What it is, in one paragraph
ChatGPT is a program you write to in ordinary sentences, and it writes back. It learned from a very large amount of text how words tend to follow one another, and it answers by writing the words most likely to come next. That makes it useful for drafting, rewording, explaining and planning. It is not a search engine, and it is not a person who knows things: it can state a wrong date or an invented fact in the same calm, sure voice it uses for a right one.
What you type is called a prompt. That is the only new word you need today, and it means nothing more than what you type. Any other unfamiliar word you meet is explained in a sentence or two in the glossary.
ChatGPT, Claude, Gemini and Copilot are all tools of this kind, and every step below works the same way in each of them. From here on, this guide calls whichever one you have open “the AI”.
If you have put this off, you have company. In a Pew Research Center survey of U.S. adults in February 2026, about half said they never use a chatbot, which is the survey’s word for these tools. Of those who never use one, 55% said that not knowing how to use them was a reason, and 79% said the same of concern about how their personal information will be used (Pew Research Center, June 17, 2026). The next 20 minutes deal with both.
- 0–2 minSee what it isA writing tool that can be confidently wrong.
- 2–7 minTry a first taskA week of dinners, with nothing private in it.
- 7–13 minAdd plain detailsSay who it is for and how long, then what you already know.
- 13–17 minPush backSay what to change, as you would to a person.
- 17–20 minLearn the two rulesKeep private details out, and check what matters.
A suggested pace. Take longer over any part you like.
Your first task: something harmless
Start with a task where a wrong answer costs nothing and you can judge the result yourself. Planning a week of dinners is a good one. It needs nothing private, and you will know at a glance whether the answer is any use.
Type this and send it:
Plan a week of dinners.
A list of dinners comes back. Read it the way you would read a suggestion from a stranger, because that is what it is. The AI does not know how many people you cook for or how much time you have, so it filled those gaps with a guess about an average household. Some of the dinners will suit you and others will not.
A clumsy question is not a mistake here. There are no special words to learn, and you can ask again as often as you like.
Make the answer fit you with plain details
The quickest way to a better answer is to tell the AI what a helpful person would ask before starting. Two details describe the result you want: who it is for, and how long it should be. One more gives it something to work from: what you already know.
The first two shape the answer. Dinners for two adults who get home at six are not dinners for a family with a toddler, and one line per dinner can be read in a minute where a page of recipes cannot. The last detail is the part only you can supply, such as what you will not eat and what is already in the kitchen.
Here is the same request with those details added. Switch between the two versions to see what changed.
Before
1/10 Needs work
- Has: What to make
- Missing: Who it's for
- Missing: Background facts
- Missing: Length
- Missing: Format
- Missing: What to avoid
- Missing: What good looks like
- Missing: An example
- Missing: If unsure
- Missing: Layout
After, with what was added marked
8/10 Strong
- Has: What to make
- Has: Who it's for
- Has: Background facts
- Has: Length
- Has: Format
- Has: What to avoid
- Has: What good looks like
- Missing: An example
- Missing: If unsure
- Has: Layout
An example. The second version says who the dinners are for and how long the answer should be, then adds what the cook already knows. The marked words are what it adds. The score counts ten plain checks, such as whether a request says who it is for.
Now try it yourself with a different task: a short note to a neighbor. Change the request in the box, in your own words. The score beside it goes up as the details go in.
2/10 Needs work
- Has: What to make
- Has: Who it's for
- Missing: Background facts
- Missing: Length
- Missing: Format
- Missing: What to avoid
- Missing: What good looks like
- Missing: An example
- Missing: If unsure
- Missing: Layout
Say who the neighbor is and what the note is about. Then add how long it should be and what you already know, such as the day and the time. The score is worked out in your browser, and nothing is sent anywhere. Open in the Prompt Grader
The score comes from the Prompt Grader, a free tool that shows what the AI would still have to guess in anything you plan to type.
When you want the whole request written out for you, copy this one and fill in each blank, which is marked with double curly brackets:
Write a short note to my neighbor that I can put through their door.
Who it is for: {{who_the_neighbor_is}}, for example "the couple next door, whom I know a little".
What it is about: {{what_you_want_to_say}}.
What I already know: {{facts_they_need}}, such as the day and the time.
Length: under 80 words, so it fits on a small card.
Format: plain sentences I can copy out by hand, with a greeting at the top and a line at the end where I can sign my name. No headings and no bullet points.
Tone: friendly and direct. Do not apologize more than once, and do not use formal phrases.
Use only the facts I gave you. If something the neighbor would need is missing, ask me before you write the note instead of guessing.
The last two sentences matter most. They tell the AI to work only from what you gave it and to ask when something is missing, which stops it from filling gaps with guesses. Notice too that the note leaves a line for your name, so you never have to type it.
Push back until it fits
The first answer is a draft, and you are allowed to say so. Reply in the same conversation and say what to change, the way you would to a person helping you. One word is a complete reply.
| When the answer is | Type something like |
|---|---|
| Too long | Shorter. One line per dinner. |
| Hard to follow | Simpler. Use everyday words and short sentences. |
| Built on a wrong guess about you | Ask me questions first, then try again. |
| Right except for one part | Keep the rest. Swap Tuesday for something without cheese. |
| Not what you had in mind | Give me two other versions to choose from. |
“Ask me questions first” is the one to remember. It turns the conversation around: the AI stops guessing and asks, and all you have to do is answer.
If you would rather answer questions than write a request at all, the Prompt Builder works that way. It asks a few plain questions and gives you a full prompt to copy into your AI.
The two rules that keep you safe
Rule 1: do not paste private details
What you type goes to the company that runs the AI, and you should assume it is stored. Leave out anything that identifies you or someone else: full names, home addresses, phone numbers, account and card numbers, passwords, and medical or financial records. The dinner plan and the neighbor note needed none of these, and most everyday tasks do not either.
Sometimes the task is a real letter or email with names and numbers in it. Safe Paste hides those details in your browser before you paste, then puts them back into the answer. Each company also publishes a privacy page that says how it uses conversations. Read the one for the tool you chose before you share anything personal.
Rule 2: check anything that matters before you rely on it
A dinner plan needs no checking. A dose, a deadline, a price, a legal rule or a figure you plan to repeat does. The AI sounds just as sure when it is wrong, so its tone is no guide. Find the same fact at a source that would know, such as the organization’s own website, or ask a person whose job it is to know.
Use it. The answer was clear and confident.
Not quite. An AI answer sounds just as confident when the number is wrong, so confidence tells you nothing about this price.
Ask “Are you sure?” and trust the price if it says yes.
Not quite. It can repeat a wrong price as firmly as a right one. Asking where the price comes from is better, as long as you then open that source yourself.
Look the price up on the town’s own website before you rely on it.
Yes. One look at a source that would know settles it. For anything about your health, your money, your legal rights or your safety, ask a person who is accountable for the answer.
Ask the same question again and use the price if it comes back the same.
Not quite. Two matching answers can share one mistake. If the two disagree, you have learned that the price is unreliable, but agreement does not confirm it.
An invented example. The rule is the same for any number you plan to act on.
Our guide on how to tell if ChatGPT is wrong lists the kinds of detail that deserve a second look. When an answer is full of facts, the Fact Check List pulls out every number, quote, source and date so you can see what there is to check.
What to do next
Use the AI on small things for a week: a packing list, a polite reply to an invitation, a recipe from what is in the cupboard, a plain explanation of a word you keep hearing. Each time, add the plain details and push back once. That is most of the skill.
When you want more, the AI Skills Check is a short quiz that shows where to start, and the free course teaches one check per lesson, each with an exercise scored as you type. If you use AI at work, Prompt engineering for beginners covers the same ground for work tasks in fifteen minutes.
Key takeaways
- A chat AI writes likely words. It is not a search engine, and it can be wrong while sounding sure.
- Start with a harmless task you can judge yourself, such as a week of dinners.
- Say who the answer is for and how long it should be, then add what you already know.
- Treat the first answer as a draft and push back in plain words, such as “shorter” or “ask me questions first”.
- Keep private details out of what you type.
- Check any fact you plan to rely on at a source that would know.
Questions
Is ChatGPT safe to use?
Two rules cover most of the risk in everyday tasks. Do not type private details such as full names, addresses, account numbers or medical records, because what you type goes to the company that runs the tool. And check any fact you plan to rely on before you act on it. For a letter with names in it, Safe Paste hides them first.
Do I need to be good with computers to use ChatGPT?
No. If you can write a text message or an email, you can use it: you type what you want in ordinary sentences and it writes back. There are no commands to learn and no wrong way to ask. What helps most is describing the result you want and adding what you already know, as the dinner example in this guide shows.
What should I ask ChatGPT first?
Something harmless that you can judge for yourself: a week of dinners, a note to a neighbor, a packing list or a birthday message. Leave medical and legal questions until you have seen how it behaves, and even then treat its answer as preparation for asking a person who is accountable for what they tell you.
Can ChatGPT be wrong?
Yes, and it does not sound unsure when it is. It writes the words most likely to come next, so it can give a wrong number or an invented source with complete confidence. Before you rely on a detail, find it at a source that would know, or ask a person whose job it is to know.
What is a prompt?
A prompt is what you type into the chat: your question or request, along with anything you paste in with it. A short prompt gets an answer built on guesses about you. A prompt that says who the answer is for and what you already know gets an answer that fits. The glossary explains other words you may meet.
Prompts to use with this
- Explain regression or statistical output in plain wordsYou ran an analysis, or were sent one, and need to know what the coefficients, p-values and R-squared actually say before you report them.
- Explain insurance coverage options to a client in plain wordsYou have quotes from several insurers and need to explain to a client what each covers and excludes, and what it costs, without overstating anything.