Setup Steps: Use AI Safely, One Pillar at a Time

The do-it-yourself version of the Quick Setup Prompt

Start Here gives you a prompt that has an artificial intelligence (AI) assistant walk you through setup one step at a time. If you'd rather do it yourself, here are the same steps, in the same order. First, secure the AI account itself, before you share anything. Then follow Own IT, Secure IT, Protect IT and Monitor IT: a simple cycle you can use for any account or device (know what you have, lock it down, share carefully, check in regularly), applied here to AI. Do them once now, then repeat Monitor IT every month.

The short version, if you're starting here. Before you share, install or click Connect, ask: What private data can it reach? What untrusted content will it read? How could it send, share or change something? Find the Stop button first. And use AI on email you chose to show it, not on email that chose you. Start Here has the full warning label.

Menu paths in this guide were checked in October 2026. This is the one place we keep them up to date, so check the date if you're reading later.

FIRST: Secure the AI account before you share anything

Pick one assistant, and sign in

Start with one: ChatGPT, Gemini, Claude or Copilot. The free version is fine to start. Sign in, or create a free account, before you share anything. Signing in lets you lock the account, turn off training, review what it remembers and delete chats. Some assistants let you turn off training without signing in, but you can't lock the account or review its memory. Signing in also keeps a history, which is why you'll check what's stored later.

Use your personal account for personal tasks. Remember that your account builds a story about you over time, from every chat, upload and saved memory. Limit what it keeps from the start; it's easier than cleaning up later. At work, use only the AI tools your company approved, with your work account.

Start in a web browser. For your first chats, open the assistant's website in your browser instead of an AI app built into your computer.

• It can't look around your computer. A chat in a browser tab sees what you type or upload, plus anything that's turned on in your AI account, like memory, project files or connections. That's why these steps check those. AI built into the computer can be set up to do more. On Windows, the Copilot app can, with your permission, see your screen (Copilot Vision), listen for "Hey Copilot," connect to your email and files, and in a test version, work on files on your computer. On a Mac, Apple Intelligence is built in: writing tools, Siri and smart replies in Mail and Messages, and on some versions it can pass requests to ChatGPT if you turn that on.
• It's easier to see what's on. In a browser, the settings, chats and connections are all in one place.
• It's easy to close the door. Signing out ends your session on that computer.

Built-in AI isn't bad. Some of it does its work on your device. But each feature you turn on is more access to keep track of. If you use it, turn on only the features you need, and check which are on: on Windows, in the Copilot app's settings; on a Mac, in System Settings, under Apple Intelligence & Siri.

Use a browser without extra add-ons (extensions), or a separate browser profile (a separate sign-in space in your browser) just for AI, because extensions can read what's on the page too.

Which way should I sign in?

Most AI apps offer a few ways to create an account. Signing in alone doesn't let the AI read your email or files; that takes a separate permission. Read the sign-in screen, though: if it also asks to see your email, files or contacts, that's a permission, not just a sign-in. Any of these ways can be safe. The differences are about what happens if one account has trouble.

How to choose. Which button you use matters less than three things:

1. How strong the sign-in is on the account you use: a passkey is best, then MFA with an authenticator app.
2. Whether you can get back in if something goes wrong: current recovery email and phone, and backup codes on paper.
3. What you said yes to: just signing in, or also access to your email, files or contacts.

Signing in with Google, Apple or Microsoft can be a good choice, especially if that account has a passkey and you won't use a password manager. The U.S. National Institute of Standards and Technology (NIST) notes that this kind of sign-in lets you use many services without keeping a separate password for each. The tradeoff is that the accounts are tied together.

Our approach: a separate email address and a unique password, with a passkey where the AI app offers one, otherwise MFA with an authenticator app, and backup codes on paper. We treat email as a critical account, so we keep AI sign-ins separate from it.

The one choice to avoid, whichever way you sign in: a reused password with no MFA.

Two extras:

• Give the AI account its own address. An email alias, like Apple's Hide My Email (Sign in with Apple can create one for you) or Firefox Relay, forwards to your inbox but can be turned off if it's ever leaked or spammed. Password resets still reach your inbox, so your email stays the most important account to protect.
• Check old sign-ins. In your Google account, go to Security, then "Your connections to third-party apps and services," to see every app you've signed into with Google or given access to. Remove the ones you don't use. Apple and Microsoft accounts have similar lists.

Lock the account

Your AI account may soon hold a lot about you, so protect it like your email.

• Best: a passkey, if the app offers one. A passkey lets your phone or computer sign you in with your fingerprint, face or PIN, with no password to steal.
• Next best: multifactor authentication (MFA) with an authenticator app. MFA is also called two-factor authentication (2FA). It means signing in takes your password plus a second proof. An authenticator app on your phone, like Google Authenticator or Microsoft Authenticator, makes a new code every 30 seconds.
• Better than nothing: a code sent by email or text. It's weaker. If someone gets into your email or steals your phone number, they get the code too.
• Use a long password that's different from every other password you have. A password manager can make and remember them for you.

Turn off training

Look in your AI app's settings for the setting that lets the company use your chats to improve its AI, and turn it off. In ChatGPT, as of October 2026, it's under Settings, then Data Controls. You usually won't notice any change. If you aren't signed in, this option may not be available.

Check memory and ads

• Memory. Memory means the app saves facts about you to use in later chats. Read what it saved, delete what it doesn't need, then decide whether memory stays on. Do this before the setup asks about your devices and email, so those answers aren't saved about you.
• Ads. Some free AI plans now show ads. In ChatGPT, for example, ads on the Free and Go plans can be personalized using your chats and memory unless you turn that off under Ad Controls. Ads aren't advice, so don't click them while you're setting things up.

OWN IT: Know what you have

List what you already have, and decide what AI is for

• List what you have, grouped by device: computers (Windows or Mac, and major version), phones and tablets, other devices like watches, game consoles, smart speakers and TVs, the web browsers on each, your email services, the AI apps and websites you use, and office, meeting, file storage and social apps. Then note the AI built into each one. Use types and major versions only (like Windows 11, not a build number), with no account names or serial numbers. This list is a map of your setup, so keep it outside any AI tool. If you ask AI to help build it, do that only in a signed-in account with training and memory off, and delete the chat when you're done.
• Check what people often forget: a work or school device, a second browser, a smart speaker, a TV or car, an old phone or tablet, or an AI app you tried once. Work and school devices go on the list, but their settings belong to the IT team.
• Note what each one is connected to: email, calendar, files or other apps.
• Keep it quick. You'll change settings in Secure IT, and anything extra can go on a "later" list.
• Decide what you'll use AI for, like writing help, explaining things or planning, and what you won't share with it: passwords, login codes, account numbers, and health, tax or money records that haven't been cleaned up.

SECURE IT: Lock the doors

Start with the most important: the AI that reads what other people send you.

Decide about the AI in your email, calendar and chat

Your email, calendar and chat are where other people, including strangers and hacked accounts, can put words in front of your AI without asking you. Start Here explains why, with the September 2026 KnowBe4 research, and what can happen if you leave it on. The choice is yours:

• Cautious: turn it off. Keep spelling and grammar. Use AI only on text you copied and cleaned up.
• Balanced: turn it off, then turn on only the feature you use. Never act on a code, link or phone number from an AI summary.
• Accepting: keep it on, with the safety settings below, and check the real email before you act.

Check even if you never turned it on. Email and device companies often switch AI on by default, and updates can turn it back on. As of October 2026, here's where to look:

Gmail

1. In Gmail on a computer, click the gear icon, then See all settings. Stay on the General tab.
2. Find Google Workspace smart features and click Manage Workspace smart feature settings.
3. Smart features in Google Workspace is the Gemini switch for Gmail, Calendar, Chat and Drive. Smart features in other Google products lets your mail and files feed Maps, Wallet, the Gemini app and Search. If you keep tax, health or money records in Gmail or Google Drive, turn off both. Click Save.
4. Back on the General tab, leave Smart features in Gmail, Chat and Meet on, so spelling and grammar keep working.

Safety settings for every level:

• Images: on the General tab, choose Ask before displaying external images. A picture that loads can tell a sender you opened the email.
• Dynamic email: uncheck Enable dynamic email. It lets a message change after it arrives. Click Save Changes at the bottom.
• Calendar invites: in Google Calendar, go to Settings, then Event settings. Set Add invitations to my calendar to Only if the sender is known, so a stranger's invite comes as an email instead of landing on your calendar.

If your email has no AI features, skip this step. If you keep the AI on, use it on mail you know, and leave any setting that lets AI delete or archive your mail turned off. You can change your choice anytime; Go Deeper has the full details.

Trim the AI built into your devices and browsers

From your list, start with the one or two you use most, like Copilot on Windows, Apple Intelligence on a Mac, or the AI button in your browser. Turn off the features you don't use, and keep the rest on a "later" list for next month. On Windows, look in the Copilot app's settings; on a Mac, in System Settings, under Apple Intelligence & Siri; in a browser, in its settings, usually under a section for AI or the assistant's name.

Check what's stored, web search and connections

1. What's stored. Look through your saved chats, uploaded files and shared links, and delete what you don't need. Check your own computer too: cleaned-up text files and downloads are copies.
2. Automatic web search. When it's on, the AI may send words from your chat to a search company. Turn it off for private tasks. When you need fresh information, search on purpose, in a separate chat with nothing private in it.
3. Connections. Remove any connections you don't use, and don't add new ones yet. Don't connect your email, calendar or files.

Find the Stop button first

Before you ask AI to do anything, find its Stop or Cancel button. Closing the window or the app usually doesn't stop work the AI service already started. The work runs on the company's computers, not yours, so it can keep going after you close the tab.

PROTECT IT: Share less, check more

Be the person in the middle

Don't copy straight from an email or document into a chat, and don't drag in whole files. Stand in the middle instead, so you decide every word the AI sees.

1. Copy just the part you need from the original, like one email or a paragraph.
2. Paste it into a plain-text app first. Use Notepad on Windows or TextEdit on a Mac. On a Mac, open TextEdit and choose Format, then Make Plain Text, before you paste. Some of these apps now have AI built in, so leave those features off.
3. Remove private details. Take out names, addresses, account numbers, login codes, and any health details the job doesn't need. You can swap a name for a pseudonym label (a made-up stand-in name) like [PERSON_A], and keep your list of who is who away from AI.
4. Read it once more. Make sure only what the job needs is left.
5. Copy the cleaned-up text and paste it into the chat.

Why not copy straight from the original into the chat? Some AI apps may start working on what you type or paste before you press Send, so you may not get the chance to take anything out.

Why not drag in a file? A file often starts uploading the moment you drop it in. And files carry hidden details: who wrote them, who changed them last, comments, old edits, and the file name itself, which can say a lot ("Smith_Divorce_Settlement.docx"). Plain text leaves all of that behind, so you see the words you're sending. It doesn't remove tricks hidden in the words, and the AI may still add what it already knows from memory or connections. So read what you paste, and check what comes back.

A word about labels: swapping a name for a pseudonym label like [PERSON_A] helps, but it doesn't make information anonymous. A job title, a date or a town can still point to a person. If you're required to make data anonymous, for work or by law, pseudonym labels alone aren't enough.

What you paste still goes to the AI company. That's why you clean it up first.

Ask well

A good request gets a better answer and shares less. Use four parts:

1. What you want: "Help me write a short thank-you note."
2. The background, without private details: "It's for a neighbor who fixed my fence." Describe the situation instead of pasting it.
3. How you want it: "Three sentences, friendly, not too formal."
4. Ask for honesty: "Tell me anything you're unsure about."

When the details matter. Sometimes AI can't help without specifics, like the wording of your lease or the numbers on a lab report. You can include them safely:

• Share the facts, not the identity. Keep the part that matters and take out who it's about. From a lease, keep the clause about early move-out, but remove your name, address, landlord and account numbers. From lab results, keep the test names, your numbers and the normal ranges, but remove your name, birth date, patient ID, doctor and clinic.
• Describe instead of identify. "An adult in their 50s" tells the AI enough. Your name and birth date don't help it answer.
• Use pseudonym labels like [LANDLORD] or [CLINIC] when the AI needs to tell people or places apart.
• Start with less, then add. Ask, "What else do you need to know to answer this?" Then decide what to share. You stay in charge of every detail.
• Prepare it the person-in-the-middle way first. Specifics still go through the plain-text app, never straight into the chat.
• Use a temporary chat (a chat the app doesn't save to your history) if your assistant offers one, and keep memory off, so the details aren't saved for later chats. Temporary chats may still be kept for a short time.
• Know when to stop. If the answer depends on details you're not comfortable sharing, ask the real expert: your doctor, your landlord, a lawyer or the company. AI can help you prepare good questions for them without the details.

Check the response

AI can sound sure and still be wrong, whether it gives you an answer, a summary, a draft or a translation. It can even make up facts.

• Ask a second AI. Paste the same cleaned-up question into a different assistant, or ask it to check the first response. If they disagree, dig deeper.
• Agreement isn't proof. Two AIs can repeat the same wrong fact.
• Check what matters with the real source: phone numbers, links, dollar amounts, dates, and medical or legal advice. Find phone numbers in the official app, on a statement or on a bookmark you trust.

Make the change yourself

Let AI help you decide and draft. Then you perform the actions: send the email, change the setting or cancel a subscription yourself, in the real app or website.

Start fresh before the chat gets tired

Long chats drift. The AI can only keep so much in view, so it quietly drops older parts, and your early instructions often go first. Mistakes pile up, and it starts trusting its own wrong answers. A long chat also holds more of what you pasted, which means more to lose if your account is ever broken into. And if a trick slipped in early, through a pasted email or web page, it keeps steering every answer after it.

General rule: one task, one chat. Start a new chat when:

• The topic changes.
• You've had a long working session with many messages or large pasted inputs. There is no universal message count at which a chat becomes unreliable.
• The app says it's summarizing or compacting the conversation.
• You notice warning signs: the AI forgets a rule you gave it, repeats itself, contradicts an earlier answer, mixes up names or versions, makes things up, ignores your formatting, or slows down a lot.

Start fresh sooner if you've pasted a lot of text, even text you cleaned up first. Every paste uses up room the AI needs to remember your earlier instructions.

To carry on, have the AI write a handoff. Paste this into the old chat:

We're going to continue this in a new chat. Write a handoff summary
I can paste there. Include:
1. The goal, in one or two sentences.
2. Decisions we made, and why.
3. Where we are now, and what's done.
4. What's left to do, in order.
5. My rules and preferences, word for word where I gave them.
6. Anything you're unsure about, marked "Check this."
Leave out passwords, codes, account numbers and personal details.
Don't add anything we didn't discuss. Keep it under one page.

Open the Handoff Prompt page to copy the prompt.

Then be the person in the middle: copy the summary into a plain-text app, fix anything wrong, remove anything private, and check the "Check this" items yourself. Then paste it into a new chat with this line on top: "Here's where we left off. Confirm you understand, and list any questions before we continue."

A handoff can carry a mistake forward just as easily as the old chat did. For big projects, keep the real decisions and final versions in a file you control. The chat is a helper, not your record.

Clean up after. Once the new chat is working, delete the old one if you don't need it. Less saved means less to lose.

MONITOR IT: Check in and clean up

Clean up and shrink your footprint

Everything you leave behind in an AI account could be stolen if someone breaks into the account or the company. That's your footprint; security teams call it your attack surface. The blast radius is how much damage one break-in could do. Cleaning up keeps both small.

• Save final outputs or what you need on your own device.
• Delete chats you no longer need, and any uploads or shared links.
• Delete your prepared copies, like the plain-text file you cleaned up.
• Review memory. Delete anything the AI saved about you that it doesn't need.
• Remove connections and installs you're not using.
• Sign out if you used a shared or public computer.
• Once a month, do the monthly check below.

Deleting doesn't always erase every copy right away, and companies may keep some copies for a while. It's still the right habit: the less sitting in your account, the less someone can take if they get in.

Your monthly check

• Look for apps that can still reach your accounts, emails you didn't send and new devices you don't know.
• Shrink your footprint: delete old chats, uploads and shared links, review memory, and remove connections and installs you don't use.
• Recheck your settings after app updates. Updates can turn features back on.
• Check your email for new AI features, including ones you never turned on, and make sure they still match the level you chose.
• Delete old chats you replaced with a handoff summary.
• When an alert email arrives, sign in to the account yourself to check it. Don't use the email's link.

What you find sends you back to Own IT, and the cycle starts again.

Next: put a monthly reminder in your own calendar for this check. When you're ready for more, Go Deeper covers connections, installs, agents and backups.

---

Own IT. Secure IT. Protect IT. was the 2019 Cybersecurity Awareness Month theme from CISA and the National Cyber Security Alliance. Monitor IT is the author's addition, because the first three are a theme and setup, but adding the fourth turns it into a repeatable lifecycle and sends you back to Own IT. This is an independent guide, not a CISA publication or endorsement. The author develops commercial security tools in this area; this guide is tool-neutral and recommends no product. Assistants and apps named here are examples, not endorsements. Settings were checked in October 2026 and change often. CC BY 4.0.

Sources

• Learn about smart features and controls | Google Help
• Choose whether to show images | Gmail Help
• More than a Password: multifactor authentication | CISA
• Ads in ChatGPT | OpenAI Help Center
• Digital Identity Guidelines: Federation and Assertions (SP 800-63C-4) | NIST, July 2025
• Manage invitations in Calendar | Google Calendar Help
• Turn off Copilot in Microsoft 365 apps | Microsoft Support
• Turn AI-powered message summaries on or off in New Yahoo Mail | Yahoo Help
• How to Get the Best From Apple Mail AI or Turn It Off | Clean Email
• Are Your AI Assistants the Next Attack Surface? | KnowBe4
• Making every Windows 11 PC an AI PC | Microsoft Windows Blog, October 2025
• Apple Intelligence brings powerful AI capabilities into everyday experiences | Apple Newsroom, June 2026

Setting names and availability vary by region, account, app, and update. Confirm in the current official help and your own account. Calendar invitation limits and blocked external images are supporting controls; they do not establish that message contents are trustworthy.

Explore the guides

All AI safety guides | Reference library