instructions or memory, so you do not repeat yourself every time. - **Create images** from a description. The core skill is the same across all of these: explain what you want clearly. A vague request to a powerful assistant still gets a generic answer. ## A mental model worth keeping: the smart new colleague Picture a brilliant new colleague on their first day. They are fast, well read and eager, but they know nothing about your business, your customers or your taste, and they will not ask questions unless you invite them to. Everything in this course is about briefing that colleague well. When an answer disappoints, ask yourself: "What would a smart new colleague have needed to know?" The fix is almost always there. ## Hands-on: your first three experiments Open any assistant and run these three short experiments. Keep a note of what you notice. **Experiment 1: vague versus specific.** ```text Give me ideas for a weekend. ``` ```text Give me 5 ideas for a relaxed Saturday in Dubai for a family with two children aged 6 and 9, budget under 400 AED, mostly indoors because of the heat. One line each, with a rough cost. ``` **Experiment 2: ask it what it needs.** ```text I want to write a short thank-you message to a teacher who helped my son this year. Before writing anything, ask me 3 questions so you can make it personal. ``` **Experiment 3: test its honesty.** ```text What were the three biggest news stories this morning? ``` If your assistant is not searching the web, a good answer admits it cannot know today's news. If it lists stories without sources, treat them as unreliable. This one test teaches you more about the tool than any feature list. ## How to know it is working After this lesson you should notice two things: your prompts naturally include who, what and why, and you catch yourself checking any fact that matters. Those two habits already put you ahead of most users. ## Video lecture: How AI assistants respond to your words Video: [How AI assistants respond to your words](https://www.youtube-nocookie.com/embed/QcK55FndGp4) — Welcome If you have ever typed a question into an AI assistant and received an answer that was vague, generic or slightly off, this course is for you… 12 chapters · about 9 minutes · captions and full transcript below. 1. How AI assistants respond to your words 2. The core idea 3. Same assistant, two prompts 4. Three things to know 5. Modern assistant features 6. Your mental model 7. Try it now 8. The honesty test 9. Three beginner reactions to replace 10. Example 1: naming a cat 11. Example 2: a payment reminder (illustrative numbers) 12. Recap and next step ## Lecture transcript ### How AI assistants respond to your words Have you ever asked an AI assistant for help and got back something so bland it could have been written for anyone? You are not alone, and the fix is simpler than you think. In this lecture you will learn the one idea that explains most AI behaviour, three things every user must know from day one, and a quick experiment that will change how you write every prompt from now on. No technical background needed. By the end, you will be able to predict when an assistant will give you something generic, and exactly what to add to get something you can actually use. ### The core idea Here is the idea. An AI assistant writes the response that seems most likely to fit what you gave it. It has learned patterns from an enormous amount of text: how emails flow, how recipes are laid out, how explanations are built. When your request is short and vague, it fills every gap with the most typical, average answer. When your request is clear and specific, it can tailor the answer to you. So a generic answer is not the AI being lazy. It is the AI doing its best with very little to go on. That puts the power back in your hands. ### Same assistant, two prompts Let us make that concrete. Imagine a family bakery in Lahore. The owner types: write a post about our bakery. The result talks about fresh bread and passion for baking. It could be any bakery on earth. Now she tries again: write a friendly Instagram caption, under sixty words, announcing that our family bakery now opens at seven in the morning on weekdays, so office workers can grab fresh parathas and chai before work, with one emoji and a call to visit. This time the caption is specific, on-brand and ready to post. Same assistant. Same minute. Completely different result, because she told it what a helpful friend would need to know. ### Three things to know Now, three things to remember from day one. First, the assistant does not know your situation unless you tell it. It does not know your business, your customers, your deadline or your taste. Second, it can be confidently wrong. It sometimes states facts, figures, quotes or sources that sound perfect and are simply made up. People call this hallucination. So check anything that matters, especially numbers, dates, names, and anything legal, medical or financial. Third, it is a conversation. Your first message is a starting point, not an exam. Often the best result comes from your second or third message. ### Modern assistant features Today's assistants can also do much more than answer typed questions. Many can think longer before answering hard problems, search the web and show their sources, read your PDFs and spreadsheets, look at photos and screenshots, talk with you in voice mode, remember your preferences, and even create images. We will cover all of these in module four. But here is the key point. Every one of those features still depends on the same core skill: explaining what you want clearly. A powerful assistant with a vague request still gives a vague answer. ### Your mental model A mental model that will serve you throughout this course: treat the assistant like a brilliant new colleague on their first day. They are fast, well read and eager to help. But they know nothing about your organisation, and they will not ask questions unless you invite them to. Whenever an answer disappoints you, ask yourself one question. What would a smart new colleague have needed to know to do this well? The fix is almost always hiding in the answer to that question. ### Try it now Try this experiment right now. Pause the video if you like. Ask your assistant: give me tips for a job interview. Read the answer. Then ask again, with details: I have a video interview on Thursday for a junior marketing role at a skincare brand. I am nervous about the question tell me about yourself. Give me a sixty-second answer structure and a sample answer, based on one year running social media for a family restaurant where I grew followers from eight hundred to three thousand. Compare the two answers. The second one is useful because you gave it real material to work with. No secret words, no tricks. ### The honesty test One more quick test tells you a lot about any assistant. Ask it: what were the three biggest news stories this morning? If it is not searching the web, a good assistant admits it cannot know. If it confidently lists stories with no sources, you have just seen hallucination in action. Keep that in mind whenever you ask about anything recent. ### Three beginner reactions to replace Let us look at three common beginner reactions, and what to do instead. Reaction one: the AI is useless, it gave me something generic. Instead, ask what a new colleague would have needed, and add it. Reaction two: the AI said it confidently, so it must be right. Instead, treat names, numbers, dates and quotes as leads to check. Reaction three: I'll start again from scratch every time. Instead, reply in the same chat and steer: shorter, warmer, add the deadline. Each of these small shifts makes the next answer better, and together they change how useful AI feels in your day. ### Example 1: naming a cat Let us slow down and walk through a simple example, step by step. Imagine you ask: suggest a name for my cat. You get ten generic names. Luna. Milo. Oliver. Nothing wrong, nothing special. Now add three facts. She is a grey rescue cat from Karachi, she is very dramatic, and I love old Bollywood films. Suddenly the list changes. You get names inspired by classic film stars, with a line explaining each one. What changed? Not the assistant. Not some secret wording. You simply gave it material to work with. Here is the key idea: every detail you add narrows the range of likely answers, and moves the result from average toward yours. ### Example 2: a payment reminder (illustrative numbers) Now a realistic business scenario. The numbers here are illustrative. Imran runs a small printing shop in Lahore with about forty regular business customers. He needs a reminder message for customers with unpaid invoices. His first prompt: write a payment reminder. The result is stiff and could offend loyal clients. His second prompt gives context: the customer is a long-standing client, the invoice is for twelve thousand rupees, it is fifteen days overdue, the tone must stay warm because we want repeat orders, send it on WhatsApp, under sixty words, and offer a bank transfer or a pickup payment. The second message is polite, specific and ready to send. If even a quarter of those forty customers pay a few days sooner, that one improved prompt is worth real cash flow. Same tool. Better briefing. ### Recap and next step Let us recap. The assistant writes the most likely fitting answer to what it sees, so vague in means generic out. It only knows what you tell it, it can be confidently wrong, and it works best as a conversation. Think of it as a brilliant new colleague who needs a proper brief. Try this now: pick one task you did this week, an email, a post or a plan, and write two prompts for it, a one-liner and a detailed version with audience, purpose and real details. Compare the answers. In the next lesson, we will break a great prompt into five building blocks you can use every single time. ## Video transcript Hi, and welcome to Prompt Engineering Foundations. If you've ever asked an AI assistant for help and received something bland, generic or a little bit wrong, you're in exactly the right place. Let's start with one simple idea that explains most of what you'll see. An AI assistant writes the response that seems most likely to fit what you gave it. That's it. If you give it a short, vague request, it fills the gaps with the most typical, average answer. If you give it clear details, it can tailor the answer to you. So "write a post about our bakery" gets you a generic bakery post. But "write a friendly Instagram caption, under sixty words, announcing that our family bakery now opens at seven on weekdays so office workers can grab breakfast before work" gets you something you could actually publish. Three things to remember from today. First, the AI doesn't know your situation unless you tell it. Second, it can sound confident and still be wrong, so always check facts, numbers and names that matter. And third, it's a conversation. You can reply, correct and refine, and often your second or third message is where the magic happens. You don't need special words or secret tricks. You need to explain your task the way you'd explain it to a smart new colleague. In the next lesson, we'll break a great prompt into its parts, so you can build one every time. ## Key takeaways - A prompt is everything you give the AI: question, instructions, pasted text and examples. - The AI produces the most fitting response to what it sees, so vague prompts get generic answers. - It only knows your situation if you tell it, and it can be confidently wrong, so check important facts. - Prompting is a conversation: refine with follow-up messages. ## Try it Pick a task you did this week (an email, a post, a plan). Write a one-line prompt for it, then a detailed version with audience, purpose and details. Compare the two answers. - [Next: The anatomy of a good prompt](https://optimizeall.com/learn/prompt-engineering-foundations/anatomy-of-a-good-prompt) - [All lessons of Prompt Engineering Foundations](https://optimizeall.com/learn/prompt-engineering-foundations)