How to use ChatGPT for work comes down to three steps: choose the right task, give it your own context and check the output before it goes out under your name. This ChatGPT guide works through those steps and, for each feature, links to the article that covers it in detail.
The guide is for professionals who use ChatGPT for client emails, reports, research, analysis or content production. If you want to turn the principles into a fixed way of working, combine the guide with a structured AI course.
How to use ChatGPT: getting started
To use ChatGPT, create an account at chatgpt.com or in the app for iOS and Android, and type your question or task in the input field. The free version is enough to get started; a paid plan mainly gives higher limits and access to the stronger models.
You can complete your first work task with ChatGPT in five steps.
- Choose a task whose subject you know well, such as a client email based on your own notes.
- State the role, goal, audience and format in your instruction, instead of a single bare question.
- Paste your own source material into the conversation or upload the file, so ChatGPT does not have to guess.
- Adjust with follow-up questions in the same conversation, for example on length, tone or order.
- Check names, figures and facts before you use the text.
An instruction that follows those steps looks like this: “You are writing on behalf of an accountancy firm. Rewrite the notes below as a client email of no more than 150 words, businesslike and friendly, and end with one clear next step.” More worked examples by task are in ChatGPT prompts.
Before you use work data, change one setting. In Settings, under Data controls, turn off the option that lets your conversations be used to improve the models. Also, do not process clients’ personal data in a personal account without a legal basis and a data processing agreement with the provider.
Three principles before you use ChatGPT for business
Using ChatGPT for business only becomes reliable when you start from three principles. They apply regardless of your plan, the feature you use or the field you work in.
First principle: a convincing sentence and a correct sentence are not the same thing. ChatGPT is a pattern generator, not a knowledge base. If you look at exactly how ChatGPT works, you will see that the model predicts which word most logically follows the previous one. The output is therefore readable and well phrased, even when the content is wrong.
During the period when I worked intensively with ChatGPT, this was the most underestimated pitfall: a well-phrased paragraph on a subject where the facts were just slightly off. The form does not give the error away.
Second principle: checking is separate from the prompt. A better prompt gives better output, but it does not solve the checking problem. A recurring pattern I saw with ChatGPT: for a keyword, the model added a search volume without being asked, even though ChatGPT has no connection to Ahrefs, SEMrush or any other keyword tool.
Those figures are hallucinations that sound plausible to anyone without access to keyword data of their own. Checking what ChatGPT says is therefore not an extra step; it is the step.
Third principle: your name is on the document. ChatGPT is not liable for what it says; you are. From my years at Rabobank, ABN Amro and Van Lanschot I bring the four-eyes principle: nothing goes out without a second check.
These three principles form the foundation of LearnLLM’s ChatGPT Course. If you still need the definition and the wider context of the tool, first read what ChatGPT is for the basic explanation.
ChatGPT features by type of work
The ChatGPT features you use depend on the work you do. For each type of task there is a feature that structures the work better than an open-ended conversation.
Images and voice
You create images with ChatGPT image generation, and you have existing images analysed by uploading an image to ChatGPT. Editing existing photos comes with clear limitations, which ChatGPT photo editing sets out task by task.
On the move or hands-free, you work with ChatGPT voice mode.
Recurring work and long texts
A ChatGPT project brings together files, instructions and conversations for each client or case, and ChatGPT memory remembers preferences across conversations. If you work with sensitive client information, switch that memory on or off deliberately.
For recurring tasks, a custom GPT was long the standard route; OpenAI is phasing out custom GPTs on 11 December 2026 and offers a migration to plugins.
You edit longer texts and code in ChatGPT canvas, a separate window next to the chat. In ChatGPT agent mode the tool carries out tasks on its own in a virtual browser, which calls for a check at every step.
Choosing or cancelling a plan
Which plan suits you is explained in what ChatGPT costs, with the features and the models you can use on each plan. Since Go arrived there has been an entry-level plan between Free and Plus; the ChatGPT Go vs Plus choice comes down to limits and access to the stronger models.
If you want to stop or switch, cancelling a ChatGPT subscription gives the steps for each platform. You cancel a subscription bought through the App Store in a different place from one bought through the website.
When can you trust ChatGPT, and when not?
You can trust ChatGPT on tasks where you already know the facts and the model only adds form, and not on tasks where ChatGPT itself has to know whether something is correct. Recognising that distinction matters more than any prompt tip.
ChatGPT is useful for an email based on your notes, a summary of a text you wrote or a job advert rewritten for a different audience. You supply the facts; ChatGPT supplies the structure and the tone.
Translation belongs in the same category as long as you are fluent in both languages: translating with ChatGPT gives usable drafts, but technical terms and tone need checking by someone who knows both languages. ChatGPT writes fluently in many languages besides English, but it still regularly carries over English sentence structure and anglicisms, so give explicit instructions on style and check the result.
Risky tasks include summarising research you do not know yourself, describing a current market development, explaining a legal regulation or looking up a statistic. Without your own source material, the model relies on training data, and that is not always up to date.
During the period when I worked intensively with ChatGPT, five patterns kept recurring: made-up sources and quotes, made-up search volumes, non-existent Python functions, partly ignored instructions and convincingly wrong summaries. Each pattern looks plausible and only stands out if you know the facts yourself.
The best test is therefore whether an expert in the field would immediately recognise the output as right or wrong. If you want to know whether AI is dangerous in business use, that article gives the wider context for this principle.
ChatGPT among other AI tools
For a long time ChatGPT was synonymous with AI, but it is now one tool among several alternatives. Choosing a tool is increasingly a choice of working environment.
Claude from Anthropic focuses more on long, complex tasks. Google Gemini is integrated into Google Workspace and therefore handy if you work a lot in Docs, Sheets and Gmail. Perplexity is an answer engine that shows sources, and Grok from SpaceXAI searches live in posts on X.
My switch from ChatGPT to Claude had to do with consistency of output on long tasks and a firmer sense of control in workflows. If you only rewrite a text now and then, you will hardly notice the differences.
For a comparison based on work scenarios, Claude vs ChatGPT is the most direct way in. Gemini vs ChatGPT sets Google’s integration against OpenAI, and Perplexity vs ChatGPT covers the difference between generated answers and source citations.
Learning more with ChatGPT
Learning more with ChatGPT means building a fixed way of working, with checkpoints that catch errors before they end up in your output. This guide gives the structure and the principles; the course gives the practice.
In LearnLLM’s ChatGPT Course you build three complete workflows for writing, analysing and planning, and you check the output of each workflow with five fixed questions. You finish with a final assignment on a real task from your own work, a personal work portfolio and a certificate you can register as a learning activity.
