ChatGPT Cheat Sheet: Ten Things It Actually Does Well

Most ChatGPT cheat sheets are a list of keyboard shortcuts and nothing else. Useful, but it is not the thing that actually determines whether you get value out of the tool week to week. This one is organised around capability instead: what ChatGPT reliably does well, a worked example of each, and the one check worth running before you act on the result. Things to ask ChatGPT that are actually worth asking covers five of these in more depth; this is the fast-reference version.

Text and writing

• —Summarising something long for a named reader — not "summarise this" but "summarise this for someone who has two minutes and has not read it".
• —Tightening your own draft — paste a paragraph and ask it to cut by a third without losing the main point.
• —Rewriting tone — formal to casual, or the reverse, with the actual audience stated.
• —Rehearsing a conversation you are nervous about, with it playing the other side realistically rather than agreeably.

Every one of these is genuinely strong because the task is transforming text that already exists, not recalling a fact from nowhere — the distinction what AI is actually bad at covers in full. Fun things to do with ChatGPT is worth a look too if the playful end of this list is more useful right now than the practical one.

Files and data

Attach an .xlsx, .csv or .pdf file and ChatGPT opens the actual content rather than guessing from a filename — OpenAI's own file upload documentation sets the real limits, and for a PDF specifically there is a separate visual-retrieval path and its own plan requirements worth knowing before you rely on it for a scanned document rather than a text one. Can ChatGPT read Excel files? walks through what survives that process correctly in a real, slightly messy workbook and what it quietly gets wrong, and can you upload Excel to ChatGPT covers the upload mechanics themselves.

Projects and memory

If you come back to the same background information every time — a client, a codebase, a running document — ChatGPT Projects groups a conversation with its own files and instructions so you stop re-pasting the same context each session. It is a genuinely underused feature next to the chat box itself.

Images and voice

ChatGPT can also generate an image from a description and edit one you already have by describing the change — a different skill from the text tasks above, and one with its own structure worth getting right. AI image prompts covers the six-part formula that makes the difference between a usable result and a generic one. Voice mode is a genuine alternative input for the same text tasks, not a separate capability — useful for rehearsing a conversation out loud rather than typing both sides of it.

Common mistakes that waste a cheat sheet

• —One giant conversation for every unrelated task. Context from an old topic bleeds into a new question in ways that are hard to notice — start a fresh chat, or a dedicated Project, per topic instead.
• —Asking one mega-prompt instead of iterating. A long first attempt that tries to specify everything at once is harder to debug than a short one you refine in two or three rounds.
• —Treating a confident number as a checked one. The confidence of the answer and the accuracy of the answer are not the same signal — see the hallucination point below before repeating a figure to anyone.
• —Re-explaining background every session instead of using a Project or pasting a short standing brief at the top of the chat.

Searching for something current

When browsing is turned on, ChatGPT can look something up rather than answer from memory alone — worth using for anything that changes week to week: a current price, a recent announcement, today's news. It still writes the answer in the same fluent voice either way, so the fact that it searched is not, by itself, a reason to skip checking the page it actually found. Ask it to show which source it used, and open that source yourself before repeating a figure from it.

A worked example: the difference one detail makes

Weak: "What can you tell me about this spreadsheet?"

Better: "This workbook has two sheets, Orders and Refunds, both keyed on order_id. For every order_id in both sheets, tell me whether the refund amount is larger than the order amount. List only those rows."

Same file, same model. The first produces a generic description of what a sheet looks like. The second names the two sheets, the shared key, and the exact condition worth flagging — a task that plays to what the tool is actually built for: matching and filtering rows against a stated rule, the same shape OpenAI's own prompting guidance recommends for any task, not only spreadsheets.

What "GPT" actually means, and why it matters here

A language model predicts the next piece of text given everything before it — it has no separate step where it checks whether it actually knows something before answering. The original paper behind that architecture is why it is strong at transforming text you give it and weaker at recalling a fact it was never shown. What does GPT stand for in ChatGPT? covers the mechanism itself if the short version here leaves you wanting more.

Where this cheat sheet stops being useful

None of the ten tasks above ask ChatGPT to recall a specific fact from memory — a stock price, your account balance, today's exchange rate. A survey of hallucination in large language models documents fluent, confident answers regardless of whether the facts behind them are real, and that is precisely where a cheat sheet built on tool capability earns its keep: every item above works from text or data you actually supplied, which is the version of a request this tool is reliable at.

The one check to run, every time

1. 01Read the result against the source you pasted or attached, not against how confident it sounds on its own.
2. 02For a spreadsheet task specifically, open two or three of the flagged rows yourself before repeating the answer to anyone.
3. 03If a draft reads generic, the prompt was probably missing a reader, a format, or the actual input — not a sign the model cannot do the task.
4. 04If the real task is a multi-step research question rather than one file or one draft, deep research prompts has a template built for that shape instead.

Coursium teaches this kind of practical judgement directly — what a tool actually does well, and the check worth running before anyone acts on the result. Stay ahead of AI by learning the tools on your phone.