Using AI tools responsibly at work

In many offices today, someone asks an AI chat assistant for a draft letter, someone else for a report outline or an explanation of an Excel formula. These tools genuinely save time, but used carelessly they can spread wrong information, expose confidential data or cause problems with honesty. In this guide we look at how to use AI chat assistants at work in a way that is both useful and safe: which tasks they suit, how to build a good request (prompt), how “bad” and “good” prompts differ, how to check the output, which data you should never enter, the rules of authorship and honesty, the situations where you need a specialist rather than AI, and how to fit AI into your workflow properly. At the end you will find common mistakes, a practice exercise and a checklist.

If you do not yet have a general idea of how artificial intelligence works, read What is artificial intelligence first; this article focuses on practice.

What an AI chat assistant is good for at work

An AI chat assistant is a program that answers your written question with text. ChatGPT, Claude, Gemini, Copilot and similar services are examples of such tools; we do not recommend any particular one, and the rules apply to all of them. They usually run in a browser or a phone app: you type a request into the box at the bottom, send it and get a reply. For a new topic it is a good habit to open a fresh conversation with the “New chat” button, so the context of the previous conversation does not mix with the new task.

The tasks where AI helps most at work:

• Draft letters. For example, a polite letter to a partner organisation about moving a meeting to another day. You provide the main points and the AI produces a first version. Business correspondence is covered in more detail in our article on email etiquette.
• Shortening text. Turning long meeting notes into a five-point summary, or pulling out the main requirements of an official letter.
• Gathering ideas. A list of topics for a careers event for young people in the mahalla, questions for a survey or an outline for a presentation.
• Draft translations. Turning a letter in Uzbek into a draft in Russian or English. It is only a draft: an official document should be checked by someone who knows the language well.
• Explaining formulas. Asking for a plain-language explanation of what a formula such as =IF(F2<10;“Order”;“Enough”) does in a spreadsheet a colleague left behind, or asking it to suggest the formula you need.
• Learning plans. A plan for Dilshod, who has just moved to a new role: “learn the basics of project management in two weeks”, with topics for each day, exercises and self-check questions.

The general rule: AI is at its best for drafts, structure and explanations. The final decision, the facts and the responsibility always stay with you.

Writing a good prompt: five elements

AI cannot read your mind. The quality of the answer depends largely on how precise the request is. A good prompt usually has five elements:

1. Context: who you are, what the situation is, who the text is for. For example: “I am an administrator at a training centre in Samarkand, and the letter will go to parents.”
2. Task: exactly what to do: “write”, “shorten”, “explain”, “suggest three options”.
3. Format: what the result should look like: a list, a table, a letter of no more than 5 sentences, an outline with headings.
4. Constraints: what must be avoided or what limits apply: “in a formal style”, “without technical jargon”, “no more than 150 words”, “in Uzbek, Latin script”.
5. Example: a sample of the style or result you want. If you have a good letter from before, you can give part of it, without personal data, as a model.

You do not have to spell out every element every time, but if the answer does not satisfy you, first check which element is missing. Also, keep the conversation going after the first reply: follow-ups such as “make it shorter”, “soften the tone” or “remove the third point” improve the result quickly.

Pairs of bad and good prompts

The examples below are illustrative. They show how a difference in the prompt affects the result.

Pair 1: a draft letter

Bad prompt:

Write a letter.

Good prompt:

I am the office manager at a small IT company. We need a letter to a partner organisation asking to move Thursday’s meeting to next week, because our director is on a business trip.
Suggest two convenient days.
Style: polite and formal, 6–8 sentences, in Uzbek.

In the first case the AI has to guess what to write about, to whom and in what style. In the second you get an almost finished draft.

Pair 2: shortening text

Bad prompt:

Shorten this.

Good prompt:

Turn the meeting notes below into a summary for the manager: 5 points, one sentence each. Separately, list “Who does what and by when”. Do not add any information that is not in the text.

The last sentence matters: it discourages the AI from inventing facts.

Pair 3: explaining a formula

Bad prompt:

The formula doesn’t work.

Good prompt:

In Google Sheets the formula =VLOOKUP(A2;Products!A:B;2;FALSE) returns a #N/A error. Column A holds product codes; the “Products” sheet holds codes and product names. What usually causes this error? Describe the steps to check it in plain language.

Here the task, the context and the format you need are all clear. Notice that you describe only the structure of the spreadsheet, not the spreadsheet itself: real customer data is not needed.

Checking the output: facts, figures, sources

An AI answer looks convincing: the grammar is correct and the tone is confident. That does not mean it is right. AI sometimes produces information that sounds plausible but is made up; this is called a hallucination. It may calmly “cite” a book that does not exist, a wrong date, an invented article of law or a broken link.

Check every result in this order:

1. Facts. Names, dates, organisation names, events: check each one against an official or reliable source. For public services and legislation, rely only on official websites.
2. Figures. Redo the calculations yourself or in a spreadsheet. AI is good with text but can make arithmetic mistakes.
3. Sources. If the AI gives a source or a link, open it: does the page exist, and does it really say that?
4. Logic. Does the answer actually address your question? Are there internal contradictions?
5. Tone and context. Does the letter suit your organisation’s style, the recipient and the situation? Does the text in Uzbek or Russian sound unnatural, like a literal translation?

A useful technique: ask the AI, “Which parts of this answer should I check separately?” It sometimes points out its own weak spots, but that does not replace checking.

Privacy: what not to enter

Whatever you type to a chat assistant is sent to the servers of the company that runs the service. Depending on the settings, it may be stored and used to improve the service. So before you write a prompt, ask yourself: “Would it be a problem if this text ended up in a stranger’s hands?”

Never enter:

• Customers’ and employees’ personal data: a full name together with passport details, personal identification number (PINFL), phone number, address, medical information or salary.
• Passwords, access codes, bank card details and login keys for systems.
• Internal documents: contracts, financial reports, plans not yet announced, the organisation’s private correspondence.

If the task needs real text, anonymise it first: replace names with “Customer A” or “Employee B”, replace exact figures with made-up ones, and remove the company name. Often it is enough to describe the structure of the text.

Further rules:

• Follow your organisation’s rules. Some organisations ban AI or allow only approved tools. If you are not sure, ask your manager or the IT department.
• Review the settings. Many services have options in Settings related to chat history and how your data is used (for example, Data controls or Privacy). Read through them.
• Protect your account. Use a strong password and turn on two-step verification. There is more on this in our article on staying safe online.

Authorship and honesty

AI writes text quickly, but that text is not your experience, your thinking or your work. The rules of honesty:

• In your studies. Do not hand in a course assignment, essay or test answers written by AI as your own. Find out your institution’s rules on AI in advance. Asking AI to explain a topic so you understand it is one thing; copying and submitting a finished piece of work is quite another.
• In a CV or application. Letting AI write your whole CV or cover letter is risky: the text comes out generic and “lifeless”, it may list skills you do not have, and at the interview you will not be able to back them up. Write about your own experience, your achievements and why you want this particular job yourself. Ask AI only for help with structure, spelling or shortening. More on this: how to write a CV.
• At work. If AI made a significant contribution to a report or presentation and your organisation’s rules require it, say so openly. The person responsible for the result is not the AI but whoever signed or sent the document.
• Other people’s texts and images. Text from AI may turn out to be very similar to someone else’s work. For an official publication or your organisation’s website, rework the text in your own words.

When you need a specialist, not AI

AI can give general information, but in the following cases it cannot be the basis for a decision:

• Legal matters. Signing a contract, an employment dispute, inheritance or a property decision: consult a lawyer. AI may misread the law or give outdated information.
• Medical matters. Symptoms, medicines, diagnosis: only a doctor. At most, an AI answer can help you prepare questions for the doctor.
• Financial decisions. Taking out a loan, investing, tax calculations: an accountant or a financial adviser.
• Decisions about people’s futures. Hiring or dismissing staff, shortlisting candidates, disciplinary measures: do not hand such decisions to AI; they are made by the people responsible.

An illustrative example: Zarina works in accounts and asked an AI how to fill in a tax return. The answer looked convincing, but it mixed in requirements from a different year. The right approach is to use the AI answer as a list of questions and check them against an official source and with an experienced colleague.

Fitting AI into your workflow

A reliable way to work with AI is a three-stage process: draft → edit → check.

1. Draft. Get a first version with a five-element prompt. Remove personal data beforehand.
2. Edit. Put the text into your own words and your organisation’s style: remove excessive praise, generic phrases and ideas you did not mean to express. Add the specific details only you know.
3. Check. Verify facts, figures, names, dates and links. Before sending an important document, read it aloud or show it to a colleague.

Practical tips:

• Save your good prompts. Keep prompts you use often in a plain text file or your notes: “weekly report summary”, “formal letter draft” and so on. Next time you only change the context.
• Set a time limit. If you have spent 10–15 minutes rewriting a prompt and the result is still not good, it may be quicker to do the task yourself.
• Keep your own skills. AI makes writing letters easier, but you still need to be able to write, calculate and think on your own; otherwise you will not spot the mistakes.

Common mistakes

• Sending the first answer without checking. The most common mistake. A wrong date or name in a letter damages the organisation’s reputation.
• A prompt that is too short. A request like “Write a report” produces a generic result that does not fit your needs.
• Entering real personal data. Uploading a full customer list “to tidy it up” breaks privacy rules.
• Treating AI as a search engine or an encyclopaedia. It can give convincing but wrong information. Check facts against sources.
• Relying on an AI answer about law, medicine or money. These questions need a specialist.
• Passing off AI text as your own. Especially in studies and job applications: it is dishonest, and it quickly shows at an interview.
• Mixing everything in one conversation. Open a new conversation for each different task so the context does not get muddled.

Practice exercise and checklist

Do this exercise with a work task you genuinely need. The example is illustrative: Shahnoza works at a district training centre and needs to prepare a message for parents about new after-school clubs.

1. Choose a task. For example, a letter, a short announcement or a summary of meeting notes.
2. Clean the data. Replace names, phone numbers, addresses and internal reference numbers in the text with placeholders.
3. Write a bad prompt first. Send a request of one or two words and save the result.
4. Then write a good prompt. Add context, task, format, constraints and an example. Compare the two results.
5. Refine. Give at least two follow-up instructions about style, length or structure.
6. Edit. Rewrite the text in your own words and put the real details in place of the placeholders only on your own computer.
7. Check. Verify every fact, date and figure against a source.
8. Reflect. Note which prompt elements helped most and save the good prompt as a template.

Checklist before using AI

• The prompt contains no personal data of customers or employees, no passwords and no internal documents.
• My organisation allows AI, and I am using an approved tool.
• The prompt states the context, task, format and constraints clearly.
• The facts, figures and links in the output have been checked.
• The text has been edited into my own words, and my personal experience is written by me.
• No legal, medical or financial decision is based on an AI answer.
• I am ready to take responsibility for the final result.

An AI chat assistant is a powerful helper, but not the final author. Use it for drafts and explanations, protect confidentiality and check every result; that way it will speed up your work without harming your reputation.

If you would like to learn to use computers and modern digital tools systematically with a teacher, take a look at our association’s free computer literacy programme and our other training programmes. When you are ready, submit an application and our specialists will get in touch.