Fun Things to Do With ChatGPT (With Working Prompts)
Five fun things to do with ChatGPT beyond jokes — a text adventure, a trivia host, a roleplay partner — with the actual prompts and what to check.
Most people use ChatGPT for email drafts and homework help, then stop there. It is also a reasonable text adventure game, a trivia host that adapts to what you already know, and a practice partner for a conversation you are nervous about. None of these need a clever prompt — they need a specific setup and, in a couple of cases, a fact-check before you repeat anything it tells you.
A text adventure with you as the main character
Prompt: "Run a text adventure game. I am a courier in a city where it rains constantly and the trains have stopped running. Describe the first scene in three sentences, then give me three numbered choices. Wait for my answer before continuing, and keep track of anything I am carrying."
The model improvises the next scene from whichever option you pick, which is a genuine strength — generating a plausible, coherent continuation from an open-ended premise is exactly the kind of task language models are built for. Keep the choices numbered and the scenes short; a vague "what do I do?" produces a vaguer game than one with a clear menu each turn.
A trivia host that matches your level
Prompt: "Host a trivia round on 1990s films. Ask one question at a time, medium difficulty, and tell me if I got it right before moving to the next one. After five questions, give me my score."
This is also where the fun version and the accuracy problem meet directly. A trivia host that states a wrong release year or a wrong actor's name with the exact same confident tone as a correct one is not a rare glitch — it is a documented property of how these models generate text, and a trivia game states a lot of specific, checkable facts in a row. Play it for what it is: a fun format for facts you already roughly know, not a source for facts you are learning for the first time.
A roleplay partner for a real conversation
Prompt: "Play the role of a hiring manager interviewing me for a project-management role. Ask one question at a time, react the way a real interviewer would, and after five questions tell me honestly which answer was weakest and why."
This works for a difficult conversation you are dreading as much as a job interview — telling a landlord about a repair, asking for a raise, practising a language you are learning by ordering a coffee in it. The specific setup does the work: naming the role, the number of exchanges, and what kind of feedback you want at the end, rather than a bare "help me practise for an interview".
A specific setup — the scene, the role, how many turns, what feedback you want at the end — is what turns a flat chat into an actual game.
A quick planning session for something you are already looking forward to
Prompt: "I have a free Saturday and £40. Suggest three completely different ways to spend it in a city, one sentence each, then let me pick one to plan in more detail." This is a genuinely useful, low-stakes way to see the model handle an open brief — it has to generate real options rather than retrieve a fact, which is closer to what it is actually good at than a trivia question is.
Where this goes wrong, and what to check
- Any specific fact stated mid-game — a date, a name, a price — gets checked separately before you repeat it to someone else. Checking an AI answer when you are not the expert applies exactly as much to a trivia answer as to a work document.
- If you push back on a wrong answer the model gave, do not expect a reliable correction — research on sycophancy in language models has found several widely used models will fold and agree with whatever you push back with, right or wrong, which makes an argument with the model a bad way to settle a trivia dispute.
- For the roleplay and interview-practice prompts, treat the "feedback" at the end as one perspective, not a verdict — it is drawing on the same pattern-completion as everything else it writes, not a certified read of your actual performance.
- Keep prompts specific rather than open-ended — showing the model exactly what shape you want back beats describing a vague style, which is why "three numbered choices" works better than "what happens next?" Writing a prompt that works on the first try covers the same specificity rule for work tasks, if the habit is worth carrying over.
If jokes and images are more your speed
The prompts above lean on a specific setup rather than trying to be funny on purpose. If you want the humour version of the same skill, funny ChatGPT prompts that actually work covers the two ingredients that make a joke land instead of falling flat, and ChatGPT photo prompts and AI image prompts cover the same specificity rule for pictures instead of text.
What to do Monday
Pick one of the four setups above and run it for five minutes on your lunch break. Both OpenAI's and Anthropic's own prompting guidance make the same point that applies here: a specific, concrete request gets a better result than a vague one, whether the task is a work email or a text adventure. Getting comfortable with that habit on something low-stakes is a reasonable way to build it before you need it for something that matters more. Coursium teaches that transferable prompting skill directly, in short lessons rather than a list to memorise. Stay ahead of AI by learning the tools on your phone.