How to use ChatGPT at work: a practical tutorial

What is ChatGPT and what is it good for at work?

ChatGPT is a conversational assistant from OpenAI that handles drafting, summarising, analysing uploaded files and searching the web, all in one chat window. OpenAI's capabilities overview describes web search for recent information, file uploads for documents, spreadsheets and presentations, image understanding and generation, voice conversations, canvas for editing longer pieces alongside the assistant, and deep research for multi-step reports with citations.

At work that maps onto a handful of reliable jobs: turning rough notes into a structured draft, summarising a long document you are allowed to share, reworking tone for a different reader, building a first-pass table from messy text and asking it to explain an unfamiliar topic before a meeting. It is weaker when the answer depends on facts it has not been given, which is why most of this tutorial is about briefing and checking.

How do I get started with ChatGPT?

Go to chatgpt.com in a browser. OpenAI's help centre says ChatGPT is free to use and can be tried before you create an account, but signed-out use keeps no saved history, so sign up if you want to come back to your work. If your employer provides ChatGPT Business or Enterprise, sign in with the work account they give you instead of a personal one.

Keep to one sign-in method. OpenAI notes that if you first registered with Google, Microsoft or Apple you must keep using that method, and separate methods can create separate accounts that are not merged automatically.

For your first prompt, skip the trivia question. Ask for something you can check in thirty seconds: 'Rewrite this paragraph for a customer who has never heard of our product, under 80 words, and list any claims you added that were not in my text.' The final instruction trains a habit you will use for the rest of this tutorial.

What is a good 10-minute ChatGPT starter exercise?

Use a fictional 12-person travel agency, Harbourline Travel. A consultant has scribbled notes after a call: a client wants four nights in Lisbon in May, two adults, a mid-range hotel near the river, one day trip, no early flights, budget agreed but not yet confirmed in writing.

Minute 1 to 3: paste the invented notes and ask ChatGPT to turn them into a confirmation email that restates every requirement, flags anything still to be confirmed and contains no prices, dates or hotel names that are not in the notes. Minute 4 to 6: ask it to list each sentence in its draft next to the line of the notes it came from. Anything without a source line is an invention to delete. Minute 7 to 8: ask for a version for a client who prefers short messages, under 90 words. Minute 9 to 10: check the result against the notes yourself and correct anything the tool got wrong.

The point is not the email. It is the loop: brief with facts, ask for traceability, shorten, then verify. That loop transfers to reports, proposals and summaries.

How do I get better results from ChatGPT?

The fastest gains come from giving ChatGPT standing context once rather than repeating it in every chat. These four techniques use features OpenAI documents in its help centre.

What should I not paste into ChatGPT, and what do OpenAI's data controls say?

Do not paste anything your employer's AI policy forbids, and if there is no policy, treat client personal data, unpublished financials, passwords and contract terms as off limits in a personal account. Follow your employer's policy over anything on this page.

OpenAI's data controls page says that on personal plans you can turn off 'Improve the model for everyone' under Settings, Data controls, so new conversations are not used to train its models. Temporary chats are not used to improve models, do not appear in history and do not create memories, though OpenAI says a copy may be kept for up to 30 days for safety. OpenAI also states that by default it does not use inputs or outputs from ChatGPT Business, Enterprise, Edu or its API to improve its models. Feedback you submit can be used, so avoid rating chats that contain sensitive content.

Where does ChatGPT fall short?

ChatGPT can produce confident, wrong answers. OpenAI's own help article on truthfulness calls these hallucinations and lists fabricated quotes, studies, citations and references to sources that do not exist. Its advice is to verify quotes, data, technical information and references to external documents before you rely on them.

It also does not know your company unless you tell it, it cannot tell which of your facts are current, and features and limits vary by plan, so a technique that works on one account may be missing on another. It drafts; you decide.

How does Tosheo.ai teach ChatGPT inside a role path?

In Tosheo.ai, ChatGPT is taught through tasks from your own job rather than generic examples. A marketing coordinator might brief ChatGPT for a campaign summary, an HR generalist might draft a policy explainer and a finance analyst might ask for a variance narrative, each from supplied fictional material.

Each lesson takes 10 to 15 minutes and ends with a hands-on task in ChatGPT, marked against a visible rubric that checks the brief, the traceability of facts and your review. Feedback is specific, and you can flag a score for review. The 7-day free trial needs no credit card.

Sources and scope

OpenAI Help Center, ChatGPT capabilities overview, checked 19 October 2026 ↗

OpenAI Help Center, What is ChatGPT: FAQ, checked 19 October 2026 ↗

OpenAI Help Center, Why can't I log in to ChatGPT?, checked 19 October 2026 ↗

OpenAI Help Center, Projects in ChatGPT, checked 19 October 2026 ↗

OpenAI Help Center, ChatGPT Custom Instructions, checked 19 October 2026 ↗

OpenAI Help Center, Data controls in ChatGPT, checked 19 October 2026 ↗

OpenAI Help Center, Does ChatGPT tell the truth?, checked 19 October 2026 ↗