How to use AI effectively — a practical guide for talking to an AI

In brief

Talking to an AI is not like doing a Google search. Google tolerates imprecision — it guesses what you mean. An AI, on the other hand, takes your wording literally. If you are vague, it answers vaguely. If you are specific, it can be too.

What you send to an AI is called a prompt: it is simply the question or instruction you type. Formulating that prompt well is the skill that makes all the difference — and it can be learned in a few minutes.

Think of a new temp who shows up in the morning without knowing what you want. If you tell them “do something,” they’ll do something — not necessarily what you expected. If you tell them “draft this email to a chilly client, conciliatory tone, three paragraphs,” they’ll do exactly that. Prompting is the same conversation, except your temp has read a hundred million books and zero clues about your context.

Be specific about what you want

The first principle is the simplest. A vague question produces a vague answer.

Compare these two formulations:

• “Explain the code.” → The AI may paraphrase, summarize, analyze the structure, or do all of the above. You have no idea what you will get.
• “Explain what this function does, line by line, assuming a reader who has never programmed.” → The response is constrained from the start.

Every detail you add narrows the space of possible responses. The AI does not have access to your intention, your context, or your implicit expectations. Anything you leave unsaid, it invents or ignores.

Good habit: specify the expected role, the desired format, and the key constraints.

In short: a vague prompt is like ordering with no menu — the AI brings whatever is at hand, not what you wanted. Three quarters of frustrations come from this.

Give context

An AI does not remember you between conversations. It does not know who you are, what project you are working on, or who you are writing for.

All of that context must be provided in the message itself. Compare:

• “Improve this text.” → Improve how? For whom? In what register?
• “Improve this text for a professional email to a client I do not know. Formal tone, 3 paragraphs maximum.” → The AI knows exactly where to go.

Providing context upfront avoids unnecessary back-and-forth and improves the response on the first try.

Show examples before asking

When you want a precise output format — a table, a list in a particular style, short answers — describing that format in words is often less effective than showing two examples.

The AI is very good at detecting a pattern and reproducing it. Take advantage of this.

For instance, if you want to classify customer reviews, you can write:

Input: "The delivery was fast."
Output: POSITIVE

Input: "The product arrived broken."
Output: NEGATIVE

Input: "The delivery was fast but the packaging was damaged."
Output:

The AI naturally completes according to the pattern you showed it. More robust than a long prose description, especially for non-standard formats.

In short: describing a format produces text; showing a format produces a copier. For structured output, two examples beat ten lines of explanation.

Ask it to think step by step

On complex tasks — calculations, reasoning, structured analyses — the AI can jump too quickly to a conclusion and get it wrong. There is a simple trick: ask it to state its reasoning steps before answering.

Adding “Think step by step” at the end of a question is often enough to markedly improve the result.

A train departs at 2:00 PM and arrives at 5:30 PM. It stops for 20 minutes along the way.
What is the actual travel time excluding stops?
Think step by step.

Why this works: by forcing the AI to make its reasoning explicit, you reduce jump errors — premature conclusions drawn without checking intermediate steps.

In short: asking for step-by-step reasoning forces the AI to write its work on the board instead of blurting out the conclusion. On a calculation, the rate of correct answers can climb from 30 to 70%.

Assign it a role

Telling the AI what role it should play changes the register and depth of its responses. This is not a magic trick: the AI was trained on texts produced by experts in varied contexts. Specifying a role steers it toward the right texts in its memory.

You are a general practitioner explaining to a patient with no medical background.
Explain what blood pressure is and why it can be a problem.

The role constrains the level of abstraction, vocabulary, and angle of approach. Particularly useful when you want the same information presented differently depending on the audience.

Iterate — the first attempt is rarely the best

The first prompt is rarely optimal. That is normal. Prompting is an empirical process: you try, you observe what is missing or excessive, you adjust one thing at a time.

This is no different from writing — the first draft shows you what is missing.

Some useful reflexes:

• If the response is too long → add “in 3 sentences maximum”
• If it is too vague → add format or context constraints
• If it is at the wrong level → assign a different role

Key takeaways

• Be specific. Anything you do not specify, the AI interprets according to its own default settings.
• Give context. It knows nothing about you — tell it what it needs to help you effectively.
• Show examples. For a precise format, two examples are worth more than a long description.
• Ask the AI to think step by step. On complex tasks, this simple instruction markedly improves accuracy.
• Iterate. There is no universal formula — adjusting one variable at a time is the right method.