ChatGPT at work: how to use it in Armenian

Choose a task you can check

For a first attempt, use material you understand: five notes from a simple meeting, for example. That makes it easier to spot a missing deadline or an invented decision. A large analysis of an unfamiliar market is a less useful starting point because confident wording can conceal an error you may not recognize.

Open the official ChatGPT website. Sign in with your account if the service asks you to. For these exercises, use a regular chat, enter the task, and send it. You do not need to connect workplace services or build an automation. OpenAI's guide distinguishes conversational use from its mode for larger tasks. OpenAI web guide.

In our October 2, 2026 test, the temporary web chat had a Russian interface, while both the request and response were in Armenian. Your account may look different. This test did not establish feature availability across accounts, free-plan limits, or an overall measure of Armenian-language quality.

A useful starting instruction is: “Answer in Armenian. Use only the facts below. Do not fill gaps with assumptions.” Add your actual task after that. If the result is too long, ask for a shorter version that preserves the facts. This article is about getting started with practical work; systematic prompt improvement is a separate topic.

ChatGPT

What we tested

Every situation below is fictional. These are not real customer messages or MekStep meeting records. We ran three tasks together in one request through the ChatGPT web interface on October 2, 2026. The visible mode label was “6 Pro”; an internal model version identifier was not shown.

The original input and response were saved. We distinguish what the tool produced from our editorial assessment. This is one small exercise, not a comparison of models or a measurement of time saved. The English explanations describe the Armenian test; they do not imply that a separate English run took place.

Task one: draft a message

The fictional situation: a partner has sent product photographs, but two images do not have sufficient resolution. A deadline for replacement files has not been agreed.

The Armenian instruction was: “Գրի՛ր գործընկերոջն ուղղված կարճ, քաղաքավարի նամակ՝ 4–5 նախադասությամբ։ Փաստեր՝ ապրանքի նկարների փաթեթը ստացել ենք, բայց երկու նկարի լուծաչափը չի բավարարում, խնդրում ենք ուղարկել ավելի բարձր որակի տարբերակները։ Ուղարկման ժամկետ դեռ համաձայնեցված չէ։ Մի հորինիր անուններ կամ ստորագրություն”։

In English, this asks for a short, polite message of four or five sentences based on the supplied facts, without inventing names or a signature.

The response thanked the partner, confirmed receipt, requested better-quality versions of the two images, and kept the statement that the delivery date had not been agreed. It did not add a name or a new deadline.

However, it was not ready to send without further work. The recipient would still need to know which two files were affected. The sender should identify the filenames. If a deadline matters, agree it with the partner rather than asking the model to choose a convenient date.

Here, the useful output is the structure of the message. Missing information remains a real work task. A good draft makes that gap easier to see; it does not automatically close it.

Task two: summarize meeting notes

The fictional notes say that the team decided to update a product description. The editor will prepare a draft on October 6. The designer will check image dimensions, but no deadline has been set. The budget was discussed but not approved. The date for sending the material to the customer has not been agreed.

We asked ChatGPT to divide the notes into decisions, next steps, and open questions without adding facts. It put the decision to update the description in the first section and the editor's and designer's work in the next steps. Unresolved dates and the unapproved budget remained open questions.

Check a summary like this against the original sentences. “Budget discussed” is different from “budget approved.” Preparing a draft is different from sending it to a customer. October 6 belongs to the editor's draft; it does not automatically apply to every action.

If real notes contain two conflicting deadlines, ask the tool to show the conflict. Asking it to pick the “most likely” one may hide the very question the team needs to resolve. Keeping uncertainty visible is often more useful than producing a neat but misleading summary.

Task three: organize an action list

Using the same notes, we requested a table with Task, Owner, Deadline, and Status fields. Missing values were to be marked “not specified.”

The editor's date remained October 6. No invented deadline appeared for the designer, and the budget remained unapproved. A completely filled table should not create the appearance of agreements that never happened.

There was still an editorial issue to resolve: the response listed both the overall description update and the preparation of its draft as separate rows. Before treating them as independent assignments, check whether one is the goal and the other a step toward it. Organizing text is not the same as creating the team's final work plan.

In a real task list, confirm each owner and deadline. Keep open questions visible. A generated table does not, by itself, mean that tasks have been placed in a calendar, sent to colleagues, or accepted by them.

Five questions before using a response

You can ask ChatGPT to flag unverified parts of its own response. That second pass does not replace the source material. OpenAI's guidance recommends stating important boundaries and reviewing the result yourself before use. OpenAI prompting guidance.

• Has it added a name, number, or date that was absent from the source?
• Has it preserved the status: proposed, discussed, approved, or completed?
• Does each fact belong to the correct action and person?
• Is the Armenian natural, including terminology, forms of address, and sentence structure?
• What must a person still clarify before sending or acting?

Choose input data carefully

Use fictional notes for practice. Do not paste an entire real customer conversation when the task only needs one sentence. Remove personal information and details you are not authorized to share. Changing a name does not always anonymize a story: a company, an unusual event, and a specific amount may still identify someone.

Do not submit passwords, identity information, unpublished contracts, or confidential customer details. Follow your organization's AI rules and use an approved account. These are editorial recommendations for handling information, not a description of every data-retention arrangement.

Check the data settings that apply to your environment. OpenAI's documentation points to ChatGPT data controls in the web and mobile settings. Do not treat the name of a special chat mode, or deleting a conversation, as permission to provide confidential material. Documentation on data controls.

A useful next exercise

Write five fictional notes about a situation similar to your work. Ask for one output, compare it with the notes, and record what you accepted, changed, and left unresolved. The goal is to learn a controlled work step, rather than to obtain perfect text on the first attempt.

If you work with customers, it can also help to understand the sales manager's role: a message is only one part of communication. For more structured learning, review the current AI from Zero course and compare its programme with your actual tasks.

Frequently asked questions