How to Use ChatGPT Well: A Trainer's Practical Guide

ChatGPT is a chat assistant from OpenAI that writes text by predicting likely next words, based on patterns it learned from a very large body of training data. It is good at drafting, summarising, explaining and reformatting. It is bad at being reliably correct on facts you have not given it, so treat its output as a first draft that you check, not as a source.

I wrote the first version of this post in mid 2023, when most of my students had only just tried it. I have rewritten it since, because the tool has changed a lot and the way people use it at work has changed even more. What has not changed is the basic advice in the last two sections.

What is ChatGPT, in plain terms?

OpenAI released ChatGPT to the public on 30 November 2022. Under the hood it is a large language model (LLM) built on the transformer architecture, which Google researchers described in the 2017 paper "Attention Is All You Need". The model is first trained on a huge amount of text to predict the next token (a word or part of a word). It is then tuned with human feedback so that it behaves like a helpful assistant instead of a text autocomplete.

That second step matters more than people think. It is why ChatGPT answers in a friendly, confident tone even when it is wrong. Confidence in the wording tells you nothing about accuracy.

The product has moved on since 2022. Newer versions can read images and files, search the web, run code and remember context across chats if you allow it. Model names change every few months, so I won't list them here. Check OpenAI's own release notes if you need the current line-up.

How does ChatGPT actually produce an answer?

When you type a prompt, the model reads your text plus anything earlier in the conversation, then generates a reply one token at a time. Each token is chosen from a probability distribution. There is no lookup in a database of verified facts unless the product has a search or file tool switched on and actually uses it.

Two practical consequences follow:

• If the answer depends on something recent, local or private (your company's sales figures, a GSEB circular from last week), the model cannot know it unless you paste it in or it searches for it.
• If you ask for a citation, it may produce something that looks like a real reference and is not. This is usually called hallucination. It has improved, but it has not gone away.

Where does ChatGPT help at work?

These are the uses I see paying off for the working professionals and students in our batches:

• Drafting and rewriting. Emails, job descriptions, meeting notes, a first pass at a report. You still edit, but you skip the blank page.
• Explaining code and formulas. Paste a SQL query or a DAX measure and ask what it does line by line. For learners this is one of the best uses, because you can ask "why" five times without feeling awkward.
• Data cleanup ideas. Describe a messy column and ask for a Power Query step, an Excel formula or a Python pandas snippet. Then test it on a copy of the data.
• Customer support drafts. Teams use it to suggest replies that an agent reviews before sending. Fully automatic replies are riskier, for the reasons in the next section.
• Study help. Turning your own notes into practice questions, or quizzing you on a topic. Our Generative AI training spends a full module on this kind of everyday workflow.

What are ChatGPT's limitations?

The limits are well documented, and they are the reason I never let a batch submit AI output without review.

• Wrong facts stated confidently. The model can mix up dates, numbers and names. Verify anything that will be published or acted on.
• Weak on ambiguous prompts. If your question can be read two ways, it will pick one without telling you.
• Bias from training data. The model reflects patterns in the text it learned from, including stereotypes. OpenAI publishes notes on this in its model system cards.
• Privacy. Anything you paste goes to OpenAI's servers. Business plans and Team or Enterprise workspaces have different data terms from the free consumer app, so read your organisation's policy before pasting client data.

How do you write a good prompt for ChatGPT?

Most bad results come from short, vague prompts. The fix is boring and it works: give the model what a new colleague would need.

1. Say who the output is for ("a store manager with no SQL background").
2. Give the material ("Here is the table structure:" and paste it).
3. State the format ("a numbered list, under 150 words").
4. Ask it to flag uncertainty ("If you are guessing, say so").

Then read the answer critically and reply with corrections. A conversation of three or four turns almost always beats one long prompt.

How should a business use ChatGPT responsibly?

Write down a short policy before people start pasting things in. It should say what data is never shared, who reviews AI-written content before it goes to customers, and how staff should disclose AI use where that matters. Keep a human in the loop for anything legal, financial or medical.

If you want your team trained on this, we run Generative AI workshops in Anand and online. For a related look at how the major models compare, see Gemini vs GPT-4.

FAQ

Is ChatGPT free to use?

OpenAI offers a free tier with usage limits and paid plans with more capacity and newer models. Pricing and limits change often, so check the plans page on openai.com for current details.

Can ChatGPT understand Hindi and Gujarati?

Yes, it can read and write many languages, including Hindi and Gujarati. In our experience the quality is noticeably better in English, so check non-English output more carefully.

Does ChatGPT give correct answers?

Often, but not reliably. It generates plausible text, which is usually right on common topics and can be wrong on specific facts, recent events and numbers. Verify before you use it.

Will ChatGPT replace data analysts or marketers?

It changes the work more than it removes it. Routine drafting and boilerplate code get faster. Knowing which question to ask, checking the result and explaining it to a manager still needs a person who understands the domain.

Is it safe to paste company data into ChatGPT?

Only if your organisation's policy allows it and you are on a plan whose data terms you have read. When in doubt, anonymise the data or describe its structure instead of pasting real records.