triox informatika

October 26, 2026 · Triox Informatika · 6 min read

How to ask well—the quality of the answer depends on the question

Most disappointment with AI comes not from the tool but from the question. Seven simple principles, with plenty of examples, that get a better answer out of any assistant—ChatGPT, Claude, Gemini, Copilot. Not technical tricks, but the way you would explain a task to a new colleague.

In brief

  • A question put to AI is not a search term, it is an assignment. Just as you would not say “letter” to a new colleague, but tell them to whom, about what, and in what tone.
  • The seven principles: context, goal, format, example, role, steps, feedback. Each is one sentence, and together they multiply the quality of the answers.
  • The first answer is rarely final. Have a conversation—the second and third round bring the good result.
  • The biggest mistake is missing context: the AI does not know your company, your customer or the situation unless you tell it.

“I tried it, but it wasn’t any good.” We hear this sentence very often, and when we ask what the person typed in, it is almost always the same story: two words, half a sentence, a question no human could have answered well either. AI assistants give the most when we talk to them the way we would talk to a smart new colleague who does not know our company. That is what we learn in this part—and it applies to every tool covered in this series.

The basic idea—not a search, but an assignment

For twenty years we typed two words into Google: “contract termination template.” This habit works against us with AI. The assistant does not look for results; it carries out a task, and for that it needs to know what the task is. Compare:

Weak: “Termination letter.”

Good: “Write a polite but firm termination letter to our office cleaning company. We have worked with them for three years; the quality has declined in recent months, and we have raised it twice. The contract sets a 30-day notice period. The tone should be fair, not accusatory, and leave the door open for working together in the future. Half a page.”

Writing the second question takes a minute. The answer is usable. The template letter you got for the first question would have had to be rewritten anyway.

The seven principles

1. Context—say what the situation is

This is the most important element, and the one most often left out. The AI does not know what your company does, who the client is, or what has happened so far. Two or three sentences of context double the quality of the answer.

“We are a trading company; we distribute industrial parts, mostly to manufacturing clients. One of our largest customers is three months late with payment, but they are an important partner and we do not want to lose them.”

If a tool supports projects or memory (Claude and ChatGPT do), you only need to give this context once, and every later conversation knows it.

2. Goal—say what you want to achieve

Not what it should do, but what it is for. Instead of “Write a summary”: “Write a summary that the managing director can read in two minutes and then decide whether to approve the investment.” Knowing the goal, the AI decides on its own what to highlight and what to leave out.

3. Format—say what it should look like

Length, structure, language, tone. “Half a page.” “A bulleted list, five points at most.” “A table with columns: risk, likelihood, action.” “Formal, addressing the reader as ‘you’ in the polite form.” “In English, but in simple language, for a non-native reader.” If you do not say, the AI uses its own default—and that is often too long.

4. Example—show it a good one

If you have an earlier letter, quote or report you liked, paste it in: “Write it in this style and structure.” One good example is worth more than ten sentences describing the style. This is where the company’s own templates and earlier materials become truly valuable.

5. Role—tell it whose head to think with

“Review this quote through the eyes of an experienced finance director and tell me what is missing from it.” “Read this letter in the role of a demanding client and tell me what would bother you about it.” The role is not a game: the AI really does bring up different considerations because of it. It is especially useful for checking: have your own material reviewed by a critical reader before it goes out.

6. Steps—break a complex task down

If a task has several parts, do not ask for all of them at once. “First, summarize the document. Then list the risks. Finally, suggest three questions we should ask the client.” Step by step is more precise, and you can also see where the reasoning departs from what you expected. Today’s tools often “think” in steps on their own, but breaking the task down is still the best tool in your hands.

7. Feedback—have a conversation, do not ask just once

The first answer is raw material. “Good, but make it shorter, and leave out the second paragraph.” “This is too formal; make it more direct.” “Do not change the numbers, only the wording.” “Now the same in English.” Every round refines it. Experienced users rarely settle for the first answer, and not because it was bad, but because the second one is better.

A few common situations—with sample questions

A difficult email. “Here is an email from a dissatisfied customer [pasted]. Write a reply that acknowledges the mistake, does not deflect, offers a concrete solution, and does not promise anything we cannot keep. Formal, short. The solution: we will ship the next delivery free of charge.”

After a meeting. “Here are my notes from today’s meeting [pasted]. Turn them into a short memo for the participants: decisions, action items with an owner and a deadline, open questions. Where the notes are unclear, mark it with a question mark; do not make it up.”

A contract. “Read this contract [uploaded] and list the clauses that are risky or one-sided for us as the customer. For each one, quote the text and explain why it is a problem. If something we would expect in a contract like this is missing, note that too.”

Excel. “This spreadsheet [uploaded] has monthly revenue by client. Which three clients saw the largest drop in revenue compared with the previous year? Show it in a table, and make a chart of the ten largest clients.”

Learning. “Explain what NIS2 is to a company director who is not an IT person. Three paragraphs, with examples. Then tell me the three things they should look at first in their own company.”

A decision. “We received two quotes for the same service [both pasted]. Compare them in a table: price, term, what is included, what is not, risks. Do not tell me which one to choose, only how they differ.”

What not to do

  • Do not assume it knows what you know. The most common mistake. What is obvious to you—the company, the customer, the history—the AI does not know.
  • Do not accept numbers without checking. Dates, amounts, statute numbers, references: always look them up.
  • Do not put in anything you would not hand to an outside service provider. Passwords, personal data and confidential client data only in a company account, and even there with care. Our article on shadow AI was about this.
  • Do not have it write the final version. The AI prepares; you are responsible. This has come up in every part of this series, and it applies here too.
  • Do not give up at the first bad answer. Say what is wrong with it. Most cases of “it doesn’t work” are really “I didn’t say what I wanted.”

One sentence worth remembering

If you take one thing away from all this: ask the way you would explain the task to a smart new colleague on their first day. What the situation is, what you want to achieve, what it should look like, whether there is an example, whose eyes to look through, and talk it over if it did not turn out right. That is all. Every other technical detail follows from this.

In the final part of the series we put the picture together: which tool you or your company should choose, what to base the decision on, and how to start so that your colleagues actually use it—and your data stays safe.

Would you like your colleagues to ask well, too?

If you have a question, a request or a specific need related to this topic, write to us. A two-hour internal workshop with your own tasks is worth more than any guide.

Write to us

Triox Informatika Kft.

← Back to the blog