AI-Literate
A practical, no-nonsense guide to using AI — whoever you are, whatever you do.
Whether you've never opened an AI tool or you use one every day, this guide will help you understand what's actually going on, how to get the most out of it, and — just as importantly — where to be careful.
13 sections · 17 min readWhat is Generative AI and How Does it Work?
Generative AI (often just called "AI" or "GenAI") is software that generates content — text, images, code, or documents — in response to something you give it. That "something" is called a prompt, and it can be words, questions, files, images, or a combination.
Well-known examples include ChatGPT, Claude, Gemini, and Copilot, among many others.
"AI" gets used loosely to mean lots of different things. What actually exists today are Large Language Models (LLMs) and Machine Learning (ML) — real, specific technologies that are very good at certain tasks, like predicting a likely next word in a sentence. A true "general" artificial intelligence — a machine that actually thinks, understands, or reasons the way a person does — does not exist. Nobody knows for certain whether today's tools are even a step toward that. Worth remembering whenever something is marketed as "AI."
AI is a tool — much like a calculator. A calculator doesn't do your maths homework for you in any dishonest sense; it helps you work faster and more accurately. GenAI is the same principle, just applied to a much wider range of tasks. It's still worth being able to do the maths yourself, though — a calculator is only useful to someone who already has a rough sense of what the right answer should look like. The same is true here.
Where Does it Get its Information?
AI tools are trained on vast amounts of text — a huge portion of the written internet. This includes academic research papers, books, forums, news articles, code repositories, and yes, the occasional angry rant someone posted online at 2am.
This means AI can be excellent on well-documented topics — coding, widely covered subjects, established processes — but may be less reliable on niche, recent, or contested topics.
AI does not work like a search engine. It doesn't "look things up" when you ask a question. Instead, it learned statistical patterns during training — which words, ideas, and structures tend to follow others. When you send a message, it predicts the most coherent and relevant response based on those patterns.
Think of it less like a library and more like a very well-read tutor who has absorbed an enormous amount of information and can synthesise it on demand — but who can also occasionally misremember things, fill in gaps with plausible-sounding nonsense, or state something completely wrong with total confidence. The last bit is important enough that it gets its own section.
Because of this, it's worth building the habit of checking things yourself with a real search engine — both before you lean on an AI's answer for anything that matters, and afterwards if something feels off.
How to Interact With it
AI tools work through a simple text-based conversation. You send a message, it responds. Many tools also let you attach files, images, or documents for the AI to read and analyse.
Most AI tools maintain memory within a conversation — it remembers what you said earlier in the same chat. Some carry memory across conversations, though this varies.
- Quality in, quality out. Vague questions tend to get vague answers.
- You can give AI a role. Starting with "Act as an experienced project manager..." can meaningfully shape the tone and focus of responses.
- You can always ask for a redo. If the first response isn't right, ask it to explain, simplify, or try again.
A simple structure for any prompt:
- Say what you're trying to do. One or two sentences on the actual problem or goal.
- Give the relevant background. Include the data, context, or constraints AI needs — don't make it guess your situation.
- Explain the approach you want. If you have a preferred method, order, or logic in mind, say so.
- Say what "done" looks like. The clearer your objective, the easier it is for AI to tell when it's actually finished the job.
You'll get to practise this in Section 13.
Ask vs. Do — Being Clear About What You Want
One of the most common sources of frustration is not being explicit about the type of response you want. There's a significant difference between asking for an opinion and asking for an action.
Vague
- "What do you think of this document?"
- "Can you help with my email?"
Clear
- "Review this and tell me what you think — don't change anything yet"
- "Rewrite this email to be more concise"
If you're not clear, AI will make an assumption — and it may guess wrong. Get into the habit of being explicit about format too: if you want bullet points, say so. If you want it short, say so.
Try: "Give me your opinion first, then ask me if I want you to action it." This keeps you in control of the flow.
Understanding Tokens — Why There Are Limits
AI tools run on a system of tokens. Every piece of text you send, every file you attach, and every word of the response generated uses tokens. The more complex or lengthy the task, the more tokens are consumed.
Most AI tools offer a free tier with token limits — for example, a cap on how many images you can analyse per day, or how many long documents you can process in an hour. Paid versions significantly increase these limits.
Start a new conversation when you switch topics. Every message you send causes the tool to re-process the entire conversation history — so a long, mixed conversation quietly racks up token usage in the background. More importantly, when you mix topics, the AI starts trying to connect things that aren't related, and the quality of its responses drops. A fresh conversation gives it a clean slate.
Every message you send to an AI tool uses real energy — and, at the data-centre level, often real water for cooling. It's not a reason to avoid AI, but it is a reason to be a bit deliberate: don't paste in huge documents you don't need to, and don't regenerate the same response five times out of habit rather than need.
Hallucinations — The Most Important Thing to Know
AI can and does make things up. This is known as "hallucination." It will sometimes state incorrect facts, fabricate references, invent statistics, or produce plausible-sounding but entirely wrong information — with complete confidence and no indication that anything is amiss.
This is not a bug that will simply be fixed one day. It is a characteristic of how these models work. This is exactly where the search-engine habit from Section 2 earns its keep.
- Never use AI output for anything important without checking it
- Be especially cautious with specific facts, figures, dates, names, and citations
- If something sounds surprising or too convenient, verify it independently
- The more niche or specific the topic, the higher the risk of inaccuracy
AI is a powerful tool for drafting, structuring, summarising, and generating ideas. It is not a reliable source of ground truth on its own.
There's a related problem that's just as important.
Hallucination is about AI getting facts wrong. This is different: AI will also construct fluent, well-structured, persuasive-sounding arguments for almost any premise you give it — including wrong, misleading, or one-sided ones. It doesn't push back. It builds on what you give it.
If your starting idea is flawed, biased, or simply wrong, AI will help you argue it convincingly — without flagging the problem. The more polished the output, the easier it is to mistake fluency for validity. Don't treat AI's confidence as confirmation that you're right.
A well-structured, articulate argument used to require real effort and at least some grasp of the subject. That bar no longer exists. The quality of the writing tells you nothing about the quality of the underlying thinking. Read with the same scepticism you'd apply to any unverified source — perhaps more.
An influencer with a flat-earth theory and access to AI can now produce polished, coherent, compelling content at scale. The arguments didn't get more accurate — they just got much easier to make, and much harder to dismiss at a glance.
Privacy and Data — What to Share and What Not To
Free AI tools have to pay for the huge amount of computing power they use somehow. Often that's by using what you type to improve their models, or — same as free social media — by building a profile of you to sell access to. This isn't unique to AI, but it's worth carrying into every free tool you use: if you're not paying for it, ask what you're paying with instead.
Anything you type into an AI tool may be stored, reviewed, or used by the provider depending on their terms of service. For personal use, it's worth reading the small print.
For professional or work use, this becomes more important. Many organisations have specific policies on what can and cannot be shared with external AI tools.
- Don't paste in confidential client data, personal data, or sensitive commercial information unless you're using a tool your organisation has specifically approved and secured for that purpose.
- Be aware that some corporate AI deployments are specifically designed to keep data within a secure environment — know which you're using.
- When in doubt, anonymise or generalise the information before sharing it.
If you use AI to write code or build a tool and you don't understand the code it produces, you have no way to know whether it's insecure, doing something you didn't ask for, or quietly broken — and if it stops working, you might not be able to fix it either. Never connect anything AI-built to real personal or financial information: passwords, bank details, personal records. This matters even more for anything installed on your device, since it runs with real access to your files and system, not inside a safe, sandboxed chat window.
AI can now generate convincing fake photos, audio, and video of real people. Don't create or share this kind of content to deceive someone — and stay a little sceptical of "shocking" images, clips, or voice recordings you see online, for the same reason you stay sceptical of AI-written text: it can look completely real and still be entirely fake.
AI as a Study Partner
Think of AI less like something that does your homework for you, and more like a tutor sitting next to you — helping you think through an idea, explaining something you're stuck on, or helping you say what you already mean more clearly. Used that way, it's not cheating. Using a calculator isn't cheating. Using a spell-checker isn't cheating. Using AI to help you develop and express your own thinking works the same way — in most contexts.
It is cheating when...
- You're being assessed on your own knowledge in an exam, test, or interview
- You submit AI-generated work as your own where original thinking is what's being graded
- You enter AI content into a competition or assignment that explicitly prohibits it
- You use it to deceive someone about your own capabilities
It is not cheating when...
- You have the idea but need help expressing it clearly
- You could do it yourself but it would take a lot longer
- You use it to explore an idea, research options, or structure your thinking
- You review, edit, and take responsibility for the final result
You own and are responsible for anything you send out, regardless of whether AI produced it. If you wouldn't be comfortable defending it if questioned, don't send it.
Rules vary by school and by class — some teachers are fine with AI for brainstorming but not for drafting, others have stricter limits. This guide gives you general principles; always check what your actual school or teacher says before assuming.
Getting the Most Out of AI
- Start with clarity. Before you open a conversation, spend a moment thinking about what you actually want. A clear goal produces a better result.
- Chunk your work. Don't ask AI to take two complex documents, compare them, and produce a new one all in one go. The output will almost certainly be poor. Break the task into steps and work through them one at a time.
- Treat it like a capable but occasionally unreliable assistant. It will sometimes misinterpret your instructions, make assumptions, or go off in completely the wrong direction. That's normal — correct it, redirect it, and carry on.
- Experiment. Some AI tools are better at certain tasks than others. It's worth trying a few to see what works best for your needs — think of it like trying different apps or different strategies in a game: the "best" one depends entirely on what you're trying to do.
This is one of the most common mistakes. When you mix topics in a single conversation, the AI tries to find connections between things that aren't connected — and the outputs quietly get worse. You may not notice it happening. When you're done with a topic, start fresh. Think of it like a substitute teacher stepping into a lesson already in progress: they do a much better job once someone fills them in on what's actually going on.
You can ask AI how to use AI. Not sure how to approach a task? Ask it to suggest a structured approach first. Then work through that approach step by step. It's surprisingly good at this.
Example Use Cases
To give a sense of the range of what's possible:
| Use case | Complexity | Example prompt |
|---|---|---|
| Answering factual questions | Simple | "What does API stand for and what does it do?" |
| Explaining concepts in plain language | Simple | "Explain machine learning like I'm 12" |
| Rewriting text in a different tone | Simple | "Rewrite this email to sound more friendly and less formal" |
| Drafting from rough notes | Simple | "Turn these bullet points into a professional email" |
| Generating ideas | Simple | "Give me 10 ideas for a team away-day on a tight budget" |
| Summarising long content | Moderate | "Summarise this 20-page report in 5 bullet points" |
| Comparing two documents | Moderate | "Here are two policy drafts — what are the key differences?" |
| Structuring a project plan | Moderate | "I need to migrate our database — suggest a phased approach" |
| Analysing data | Moderate | "Here's our sales data for Q1 — what trends do you see?" |
| Writing or debugging code | Moderate | "Here's my Excel formula — why isn't it working?" |
| Multi-step research and drafting | Complex | "Help me research, structure, and draft a business case for X" |
Working Through a Bigger Project
For more complex tasks, a step-by-step approach works best. Here's how that might look in practice:
- Describe the overall goal to AI and ask it to suggest a structured approach. Save that as your working plan.
- Feed the plan back in and say "Let's start with step 1."
- Work through step 1, review the output, make any edits, and save the result.
- In the next message, include the updated plan and the output from step 1, then say "Let's move to step 2."
- Continue this way, keeping the AI informed of where you are and what has already been done.
- Keep going until you're satisfied with the outcome.
This approach keeps the work manageable, the context clear, and the quality of output significantly higher than trying to do everything at once.
This is a living guide. As AI tools evolve, some specifics will change — but the core principles of clarity, scepticism, ownership, and common sense will remain relevant.
Can You Tell if Something Was Written by AI?
As AI becomes part of everyday writing, being able to recognise AI-generated text is increasingly useful — whether you're reviewing work, reading online content, or wondering if that unusually polished email came from a person or a prompt.
No single signal is definitive. But certain patterns appear so consistently they've become well-known tells.
Structural habits
- Bullet points and numbered lists for everything, even when prose would be more natural
- Paragraphs that are all roughly the same length
- Everything neatly balanced — both sides covered, tidy conclusion — even when not asked
- Responses that end with a sentence restating what was just said
- Headers and sub-headers in what should be a simple message
Vocabulary tells
These words appear far more often in AI output than in ordinary human writing:
Tone and style
- Hedges constantly — "it's worth noting that", "many experts believe", "it's important to recognise"
- Diplomatically balanced on everything, rarely takes a strong position even when one was asked for
- No typos, no colloquialisms, no regional expressions or slang
- No personal anecdotes or specific detail — examples are always generic and safe
- Perfect punctuation throughout, even in casual messages
- Heavy use of the long dash — like this — where a comma or nothing would be more natural
When someone who normally writes casually suddenly produces three well-structured paragraphs with a balanced argument and a closing thought — the words might be fine, but the voice isn't theirs. For personal or workplace messages, this is the biggest flag of all.
Read it back as yourself before you send. Cut the unnecessary structure. Add a specific detail, your actual opinion, or something only you would say. Make it sound like you — because ultimately, you're responsible for it.
Look for clusters of signals, not one thing in isolation. Compare it to how the person normally writes. If it matters — a job application, a piece of journalism, a classmate's essay — it's reasonable to ask directly.
Tools like GPTZero exist to flag AI-written content automatically, but they have high false positive rates — legitimate human writing regularly gets flagged as AI. Don't use them to accuse someone. Treat them as one signal among several, not a verdict.
Try It Yourself
The best way to internalise any of this is to put it into practice. Each exercise below presents a realistic scenario — take a moment to think before revealing the answer.
Fix the prompt
Someone needs help with an upcoming presentation and sends this to an AI tool:
What's missing? How would you improve it?
Original — missing
- Who the audience is
- What the topic or goal is
- How long it needs to be
- Any tone or format guidance
Improved — gives AI
- Audience: your class and teacher
- Topic: your science fair project results
- Length: 5 minutes
- A structure and a tone to aim for
The AI now has the audience, the goal, the structure, and the tone. It can give you something genuinely useful rather than a generic template.
Spot the hallucination
You ask AI: "When was Slack founded, and who created it?" It responds:
Is everything here accurate? What would you check before sharing this?
Slack was acquired by Salesforce — not Microsoft — in 2021, for approximately $27.7 billion. Microsoft is a direct competitor to Slack via Microsoft Teams. The rest of the response is broadly accurate.
This is a classic hallucination pattern: most of the answer is correct, which builds false confidence. The error is specific, plausible-sounding, and exactly the kind of thing that gets copied and forwarded without a second look.
Any specific claim — names, dates, figures, company events — is worth a 30-second check before you share it. The more specific it sounds, the more suspicious you should be.
Plan the approach
Your manager asks you to produce a short competitor analysis comparing your main product to three rivals — using AI to help. Before you open a single conversation, how would you approach it?
Think through the steps involved, then reveal a suggested approach.
- Open a fresh conversation. Tell AI the goal and your context — what your product does, who the three rivals are, and what the analysis is for.
- Ask AI to suggest a structure before you start. Agree on the headings (e.g. pricing, key features, target audience, weaknesses).
- Research each competitor in a separate conversation — one per chat. Ask what AI knows, and note anything surprising to verify later.
- Bring the findings into a new conversation and paste in your notes. Ask AI to compare the competitors side by side against your agreed structure.
- Ask AI to draft a summary. Review it — add anything AI couldn't know (internal context, recent news, your own experience of the rivals).
- Verify any specific claims — prices, features, company details — before the report goes anywhere.
Breaking the task into steps keeps each conversation focused. AI produces significantly better output when it has one clear job, rather than trying to do everything at once.
What shouldn't be in there?
Someone is about to send this to a public AI tool to get help writing a performance review:
What should be removed or changed before sending?
- Full name — "Sarah Johnson" is personal data. Use "a team member" instead.
- Employee ID — EMP-4471 is an identifiable internal record. Remove it entirely.
- Exact figures tied to an individual — the specific sales numbers combined with a name create a detailed personal record. Use a percentage shortfall if the context is needed.
- Disciplinary details — verbal warnings are sensitive HR data that should never go into an external tool without explicit organisational approval.
Before sending anything to an external AI tool, ask: if this conversation were made public, would anyone be identifiable or embarrassed? If yes, anonymise it first.