Your first prompt: instructing AI

You know how a model works inside. Time to take the controls — the skill is called prompting, and it's near the top of the most wanted skills of 2026 (remember the lesson on the job market).

Why the prompt decides

A prompt is the text you send to an AI. The model doesn't read minds: everything it knows about your task is your text. A weak prompt makes it guess (and fill in what's likely — you know how that ends). A strong one aims the prediction straight at the target.

The RTCF formula

1. R — Role. Who should the model "be"? "You are an experienced HR specialist". The role is a filter: which part of its knowledge to answer from, and in what tone. Remember the model learned from the whole internet? The role picks the right shelf.
2. T — Task. What exactly to do. One verb beats five: "write a plan", "find the errors", "cut it in half".
3. C — Context. What the model needs to know about the situation: who it's for, what's already done, what the constraints are. The rule: what would a new colleague ask if you handed them this task? — put that in.
4. F — Format. What the answer should look like: "a table", "5 bullet points", "under 100 words".

See for yourself

❌ "Write something about remote work" — the model is guessing: an article? a tweet? for whom? for or against?

✅ "You are the editor of a company blog (R). Write the opening of an article about moving to remote work (T) for a 50-person IT company; friendly tone, the readers are the employees themselves (C). 3–4 sentences, plain language, no corporate jargon (F)".

The first prompt gets you the internet's average. The second gets you exactly what you need. The mechanism under the hood is identical — the only difference is your text.

Iteration: a prompt is a conversation

The first answer is a draft, not a verdict. "Make it simpler", "add an example", "different tone", "cut it in half" — the model remembers the conversation and revises. Two or three passes almost always beat the first. What sets professionals apart isn't a perfect first prompt, it's how fast they iterate.

Three beginner mistakes (save yourself weeks)

• Telegraph-style shorthand: "email client discount" — the model isn't a search engine, it needs sentences. Write the way you'd explain the task to a person.
• Everything in one request: "write the article, come up with headlines, make a post and translate it" — quality drops at every join. One task → one request; the next one goes in the next message, and the context carries over.
• Giving up after the first answer: a beginner gets a mediocre answer and decides "AI can't do this". That's the moment a pro is just getting started: clarifying, adding context, asking for options.

Seven ready prompts — they work as they are

Nothing to invent: copy the whole prompt, paste it into Gemini and look at the answer. The text inside is already filled in — you'll swap in your own later. The examples deliberately span different fields, not just office ones — take the nearest to your line of work.

1. "You are an editor. Rewrite this email: half the length, friendlier, keep every fact and date. Email: 'Dear Mr Bennett, we hereby notify you that the equipment delivery scheduled for 14 March has been postponed to 21 March due to a delay at customs. We apologise for any inconvenience caused.'"
2. "You are a planner. Lay out my day hour by hour: start at 9:00, meetings at 12:00 and 16:00, an hour for a report, gym in the evening. Leave slack between things. Answer as an hour-by-hour list"
3. "Explain in plain words what an interest rate is, to someone hearing it for the first time: 3 sentences + one everyday analogy"
4. "You are a critic. Here's my plan: 'start a newsletter about productivity, publish 3 times a week, add advertising after a month'. Find 3 weak spots and ask me 3 questions I'm missing"
5. "Give me 15 name ideas for a small seaside coffee shop. Group them by style (cosy, bold, minimal), and one line each on why it would work"
6. "You are the owner of a small restaurant. Describe a dish for the menu — seafood pasta — appetisingly and briefly: 2–3 lines, no grand claims, so people want to order it. Give me three options to choose from"
7. "You are an experienced teacher. Explain fractions to a 9-year-old who's scared of them: plain words, through a pizza example, 4–5 sentences, and finish with one easy question to check they got it"

Notice: every one has a role or a format, and usually both. That's RTCF in its natural habitat. And look at the fourth prompt — "find the weak spots in my plan": almost no beginner uses AI-as-critic, and it's often more useful than AI-as-doer. A good question about your own plan is worth more than someone else's finished text. The last two — the restaurant and the school lesson — are a reminder that the formula works far outside the office. Run a few? Now take any of them and swap the text inside the quotes for yours — your task, your email, your plan.

When you don't need the formula

"What's the capital of Australia?" needs no role and no format. RTCF is for tasks where the result matters and one phrase won't describe it. Simple questions, simple prompts.

Do this now

<