How to Talk to Your AI: A Guide for Normal People
No coding knowledge required. No jargon. Just practical things that make AI actually useful.
First, the thing everyone gets wrong
Most people treat AI like a search engine: type a vague question, hope for a magic answer, get frustrated when it’s generic. That’s not how to use it.
Think of it more like a very capable new colleague who has just started. They’re smart, well-read, and fast. But they don’t know your situation, your preferences, your history, or what you actually mean by “can you sort this out.” The more you brief them, the better the work.
The three things that make the biggest difference
1. Say what you actually want
Not a gesture toward it. The actual thing.
- Weak: “Can you help me with this email?”
- Strong: “I need to decline a meeting invitation from a client without damaging the relationship. We may want to work together later. Keep it warm but firm, under 100 words.”
The second version will get you something you can actually send. The first will get you a template that sounds like no human ever wrote it.
2. Say who it’s for
Writing for your boss, your mum, and a journalist are three completely different tasks — even if the subject is the same. AI doesn’t know which one you mean unless you say.
“Write this as if the reader is a busy person who already knows the basics and just needs the key decision.”
3. Give an example of what good looks like
If you want a specific tone or style, paste an example of it. “Write this in my voice” means nothing. “Write this in my voice — here’s a paragraph I wrote last week: [paste it]” means everything.
Things to include that most people forget
- Who will read this — their background, their mood, their relationship to you
- What format you want — bullet list? Short paragraph? Table? Email? Just say
- What constraints exist — word count, things to avoid, platform it’s going on
- What you’ve already tried — saves you getting the same suggestion twice
- What “done” looks like — rough draft for you to edit, or ready to send?
- What it should optimise for — speed, accuracy, brevity, creativity, or caution all pull the answer in different directions
What AI is actually doing (without the jargon)
It doesn’t “think.” It predicts what the most useful next word is, based on everything it’s been trained on and everything you’ve said in the conversation. This means:
- It can be confidently wrong. It won’t always say “I don’t know.” It will sometimes fill gaps with plausible-sounding nonsense. Anything that matters (names, statistics, facts, medical or legal specifics) — check it independently.
- It has no memory between conversations. Yesterday’s chat is gone. Every new conversation is a blank slate. If context matters, paste it in at the start.
- The quality of your input directly shapes the quality of the output. Generic in, generic out. This is the main lever you control.
When something goes wrong
If the answer is off: Don’t just try the same prompt again. Tell it what specifically was wrong. “That’s too formal — I need something warmer” or “You missed the point, the real issue is X” will fix it faster than starting over.
If you’re going round in circles: Start a new conversation. Paste a clean brief at the top. Sometimes a conversation gets so tangled that starting fresh is faster than untangling.
If it’s not saying what you think it should: You probably have a strong instinct about the right answer — share it. “I think the answer might be X, but I’m not sure — can you help me think through it?” is a much more productive prompt than waiting for it to independently arrive at your answer.
Green flags (you’re doing it well)
- You get something usable on the second or third exchange, not the first
- You’re refining rather than restarting
- You treat the first response as a draft, not a verdict
- You push back when something’s wrong, with reasons
Red flags (something to adjust)
- You keep getting generic responses → your prompts are probably too vague
- You’ve rewritten the same prompt five times → tell it what was wrong, not just try again
- You’re accepting outputs you privately think are bad → say so, it won’t be offended
- You’re pasting walls of text with no instruction → tell it what you want it to do with them
The one rule
Treat it like a smart colleague, not a slot machine. It’s not about finding the magic combination of words that produces a perfect output. It’s a conversation. Brief it well, react to what it gives you, redirect when needed. That’s it.
Sources: All contributors to this project