Your first useful task with ChatGPT or Claude

You do not need programming to make an AI assistant useful at work. Start by turning rough notes into a weekly update: you know the facts and can judge the result. This guide works in ordinary ChatGPT or Claude chat, in a browser or their desktop apps. Choose one. The fictional example is a complete practice task and an editorial demonstration, not a measured product test. You remain responsible for the final message.

Before you start

• Access to ChatGPT or Claude through a route your organization permits. A browser is sufficient.
• No terminal, programming, API key, paid API or account connection is needed.
• A document for the approved text. Check organizational data rules before replacing the fictional brief.

Tools: ChatGPT, Claude. Use an assistant approved by your organization. This text-based first version does not require new integrations; account features and data policies still apply.

The workflow, step by step

1. Open a chat and choose one assistant

   For the simplest start, open chatgpt.com or claude.ai in your browser and follow the service’s sign-in flow. Use an approved work account if provided. Start a new conversation. In the current ChatGPT interface, select Chat; Work and Codex are other options with different purposes. In Claude, use Chat rather than Code or Cowork. A desktop app is optional. Official documentation lists ChatGPT for macOS, Windows and Linux, and Claude for macOS, Windows and Linux beta. Download only through the provider’s official route. Features differ by platform. This exercise needs only a text conversation, so stay in the browser if installation adds friction.

2. Choose a small job with a visible finish

   Ask for one deliverable you can inspect: a weekly update for four colleagues. A useful finish is concrete: the draft records completed work, identifies the open item and names the next action without inventing a deadline. Read the fictional practice brief below. Notice the difference between a completed check and an unresolved record. Twenty-four supplier records were checked; three still need a contact. Maya’s request is planned work. The review meeting has no date. Preserve those distinctions while improving the wording.

3. Give the assistant a brief, a format and boundaries

   A prompt is simply your request. A good first prompt says what to produce, who will read it, which facts to use and what must stay uncertain. Copy the complete starter prompt below into the conversation and submit it. The brief is included. The requested headings make the answer easy to scan, while the word range keeps it short. The separate review note stops a missing detail from disappearing into polished prose. These are writing instructions, not a guarantee that the model will obey every rule. Check the answer yourself; no advanced setting is needed.

4. Read the answer against the brief

   Keep the original notes beside the reply. Check each number, name and statement of status. Has the assistant said that all supplier information is complete? Has it invented Friday as the meeting day? Has it claimed Maya already contacted anyone? Correct any such changes. Then check usefulness: would a colleague understand what happened and what remains open? A confident tone does not establish accuracy. The illustrative expected result below shows one acceptable structure, but matching its wording is unnecessary. Judge facts before readability; verify the assistant’s claims yourself.

5. Improve one thing through a follow-up

   Continue in the same conversation with a specific revision request. For example: “Keep all facts unchanged. Make the opening warmer and the Next section more direct. Keep the review date explicitly unconfirmed.” If it made a factual error, name the error and repeat the correct fact from your brief. Read the revised version again; improvements in tone can introduce new mistakes. If you want to compare providers, start a fresh chat in the other service and paste the same complete prompt. Compare both drafts against the brief. Agreement does not verify facts; one preferred draft does not establish a universally better product.

6. Save the reviewed draft and repeat deliberately

   Copy the approved text into your usual document or message editor. Confirm that the missing meeting date is still visible before you decide whether the update is ready to share. Sending is a separate human action in this exercise. Save the briefing template so next week you can replace the facts and audience instead of beginning from a blank request. You may encounter usage limits or different controls on different plans. Claude’s desktop documentation lists Chat on its free and paid plans; that does not make Code or Cowork free. Do not purchase an upgrade just to follow this text exercise. If access is unavailable, keep the template and resume when your approved service is available.

Try it with a sample

Use this synthetic input to see the distinctions your output should preserve.

What a good result should preserve

A prompt you can adapt

Help me turn the fictional brief below into a weekly internal update. Use only these facts. Write in plain, friendly English for a small operations team.

Return:
1. A subject line.
2. An update of 100–140 words under Done, Still open and Next.
3. A separate “Check before sending” note naming missing information.

Keep the numbers and owner exactly as supplied. Make the unknown meeting date explicit: “Review meeting date: to be confirmed.” Do not invent dates, deadlines, approvals, savings or work already completed. Do not turn a planned action into a finished one. If a fact is missing, identify it separately rather than guessing. Draft text only; do not send, schedule or access other tools.

BRIEF:
FICTIONAL PRACTICE BRIEF: no real business or customer data.
Audience: the four-person operations team at Cedar Lane Supplies.
Purpose: weekly internal update for the week ending 16 October 2026.
Completed: reorganized the stockroom shelves; checked 24 supplier records.
Open: three supplier records still lack a delivery contact.
Owner: Maya will ask the suppliers for those contacts.
Known next step: compare two delivery options after the contacts arrive.
Unknown: the review meeting date has not been agreed.
Boundary: draft only; nothing has been sent or scheduled.

Check the result

• Preserve 24 checked records and three missing contacts; do not imply every record is complete.
• Keep Maya’s supplier request and the delivery comparison as future work.
• Leave the meeting date unconfirmed; the reporting week supplies no meeting date or deadline.
• Use the requested headings and separate review note; check readability and length.
• Compare against the original brief for invented savings, approvals or completed actions.
• Use fictional text only for practice; no sending or account connections are required.

What to keep in mind

This researched guide uses fictional input and an illustrative draft. It claims no recorded product result, verified savings, installation or live-account test. Availability, limits and interface labels can change. Chat can share an app with agent modes; chatting does not imply coding or computer-control access. Prompt instructions do not enforce permissions. Follow organizational service and data rules for real work.

Sources & review notes

These sources support the tool context and review principles. The workflow, prompt, and synthetic example are original editorial material.

• OpenAI: Quickstart

  Explains web and desktop entry points and the current Chat, Work and Codex choices. This exercise selects Chat and uses pasted text.

• OpenAI: ChatGPT on the web

  Official web entry point and conversational task guidance. No particular model, plan entitlement or usage allowance is required by the exercise.

• Anthropic: Get started with Claude

  Documents browser access, conversational prompts, refinement and usage limits. Its brief platform overview omits Linux; the dedicated installation page is used for that detail.

• Anthropic: Install Claude Desktop

  Dedicated current platform and plan matrix. Linux is beta. No computer-control, extension, Cowork or Code setup is part of this beginner exercise.