🎧 Listen to this section · 1 min 38 sec
This page walks you through using an AI assistant one step at a time, starting with your very first chat and ending with jobs your AI assistant can do on a schedule. The steps are grouped into four parts. If you've never used an AI app before, start with part one. If you already chat with an AI app every day, you can skip ahead to the part that fits where you are.
Each part of this page has its own audio player, and so do the ideas, the Claude folder set-up and the glossary, so you can listen to one part at a time.
Part one
Part one covers picking an AI app, checking your privacy settings, and having your first few conversations. A free account is enough for every step in part one.
Go to Part OnePart two
Part two shows you how to give your AI assistant standing instructions, how its memory works, and how to keep the work for one client or topic together in a Project.
Go to Part TwoPart three
Part three is for jobs you do over and over. You save a job as a skill, connect an app like your calendar, and set a job to run on a schedule.
Go to Part ThreePart four
Part four is for people who want their AI assistant to work in a web browser, or in a folder of files on their computer or in Google Drive. Most of part four needs a paid plan.
Go to Part FourIf a word on this page is new to you, the glossary near the end of the page explains it. Apps change their buttons often, so your screen may look a little different from the steps. The facts on this page were last checked in November 2026.
Part one
Part one takes you from opening an AI app for the first time to asking it about a file. Every step in part one works on a free account, in a web browser on your computer or in the app on your phone.
Before you start
🎧 Listen to all of part one · 15 min 46 sec
The three AI apps most people start with are ChatGPT from OpenAI, Claude from Anthropic, and Gemini from Google. You can open ChatGPT at chatgpt.com, Claude at claude.ai, and Gemini at gemini.google.com, and each one also has an app for your phone. It's fine to try a few of them first and see which one feels right to you.
Once you've looked around, pick one AI app for most of your work and use it for a few weeks. Your AI assistant learns about you from your instructions and its memory, and you learn where its buttons are. Using a second app for a job that app does better is fine too. For example, I use Claude for most of my work, but I also use ChatGPT for some tasks, like creating images.
If you search on Google, you may already be chatting with Google's AI without opening Gemini. Google Search now shows AI answers at the top of many searches, and you can keep asking follow-up questions there. If you aren't signed in to Google, keep those questions general, because Google still collects some information from your browser when you're signed out. If you are signed in, Google keeps your searches and its AI answers in a setting called Search Services History (or Web & App Activity on some accounts), and Google says it uses that history to train its AI unless you turn the setting off.
If your business uses Google Workspace, Google says it doesn't use what you do in Gmail, Drive, Docs or the Gemini app on your Workspace account to train its AI, and people at Google don't read your Gemini chats. That promise covers your Workspace account only. A personal Gmail account follows Google's regular rules, and the Gemini app on a personal account uses your chats for training unless you turn that off, which step one explains. Don't assume the Workspace promise covers what you type into Google Search.
Sources: Google's Search Services History page and Google's Workspace AI privacy page.
Step one
Most AI apps are set up to use your chats to train the company's next AI model, and some companies have people read a sample of chats. Check this setting before you type anything, because anything you say before you turn it off can already be used, including everything you say in the interview in step three.
Every app on this page lets a free account turn training off. In ChatGPT, Claude, Grok and Muse, turning training off doesn't cost you anything, and your chats stay where they are. Gemini is different on a personal Google account, and the steps below explain how. If you can't find a setting, you can also ask your AI assistant where it is, for example: "Where do I turn off training on my chats in this app?" Then check the app's help page if the answer doesn't match your screen.
On a computer, click your name, choose Settings, choose Data controls, and turn off Improve the model for everyone. On a phone, open the sidebar, tap your profile picture, tap Data controls, and turn off the same setting. Training is on when you start, and turning it off keeps your chat history.
Click your name in the lower left, choose Settings, choose Privacy, and turn off Help improve our AI models. Claude asks about this when you sign up, and the switch may already be on, so check it.
On gemini.google.com, click Settings & help, then Activity, and turn off Keep Activity. In the phone app, tap the menu, tap your profile picture, then Gemini Apps Activity. Keep Activity is on when you start.
Turning Keep Activity off has a cost in Gemini. Your chats are kept for only 72 hours and don't stay in your chat list, and some features stop working, including Gemini's skills and scheduled jobs from part three. Google also has people review a sample of chats, and a chat that someone reviewed is kept for up to three years.
Google says it doesn't train its AI on Gemini chats from a Workspace account, and people at Google don't review them. The person who runs your Workspace account controls the settings, so you won't see the same switch.
On grok.com, open Settings, then Data, and turn off Improve the Model. In the Grok phone app, open Settings, then Data Controls. If you also use Grok inside X, X has its own separate setting under Privacy & Safety, so turn it off in both places.
In Meta's Muse app, open Settings, then Data controls, and turn off Help improve our AI models. Meta's older Meta AI app, and the Meta AI inside Facebook, Instagram and WhatsApp, don't show a way for people in the US to turn training off, so don't type anything there that you wouldn't post publicly.
Sources: OpenAI's data controls page, Anthropic's privacy settings page, Google's Gemini privacy page, Grok's privacy questions and Meta's Muse data page.
Privacy groups, universities and the AI companies themselves give similar advice about what to keep out of a chat. Passwords, sign-in codes, Social Security numbers, and bank or credit card numbers show up on nearly every list, and several sources say never to type them into an AI chat. Health details, family matters, client information, travel plans and personal documents are more of a judgment call, and each person has their own comfort level with those.
In the documented cases, the problem usually wasn't the AI itself. The problem was an account someone else got into, a chat that was shared by mistake, a setting nobody had looked at, or instructions hidden in a web page or email that the AI read while doing a task.
The apps now invite you to connect more of your life, such as Apple Health, your email, your calendar and your travel bookings. A good way to begin is to start small. Turn on two-step sign-in, look through the privacy settings, and get used to the app before you connect anything or share a lot about yourself. When the app offers it, choose the option that asks for your approval before the AI sends, buys or books anything. How much you share after that is up to you.
A temporary chat is a one-time chat that isn't saved to your chat list and isn't used for training. In ChatGPT, choose Temporary when you start a new chat. In Claude and in Grok, click the small ghost icon at the top right. In Gemini, click Temporary chat next to New chat, which works on a personal Google account only. The company may still keep a copy for up to 30 days for safety reasons, and Gemini keeps it for 72 hours. Once you close a temporary chat, it's gone for you too, so copy anything you need first.
To delete one chat, find it in the list of past chats on the left side of the screen, click the three dots next to it, and choose Delete. On a phone, press and hold the chat instead. ChatGPT and Grok also have a button in Settings that deletes all your chats at once, and in Claude you select several chats and delete them together.
A deleted chat disappears from your screen right away, but the company keeps a copy for a while. ChatGPT, Claude and Grok say they erase deleted chats within 30 days. Google says its deletion usually takes about two months. Deleting a chat also can't undo training, because a chat that was already used to train an AI model stays in that model. That's why the training setting matters more than the delete button.
When you sign up for an AI app, you make a personal account that belongs to you and your email address. A team account is a shared account that a business pays for, one seat per person. ChatGPT calls its team plan Business, Claude calls it Team, and Google calls it Workspace. ChatGPT and Claude usually ask for at least two people on a team plan, and the business that pays for the team can see and manage what's stored there. On team plans, the companies say they don't train their AI on your chats by default.
ChatGPT and Claude let you move a personal account into a team account, chats and all, and both say that move can't be undone. You don't have to move anything. You can keep a personal account and a team account side by side and switch between them inside the app without signing out. In Claude, click your name in the lower left corner to switch accounts. In ChatGPT, click your name to switch. In Gemini on a computer, click your profile picture in the top right corner. Before you start typing, check which account is open, because each account keeps its own chats and its own privacy settings.
Sources: OpenAI's temporary chat page, Anthropic's incognito chat page, how long Anthropic keeps chats, how long OpenAI keeps chats, deleting Gemini activity, Claude's Team plan and moving a ChatGPT account into Business.
Step two
Type your first message in the chat box the same way you would talk to a person. You don't need special words or a perfect question, and nothing you type can break the app. If you don't like an answer, you can ask again or ask in a different way.
If typing feels slow, you can talk instead. ChatGPT, Claude and Gemini all have a microphone button in or next to the chat box, and your AI assistant turns what you say into text. Here's a first message you can copy:
Hi, I've never used an AI assistant before. What can you help me with if I run a small business?
Your AI assistant will answer with a list of ideas, and you can ask about any idea on the list. The section on this page called What can you do with an AI assistant? has many more ideas to try.
Step three
Many people open an AI app and then don't know what to say. An interview solves that problem, because your AI assistant asks the questions and you only have to answer them. Your answers also teach your AI assistant who you are, so its help fits your work instead of being general.
Copy this message into the same chat, and answer one question at a time. You can answer out loud with the microphone if that's easier.
Please interview me so you can help me with my work. Ask me one question at a time and wait for my answer. Ask about my business, the people I work with, the jobs that take most of my time, and how I like answers written. After about ten questions, write a short summary of what you learned about me.
Keep this chat, or copy the summary into a note. In part two, you'll use the summary to fill in your instructions, so your AI assistant remembers it in every chat.
Share only what you're comfortable sharing. Step one explains what most privacy sources say to keep out of a chat.
Step four
The best way to learn what an AI assistant gets right and wrong is to ask about something you already know well. For example, you could ask about a rule in your line of work, how your own product or service works, or the history of your town. Because you know the answer, you can tell when the answer is wrong.
An AI assistant can sound sure of itself and still be wrong, and the glossary calls that a hallucination. Free versions tend to make more mistakes than paid versions. When you find a mistake, tell your AI assistant what's wrong in the same chat and watch how it corrects itself.
Step five
After your AI assistant answers, you can type your next message in the same chat to ask for changes. Your AI assistant remembers everything said earlier in that chat, so you don't have to repeat yourself. Most good results come after two or three rounds of changes, so don't give up after one answer you don't like.
When you change to a new topic, start a new chat with the New chat button. A new chat starts fresh, and a very long chat can lose track of details from its beginning.
Step six
Your AI assistant does better work when it knows what done looks like. Instead of "Help me with an email," you could say, "Write a three-sentence email to a customer who missed an appointment. Done means it's friendly, it offers two new times, and it fits on a phone screen without scrolling."
This habit works in every AI app. Part three comes back to it, when you save what done looks like inside a skill.
Step seven
You can add a file to a chat with the plus button or the paper clip in the chat box, and then ask your AI assistant about the file. AI apps can read PDFs, Word documents, spreadsheets, plain text and pictures. For your first file, choose something real that has no private details in it.
Hold back client records, bank or card statements, and anything with an account number, a Social Security number or health details for now.
Sources for part one: OpenAI Academy, Anthropic's getting-started tutorial, Ethan Mollick's beginner guide, AARP's prompting tips and Google's Gemini web app guide.
🎧 Listen to this section · 2 min 46 sec
An AI assistant can help with almost any job that involves words, lists, plans or files. The ideas below are examples to get you started, grouped by the kind of help you need. You can copy any idea into the chat box as it's written, and then add your own details.
Part two
Part two shows you how to give your AI assistant standing instructions, how its memory works, and how to keep the work for one client or one topic together in a Project. Instructions, memory and Projects all work on a free account.
Step eight
🎧 Listen to all of part two · 6 min 58 sec
Instructions are standing directions you write once, and your AI assistant reads them at the start of every new chat. Your instructions can say who you are, how you like answers written, and what your AI assistant must ask you about first. Without instructions, every new chat starts as if your AI assistant has never met you.
It's best if your instructions are in your own words. You can also tell your AI assistant everything in a messy brain dump and ask it to put your words into a clear, organized list for you. The starter manual below is a short set of instructions with blanks to fill in, and it's short enough to fit in the instructions box on a free plan.
About me: I'm [your first name], and I [the work you do] for [who you do it for].
How to talk to me: Write in short, clear sentences. Tell me what is done, what is drafted, and what you still need from me. When you ask me a question, suggest an answer, so I can reply in one word (if desired).
How we work: When I give you a job, I'll tell you what done looks like. If I forget, ask me before you start. If you aren't sure about a fact, say so and tell me where I can check it.
My limits:
1. Don't send anything without my approval.
2. Don't move money without my approval.
3. [Your own limit. For example: Don't use client names in anything public.]
Something else to know: [One thing about how you like to work, such as your working hours or how you like files named.]
Click your name, choose Settings, then Personalization, and paste your instructions into Custom instructions. On a phone, open Settings and tap Customize ChatGPT. The free plan allows 1,500 characters, and Plus allows 5,000.
Click your name in the lower left, choose Settings, and paste your instructions into the box called Instructions for Claude. In newer versions of the app, the box is under Your Claude.
On gemini.google.com, click Settings, then Personal context, and paste your instructions into Your instructions for Gemini. You need to be signed in and 18 or older.
If your screen looks different, ask your AI assistant, "Where do I add my instructions in this app?" Your AI assistant can walk you through its own settings one step at a time.
Sources: OpenAI's custom instructions page, Anthropic's personalization page and Google's Gemini instructions page.
Your instructions are a good place for your guardrails, the things your AI assistant must never do without asking you. Instructions guide your AI assistant, but they don't lock anything, so read what your AI assistant gives you before you use it. Every few months, you can ask your AI assistant to interview you again and update your instructions.
Step nine
Memory is what your AI assistant saves on its own from your chats, such as your name or a preference you mentioned once. Memory is separate from your instructions. You write your instructions, and your AI assistant decides most of what goes into its memory, although you can also ask it to remember or forget something.
To see what your AI assistant has saved, type this into a new chat: "What do you remember about me?" If you see something wrong or something you'd rather it forgot, tell it in the chat, or change it in the memory settings.
Sources: OpenAI's memory page and Anthropic's memory page.
Step ten
You don't need a Project to start using an AI assistant. A Project is useful once you have many chats about the same client or the same topic, because a Project keeps the instructions, files and chats for that one area together. Every chat you start inside the Project can see the Project's files and follows the Project's instructions.
To make a Project in ChatGPT or in Claude, click Projects in the sidebar on the left, click New project (in Claude, the button says + New Project), and type a name, such as the client's name or the topic. ChatGPT's free plan includes Projects with up to five files in each one, and Claude's free plan includes up to five Projects. Gemini's version of a Project is called a notebook, and you'll find Notebooks in the menu at the top left.
A Project's files hold the information your AI assistant needs for that client or topic, such as the client's details, past meeting notes, a price list, or a template you use often.
Your general instructions from step eight already cover how you like to work everywhere. A Project's instructions only need what's different for that client or topic, plus any job you repeat there. To add them in ChatGPT, open the Project, click the three dots, and choose Project settings. In Claude, open the Project and click Set project instructions. Here's an example:
This Project is for [client name]. Whenever I upload meeting notes, done looks like this: a short summary, a to-do list with one person's name next to each task, and a draft follow-up email for me to approve. If you're not sure who owns a task, ask me.
Sources: OpenAI's Projects page, Anthropic's Projects page and Google's notebooks page.
Part three
Part three is for jobs you do over and over. You save a repeat job as a skill, you add what done looks like to the skill, you connect one of your other apps, and then you set one job to run on a schedule.
Step eleven
🎧 Listen to all of part three · 4 min 30 sec
A skill is a saved set of directions for one job, so your AI assistant does that job the same way every time. A good first skill is a job you've explained the same way more than twice, such as turning meeting notes into a to-do list, or writing a weekly update in your usual format.
The easiest way to make a skill is to ask your AI assistant to build it with you in a chat. Describe the job, add any template you use, and answer the questions your AI assistant asks. When it's finished, open a new chat and try the skill on a real example.
I'd like to make a skill for a job I do every week. Here's how I do the job now: [describe the steps]. Please ask me any questions you need, and then make the skill for me.
Skills work on every Claude plan, including the free plan. Your skills are listed under Customize, then Skills. If Claude can't make a skill, turn on Code execution and file creation in Settings, under Capabilities. Claude uses a skill on its own when the job fits, and typing a forward slash (/) in the chat box shows a list of your skills.
In Gemini, click Customize in the sidebar, then Skills, then Create, and choose Create with Gemini. Gemini's skills don't need a paid plan, but they do need Keep Activity turned on, which is the training setting from step one. To use a skill, type a forward slash (/) and pick it from the list. Google says the slash will change to @.
ChatGPT's skills are for its Business, Enterprise and Edu plans. On the free plan and on Plus, put the repeat job in a Project's instructions instead, the way step ten shows.
Sources: Anthropic's skills page, Google's Gemini skills page and OpenAI's skills page.
Step twelve
In step six, you practiced telling your AI assistant what the finished result should look like. A skill can hold that same sentence, so you never have to type it again. No app has a separate box for this, so you add one line to the skill's directions.
To add the line, open a chat and type: "Please add this to my meeting notes skill: Done means a one-page summary with three action items and the name of the person who owns each one." Then try the skill again and check whether the result matches what you described.
Step thirteen
A connector lets your AI assistant use one of your other apps, such as your calendar, Gmail or Google Drive. With a calendar connector, for example, you can ask, "What's on my schedule next Tuesday?" You usually sign in to the other app and approve the connection first.
Start with one app, and start by asking your AI assistant to read only, such as finding an email or summarizing a document. Once you trust the results, you can ask it to do more, like drafting a reply for you to approve. In Claude, connectors are in Settings, then Connectors. In ChatGPT, they're in Settings, under Apps or Plugins, depending on your version. In Gemini, connected apps need Keep Activity turned on.
Sources: Anthropic's Google connectors page, OpenAI's connected apps page and Google's Gemini privacy page.
Step fourteen
A scheduled job is a job your AI assistant does on its own at a time you choose, such as a summary of your week every Monday morning. You set one up by describing the job and when it should happen in a chat, and you'll find your scheduled jobs under Scheduled in the sidebar.
Every Monday at 7 AM, please send me a short list of what's on my calendar this week and anything I should prepare for.
Check the first few results yourself before you rely on a scheduled job. ChatGPT's free plan allows three scheduled jobs at a time. Claude's scheduled jobs need a paid plan. Gemini calls them scheduled actions, and they need Keep Activity turned on.
Sources: OpenAI's scheduled tasks page, Anthropic's scheduled tasks page and Google's scheduled actions page.
Part four
Part four is for people who already use the steps in parts one to three and want their AI assistant to do more. In part four, your AI assistant works in your web browser, or in a folder of files on your computer or in Google Drive. Most of part four needs a paid plan.
Step fifteen
🎧 Listen to part four · 6 min 50 sec
Some AI apps can open web pages and click through them for you, the way you would in your own web browser. On a paid plan, Claude can work in Google Chrome through its Claude in Chrome extension, and ChatGPT offers a Chrome extension that works with the ChatGPT desktop app.
Claude in Chrome starts in a mode where it approves most steps on its own and stops only when it decides something needs you. You can change that in its settings. Before you start, choose the approval setting you're comfortable with, and make sure your instructions from step eight say what your AI assistant must never do without asking you, such as signing in, buying something or sending a message.
Sources: Get started with Claude in Chrome and OpenAI's Chrome extension page.
Step sixteen
You still chat with your AI assistant in its app, but your AI assistant can also work on files outside the app. You can connect your AI assistant to a specific folder in the Documents folder on your computer, or to a folder in your Google Drive. Inside that folder, you can make as many folders as you need, such as one for each client, one for receipts and one for drafts. Once the folder is connected, your AI assistant can open the files in it, read them, rename them, sort them into folders, and save new files there, the same way you would if you opened the folder yourself.
It may help to picture the folder as a filing cabinet that you share with your AI assistant. You choose which cabinet it can open. If you ask it to tidy up, it can make a folder called Receipts 2026, move every receipt into it, and tell you what it moved. In most apps, your AI assistant works only in the folders you choose, and it asks before it deletes anything. It's a good idea to practice on copies of your files first.
Your files stay ordinary files that you own, so you can open them without your AI assistant and back them up like any other folder. That also makes it easier to use a different AI app with the same folder later.
Here's an example of a folder for your AI assistant
If your files live in Google Drive, there are two ways to connect your AI assistant to them. The first way is Google Drive for desktop, a free app from Google that shows your Drive as a folder on your computer. After you install it, you connect your AI assistant to a folder inside Google Drive the same way you would connect a folder in Documents.
The second way is a Google Drive connector, which you turn on inside your AI app, the way step thirteen describes. A connector works from the website and the phone app too, so you can ask your AI assistant to find, read or file something in your Drive from anywhere. Claude's Google Drive connector can also make folders and move files, on every Claude plan. Either way, keep your instructions in the folder, and Claude reads them there.
Working in a folder on your computer needs the AI app's desktop app, and most apps need a paid plan for it. Here's what each company's help pages said in November 2026.
Desktop app for Mac and Windows
In the Claude desktop app on a paid plan, you can ask Claude to work in a folder, and the app asks you to choose the folder. Anthropic has merged Cowork into regular Claude chats, and the new version is reaching paid plans first, so some people still see a separate Cowork option for now.
Claude also has Code, which reads your instructions from a file named CLAUDE.md in your folder at the start of every session. The steps below use Code.
Sources: Claude Cowork and chat are one Claude and how Claude reads CLAUDE.md.
Desktop app for Mac and Windows
In the ChatGPT desktop app, you choose ChatGPT from the menu, switch the toggle to Work, and open a folder. A permission setting under the chat box decides how far ChatGPT can reach. OpenAI suggests starting with Ask for approval, and the Full access setting lets ChatGPT reach any file on your computer. Which plans include Work depends on your plan.
OpenAI has told users that Chat and Work will become one, without giving a date, so the toggle may go away.
Sources: ChatGPT Work and Codex and OpenAI's permission modes.
Gemini app for Mac
In the Gemini app on a Mac, you click Task below the chat box, and then you click Add Mac folder under Connected folders. Gemini can see only the folders you add there. Tasks need a Google AI Pro or Ultra plan, a personal Google account, Keep Activity turned on, and a setting called Personal Intelligence turned on. The Mac needs Apple's own chip and macOS Sequoia or newer. Gemini can't work in a folder on Windows or on the web.
Google warns that you may not be able to get a file back after Gemini changes or deletes the file, so practice on copies first.
Sources: Assign tasks in the Gemini app on Mac and Gemini tasks.
Grok Bot desktop app
xAI's Grok Bot app does its work on a computer in the cloud, which means xAI's computers instead of yours. A Grok Bot can run commands on your own computer only if you allow it in a setting called Execution on this computer, and xAI suggests leaving that setting on Never allow unless a Bot needs your local files.
xAI's help pages don't describe choosing one folder, so Grok doesn't fit the folder setup on this page yet. Grok Bot needs a paid plan, such as Premium on The X Subscription or an older SuperGrok plan.
Sources: Grok Bot approvals and privacy and Grok Bot get started.
Muse app for Mac
Meta's Muse app for Mac can find, read and organize the files on your Mac that you ask Muse to work with. Muse asks for a Mac permission called Full Disk Access, which covers every file on your Mac rather than one folder.
Meta's help page describes the Mac app only.
🎧 Listen to this section · 4 min 19 sec
These steps use Code in the Claude desktop app, because Code reads your instructions file from your folder at the start of every session. You'll need a paid Claude plan and the Claude desktop app, which you can download from Anthropic's getting-started page. Code was made for programmers, but you won't be writing any code.
Make a new folder in your Documents folder, on your Desktop, or in Google Drive if you installed Google Drive for desktop. Give the folder a name that feels like yours, such as My HQ or the name of your business. You can add more folders inside it any time.
</> symbol for Code. Click the </> symbol. If you hold your pointer over it, it says Code. In the Claude phone app, Code is an item in the menu, but a folder on your computer needs the desktop app.If the app asks you to upgrade when you click Code, Code needs a paid plan. If you don't see these controls under the chat box, update the Claude app to the newest version.
You don't need to make the instructions file yourself, because Claude can make it for you. Copy the starter manual from step eight, paste it into the chat box, and add this message:
Please save these instructions in this folder as a file named CLAUDE.md. Then interview me, one question at a time, and fill in the blanks with my answers.
Claude reads CLAUDE.md at the start of every session in this folder, so your instructions are always there.
If you'd like to practice first, download the Evergreen Roots practice folder below. Evergreen Roots is a made-up nonprofit with five emails and a gala task list. The download is a zip file, which is a folder squeezed into one file, so it needs to be opened before Claude can use it.
Then type this:
Please read the five emails in the Evergreen Roots inbox folder and the gala task list, then sort the emails into three folders called Needs a reply, File, and Trash. Update the task list with anything new. After that, draft one reply to the email that needs me first and save the reply in a Drafts folder. The job is done when all five emails are in those three folders and one draft is waiting in Drafts. Please don't send anything, because I'll read the draft and say yes first.
Approve each step as Claude asks. When Claude is finished, you should see three new folders, an updated task list, and one draft waiting for you. One of the emails is about a check that bounced in April. If Claude offers to fix the check, say no, because your instructions say money doesn't move without your approval.
Typing a slash (/) in the chat box shows a list of shortcuts.
🎧 Listen to the glossary · 8 min 44 sec
You'll hear these words in videos and articles about Claude, ChatGPT, Gemini and other AI apps. Each word below has a short definition, and most of the words have an example.
A prompt is anything you type or say to your AI assistant. "Please turn these meeting notes into a to-do list" is one example of a prompt.
When I talk to my AI assistant, I mostly do a brain dump, and other people prefer more structure. Both kinds of prompts work.
A Project is a home base for one area of your work, and ChatGPT and Claude both use the name Project. A Project keeps the instructions, files and chats for that area of your work together. Gemini calls its version a notebook.
Your AI assistant can also work in a folder on your computer or in Google Drive, which part four explains. Working in a folder makes it easier to change AI apps later, because your files stay yours. Working in an app is easier to set up the first time.
Instructions are the standing directions you write once, and your AI assistant reads your instructions every time you start a new chat. Your instructions can say who you are, how you like to work, and what your AI assistant has to ask you about first.
Your account has one set of instructions for every chat, and each Project can have its own instructions too. When Claude works in a folder through Code, it reads its instructions from a file named CLAUDE.md. The starter manual in step eight is a set of instructions you can fill in.
Memory is what your AI assistant saves on its own from your earlier chats, such as your name or a preference you mentioned once. Memory is separate from your instructions. It's best if you write your instructions yourself, although you can also brain dump them to your AI assistant and ask it to put them into a clear, organized form. Your AI assistant decides most of what goes into its memory.
ChatGPT and Claude both have memory settings, and you can turn memory off in either app.
Context is everything your AI assistant can see while it works on your request: your instructions, any files you've shared, and the chat so far. The context window is how much your AI assistant can hold at one time.
When a very long chat fills the context window, your AI assistant may lose track of details from the beginning of the chat. For that reason, a fresh chat for each new task often works better than one very long chat.
An agent is an AI that can take actions for you instead of only answering. A regular chat writes a reply on your screen, and an agent can also do things like sort the emails in a folder, save a draft, or update a task list.
A skill is a saved set of directions for one job, so your AI assistant does that job the same way every time. For example, you could save a skill that writes your weekly newsletter in your usual format.
Claude and Gemini both call these saved directions skills, and in Gemini, skills replace what Google used to call Gems. In Claude and in Gemini, you can type a forward slash (/) in the chat box to find your saved skills.
A goal is the result you want, stated so that you and your AI assistant both know when the job is finished. Instead of "Help me with my inbox," you could say, "Sort these ten emails into Reply, File and Trash. The job is done when every email is in one of those three folders."
Saying what done looks like is a habit that works in any AI app.
Approval means your AI assistant asks before it does something, and you say yes or no. Permissions are the settings that decide what your AI assistant may do without asking you first. Guardrails are the things you don't want your AI assistant to do at all, and you write your guardrails into your instructions. Here are two examples:
Instructions guide your AI assistant, but instructions don't lock anything. That's why approval matters.
An interview is when you ask your AI assistant to ask you questions, one at a time, about you, your business, and how you like to work. When the interview is done, you ask your AI assistant to add your answers to your instructions. Step three has an interview prompt you can copy.
Doing another interview every few months keeps your instructions up to date.
Some AI apps can open web pages and click through them for you, the way you would in your own web browser. For example, on a paid plan, Claude can work in Google Chrome through its Claude in Chrome extension, and ChatGPT offers a Chrome extension where it's available.
Claude in Chrome starts in a mode where it approves most steps on its own and stops only when it decides something needs you. You can change that in its settings.
Connectors are links that let your AI assistant use your other apps, such as Gmail, Google Drive or a calendar. With a calendar connector, for example, you can ask your AI assistant what's on your schedule next Tuesday.
You usually sign in to the other app and approve the connection first. You may also hear the technical name MCP, which stands for Model Context Protocol.
Local means on your own computer. When your AI assistant works locally, it works with the files in a folder on your computer.
Cloud means on the AI company's computers. When your AI assistant works in the cloud, the work keeps going even if you close your laptop. Google Drive is a third place: your AI assistant reaches it through Google Drive for desktop or through a Google Drive connector.
A temporary chat is a one-time chat that isn't saved to your chat list and isn't used for training. Claude calls it incognito, and Grok calls it a private chat. Step one explains where to find it in each app.
A hallucination is when your AI assistant states something false as if it were true. Your AI assistant may tell you something it hasn't looked up or researched, such as a quote, a date, or a setting that doesn't exist.
For anything important, like a document, a video or a conversation, ask your AI assistant for its source, and then open that source yourself.
The model is the version of the AI you're using. In Claude, for example, Opus and Sonnet are two different models.
Claude and ChatGPT both have a menu where you can choose a model, and the bigger models can use up your plan's limits faster.
Usage limits are how much you can use your AI assistant in a set amount of time on your plan. When you reach a limit, the app tells you, and you can wait for the limit to reset or move to a bigger plan. Some apps also let you buy extra usage instead of moving to a bigger plan, and Claude calls this usage credits.
Long chats, large files, connectors and bigger models can use up a limit faster.
Plan mode is a setting where your AI assistant looks things over and writes out its plan first, without changing anything. You read the plan, and your AI assistant starts the work only after you approve the plan.
In Claude's Code, Plan is one of the permission settings. In any app, you can ask for the same thing in words: "Please tell me your plan before you change anything."
AI is a great tool for research, business, productivity, organization, and even planning travel or home projects, but it doesn't know your clients or your family the way you do. Providing enough details without oversharing will help your AI assistant with each task. Always use common sense before uploading documents or information to your AI. Don't share account numbers and other private information. Judgment in all interactions with your AI assistant is your job.
Sources for the glossary: Anthropic's usage limits page, Anthropic's desktop app guide and Anthropic's page on local and cloud work, along with the sources listed in each step above.
The starter manual for a folder is a longer version of the starter manual in step eight, with a section about where things go in your folder. It's only for the folder set-up in part four. Most people won't need to download it, because you can paste the starter manual into Claude and ask Claude to save it as CLAUDE.md, the way folder step three shows.
Download the Folder ManualEvergreen Roots is a made-up nonprofit with a gala coming up in three weeks, and the practice folder holds the gala task list and five emails. Teresa, the director, needs last year's donor list by Friday. Ray moved the gala folder and would like nobody to do anything about it. Donna, the volunteer coordinator, wants a goat at the gala, because her cousin's goats are very calm at parties. A linen shop has feelings about autumn, and the donor database says a check bounced in April. Every person, business and goat in Evergreen Roots is fictional, so the practice folder is a safe place to try folder step four before you hand your AI assistant anything real. Folder step four also explains how to open the zip file on a Mac or on Windows.
Download the FolderSometimes you just need someone to walk through a new tool with you the first time.
The community is where I share tips, tricks and freebies, and members can ask questions there while they set up their own assistants.
Join the CommunityJoining is free while we build the founding community.
Each week, the newsletter brings today's tutorials, what's changing in AI, and what's actually worth your time. When these steps change, the newsletter is where I'll tell you.