How to use ChatGPT effectively, in short
Give ChatGPT proper context, tell it who to be and what a good answer looks like, and treat the first reply as a rough draft rather than the finished article. State the role, the task, the audience, the format and the length, paste in your own material, then push back until it earns its keep.
- A prompt is a brief, not a question. One line in gets you one bland line back.
- The first answer is where the conversation starts, not where it ends.
- Turn on Custom Instructions and use Projects. Two minutes, permanent payoff.
- Check anything that matters. It will be confidently wrong and never tell you.
Stop asking ChatGPT questions. Start writing briefs.
I run The Oxford AI School, so I watch people meet ChatGPT for the first time most weeks. The pattern is always the same. They are nervous, they ask it something vague, it gives them something bland, and they quietly decide the whole thing is a con. Then I show them the difference a decent brief makes and the penny drops in about ninety seconds.
A prompt is a brief. If you would not hand a one-line brief to a freelancer and expect brilliance, do not hand one to ChatGPT either. The more it knows about what you are doing and why, the less generic the answer.
Why one-line prompts give one-line answers
With nothing specific to work from, ChatGPT averages out everything it has ever read. That average reads like a textbook, because it is one. The fix is not clever wording or secret phrases. It is context: what you are trying to achieve, who it is for, and what the finished thing should look like.
The five things a good ChatGPT prompt includes
- Role. "Act as a plain-speaking accountant explaining this to a nervous sole trader."
- Task. The actual job. "Rewrite this email so it chases a late invoice without sounding rude."
- Audience. Who reads it. A board, a customer, your mum. It changes everything.
- Format and length. Bullet points, a table, a 200-word paragraph, a script. Ask for the shape you want.
- Your material. Paste the email, the notes, the data, the draft. Never make it invent the facts.
A ChatGPT prompt template you can copy
Steal this and fill in the blanks. It works for almost any written job:
"You are [role]. I need you to [task]. It is for [audience], who care about [what matters to them]. Give me [format] of about [length], in a [tone] tone. Here is my material: [paste your notes, figures, email or draft]. Ask me anything you need before you start."
That last sentence is the one people skip and the one that pays best. Invite the questions and you will often get three good ones that improve the brief before a single word is drafted.
That is the whole secret. Not a prompt library, not a course in "prompt engineering", just the difference between "write me a marketing plan" and "here is my business, my budget and my last three months of sales, draft me a one-page plan for October aimed at local customers." One gives you Wikipedia. The other gives you something you can use on Tuesday morning.
Treat the first answer as a draft, not a verdict
The single biggest habit shift is this: the first reply is where the conversation starts, not where it ends. ChatGPT is not a vending machine. It is closer to a very fast, very willing junior who has read most of the internet and remembers none of your standards until you tell them.
The follow-up prompts that do the heavy lifting
So tell it. Good work comes from the second, third and fourth pass, exactly as it would with a person.
Four phrases worth memorising
- "Too formal, loosen it up." Tone is the fastest thing to fix and the thing people put up with longest.
- "Cut it in half." Almost every first draft is twice as long as it needs to be.
- "You have invented a statistic. Remove anything you cannot stand behind." Say this often.
- "Now give me three versions and I will pick." Choice beats iteration when you are stuck.
Two settings worth turning on today
Custom Instructions let you tell ChatGPT once who you are and how you like things written, so you stop repeating yourself in every chat. Projects let you keep related chats and files in one place, which matters the moment you are doing anything longer than a single question. Both live in your settings and take two minutes to set up. They are the closest thing to a free upgrade.
A word of warning that never goes out of date: ChatGPT will state things with total confidence and be completely wrong. It does not know it is wrong, because it does not know anything. Check anything that matters, especially numbers, names, dates and law.
What ChatGPT is genuinely good at, and what to keep it away from
Point it at these
- Drafting anything written. Emails, proposals, adverts, descriptions, posts, replies.
- Rewriting and fixing. Shorter, friendlier, less salesy. It is a fast, patient editor.
- Summarising. Paste the long document, thread or transcript and ask for the five things that matter.
- Explaining. "Explain this to me like I run a shop, not like I have a degree in it."
- Brainstorming and structuring. Names, angles, outlines, agendas, plans.
Keep it well away from these
Your final figures, unless a human has checked them. Legal, tax or medical questions treated as gospel rather than a starting point. And confidential information on a free account, because on the free plan you cannot assume your data stays private. Client details, staff records and commercially sensitive material stay out unless you are on a proper business plan with the settings sorted. If you want the practical version of that, read how to keep information safe using AI tools.
Making images: the ChatGPT caricature trend, explained
ChatGPT does not just write. It draws. The ChatGPT caricature craze, where people turn a selfie into a cartoon, a pop-art poster or an action-figure-in-a-box, is the most fun most people have with AI in their first week, and it is a genuinely good way to learn how image prompts work.
How to make a ChatGPT caricature
The method is the same as with text. Upload a clear photo, describe the style you want, and say what to exaggerate and what to keep. "Turn this into a friendly hand-drawn caricature, play up the eyebrows, keep it recognisably me, warm colours, white background." Then refine. If the first go makes you look like a startled owl, tell it so and try again.
Two rules before you upload a photo
- Only upload photos you have the right to use.
- Do not make caricatures of other people without asking them first. A cartoon of your own face for your LinkedIn banner is a bit of fun. A cartoon of your boss is a conversation you may not want to have.
Beyond the chat box: Codex and ChatGPT Atlas
Two names come up constantly and confuse people, so here they are in one line each. Codex is OpenAI's coding agent, aimed at developers, which writes and runs code across a whole project rather than handing back a snippet. ChatGPT Atlas is OpenAI's own web browser, with ChatGPT sitting alongside every page so it can read, summarise and, in agent mode, actually do things.
Both matter because they show the direction of travel: less "AI that answers questions", more "AI that does the task while you supervise". If you want the full plain-English version, read ChatGPT Atlas and Codex, explained.
Claude vs ChatGPT: am I on the wrong one?
This is the question I get asked most, usually phrased as "everyone keeps mentioning Claude, am I on the wrong one?" The honest answer is that it is not a fight with a winner. They are both excellent, they are good at slightly different things, and plenty of people I train end up with both open.
Rough rule: want one tool, want to make images, want the biggest ecosystem? Start with ChatGPT. Write a lot, work with long documents, care how natural the output sounds? You will probably prefer Claude. Serious about this? Learn both, and read the full Claude vs ChatGPT comparison.
A note on the Boris Johnson ChatGPT moment
You may have seen the clip. Boris Johnson cheerfully admitted he loves ChatGPT "because it says I'm brilliant", delivered in a pronunciation so plummy and drawn-out that the internet stopped listening to what he was saying and started replaying how he said it.
There is a small lesson buried in the comedy. It does not matter how you say the name, or whether you sound like you were born in a rowing club. What matters is what you get the thing to do. Plenty of people who say "ChatGPT" perfectly get nothing useful out of it. The measure of using it well is the work on the other side, not the vocabulary.
The one habit that changes everything
If you take a single thing from all of this, take this: talk to ChatGPT like a capable colleague, not a search box. Give it the context a colleague would need, tell it what good looks like, and keep going until the work is right. Do that and it stops being a party trick and starts being the most patient, fastest assistant you have ever worked with. Ignore it and you will keep getting one bland line for every bland line you type. The tool was never the problem.
How to Use ChatGPT Effectively: frequently asked questions
How do I use ChatGPT effectively?
Give it context, tell it who to be and what a good answer looks like, and treat the first reply as a draft. State the role, the task, the audience, the format and the length, paste in your own material, then push back and refine. Most people who feel let down by ChatGPT simply asked a one-line question and accepted a one-line answer.
What makes a good ChatGPT prompt?
Five things: a role, the task, the audience, the format and length, and your own material. "Write me a marketing plan" gives you Wikipedia. "Here is my business, my budget and my last three months of sales, draft me a one-page plan for October aimed at local customers" gives you something you can use on Tuesday morning.
Do I need to learn prompt engineering to use ChatGPT?
No. You need context, not magic words. There is no secret phrase and no prompt library worth paying for. If you can brief a colleague or a freelancer clearly, you can prompt ChatGPT well, and that is most of the skill.
Why are ChatGPT answers so generic?
Because the prompt was generic. With nothing specific to work from, it averages out everything it has read, which reads like a textbook. Give it your facts, your tone and your constraints and the blandness disappears almost immediately.
What are ChatGPT Custom Instructions and Projects?
Custom Instructions let you tell ChatGPT once who you are and how you like things written, so you stop repeating yourself in every chat. Projects keep related chats and files together in one place, which matters as soon as you are working on anything longer than a single question. Both live in your settings and take two minutes to set up.
Can ChatGPT make a caricature of me?
Yes. Upload a clear photo and ask for a caricature in the style you want, for example a friendly cartoon, a pop-art poster or a hand-drawn sketch. Tell it what to exaggerate and what to keep, then iterate. Only upload photos you have the right to use, and avoid using other people’s faces without their permission.
Is it safe to paste work documents into ChatGPT?
On a free personal account, treat everything you type as potentially non-private, so keep client data, staff records and commercially sensitive material out. On a paid business or team plan with the right settings, your data is not used for training and it is far safer. Set it up properly once and you can use it with confidence.
How do I stop ChatGPT making things up?
You cannot stop it entirely, so build the habit of checking. Give it your source material rather than letting it recall from memory, ask it to flag anything it is not certain about, and tell it to remove any statistic or quote it cannot stand behind. Always verify numbers, names, dates and law yourself.
Harry Lang runs The Oxford AI School, teaching teams and businesses the basics of AI in a fun, easy and practical way. A marketing leader turned AI trainer, he can be reached at Learn@TheOxfordAISchool.com.
Practical ChatGPT training for you or your team
The Oxford AI School delivers plain-English AI training in Oxfordshire and London, across the UK, and live online for teams in the USA and other international markets. Sessions cover prompting, checking output, data safety, images, agents and the tools your team already has open.
Where teams usually start
Basic AI Training
Learn prompting, checking, data safety and practical everyday use in a morning.
Explore this trainingAI Basics Deep Dive
Go further on prompting, projects, images and research with hands-on practice.
Explore this trainingWhich AI Tool to Start With
Compare ChatGPT with Claude, Gemini, Copilot and Perplexity for your business.
Explore this trainingAI Policy Workshop
Set clear rules for safe ChatGPT use before your team pastes something it shouldn't.
Explore this trainingThe ChatGPT series
Four plain-English guides to the questions people actually type into Google and ChatGPT about ChatGPT.
Get more than bland answers out of ChatGPT
The difference between useless and brilliant is a handful of techniques. We teach individuals and teams those in a single practical morning, in person or online.
Book a call