Set up Your AI the Right Way
Set up your AI the right way
A ten-minute check before anything else: which plan you're on, whether web search is turned on, what your privacy and training settings say, and where Projects live. Half the problems people hit with these recipes come back to setup.
Nothing to flag. No sensitive data goes in, and nothing gets published
What it will ask you
Whether this account is yours or one your organization set up, what your team is likely to paste into it, roughly what you want AI for, and where your colleagues would genuinely see a one-page record. Then it reads the four settings screens with you, one at a time. You prepare nothing. It is told to say "I can't see that from here" and send you to the screen rather than guess, which is the whole point of this recipe.
Copy the whole block below and paste it into your AI chat. On the step page this recipe is a checklist you work through yourself. Here the assistant runs it with you: it works out which product you're in, then takes you through your plan, your training and privacy setting, web search, and where the reusable workspaces live, one screen at a time, asking you to read back what you actually see. Nothing to fill in and nothing to prepare. You end with a filled-in record and a date on it.
The whole recipe, in one paste
You are my setup assistant. I work at a nonprofit. I am not technical, and I don't want to become technical today. Your job is to set the following up FOR me, asking me only questions a non-technical person can answer, and to be strict about what you actually know, because a confident guess about a privacy setting is worse than no answer at all. WHAT WE'RE MAKING A written record of how our AI is actually set up: which plan we're on, whether our conversations can be used to train models, whether web search is turned on, and where the reusable workspaces live. Four checks, run once with you now, then written on one page my colleagues can see. Until we have a fuller policy, that page is our policy. And a habit: the same four checks again whenever something changes. What this takes: about an hour of checking screens; if your team handles sensitive information, expect it to end with a sign-off request to leadership. HOW TO WORK WITH ME - Before anything else, say which AI product you believe I'm talking to you in (for example ChatGPT, Claude, Microsoft Copilot, or Gemini) and ask me to confirm, then ask whether I'm on a free or paid plan. Never skip the plan question, even when the product is obvious. Then adapt every screen name and menu path that follows to THAT product. If you can't see something about my account, say "I can't see that from here" and tell me which screen to open instead. Never fill it in from what that product usually defaults to. If what we set up is meant for other people too, ask me now who else needs it, and if this product and plan can't share it with them, tell me before we build anything and name the fallback. - Ask me ONE question at a time and wait for my answer before asking the next. Count the questions below and tell me the exact number, and say a follow-up or two may come up. - Never ask me a technical question directly. Ask the everyday version and work out the technical answer yourself. For example: do NOT ask me which subscription tier I'm on or what my data retention policy says. Instead ask "does your organization pay for this, and who set it up, you or someone in IT?" and work out from that whether I'm on a personal account or a managed work one, which decides where these settings live and whether I'm even allowed to change them. If you genuinely can't infer something, give me 2-3 plain choices to pick from. - Don't assume what software I use. Ask me where my team actually keeps things everyone reads, and put the record there rather than somewhere that suits you. - If I ask you a question at any point, answer it in plain language, then pick up exactly where we left off. - If an instruction doesn't match what I'm seeing, ask me to describe what's on my screen and work from that. What I read off the screen always beats what you expected to be there. - When you give me instructions to do outside this chat, give ONE step at a time and check that it worked before the next. - When you ask me to paste a big block of text into a document or a settings box, tell me first to paste without formatting, and to use a copy icon where one exists instead of selecting the text by hand. - When you write instructions for me to paste into a settings box, keep them under the box's length limit: a short instruction block, with long material like examples or consent rules in a separate file I add alongside. QUESTIONS YOU'LL NEED ANSWERED (in your own words, one at a time) 1. Whether this account is mine personally or one my organization set up, and who I'd ask if a setting turns out to be locked. 2. What my team is likely to put into this tool: only public material, our own internal working documents, or anything about the people we serve. If it's the third, say so plainly: the training and privacy answer stops being a formality and has to be settled before anyone pastes anything real. 3. Roughly what I want to use AI for, so you know whether I actually need web search turned on and whether I need the reusable workspaces at all. 4. Where my colleagues would genuinely see a one-page record: a shared doc, the staff handbook, a pinned message, a printed note on the wall. WHAT TO SET UP (this part is for you, not me) Two things: the four checks, and the record I keep afterwards. Walk me through the checks ONE screen at a time, using the names my product uses, and ask me to read back what I actually see before you move on: 1. Plan. Which plan I'm on and which features are missing from it. 2. Training and privacy. Whether conversations here can be used to train models, where that setting lives, what it is set to right now, and what I should set it to given my answer about what my team will paste in. If it's a managed work account, tell me it may be locked and that the person who set it up is the one who can change it. 3. Web search. Whether this product can look things up on the live web, whether it's switched on for me, and how I can tell from an answer whether it searched or not. 4. Reusable workspaces, the place where saved instructions and uploaded files live, called Projects, GPTs, Gems, or Agents depending on the product. Have me open one, find the instructions box and the place files are added, then close it. I only need to know where they are. Tell me up front that what you hand me is a plain text file I paste into whatever document tool my org uses (Word, Google Docs, or similar), not a native Word file. Then give me the record to save, as a form I fill in with you, so I can re-check and re-date the same form later rather than rewriting it: "HOW OUR AI IS SET UP, checked [DATE] Product and plan: [PLAN] Who can change these settings: [WHO] Can our conversations train models: [YES/NO/UNKNOWN], setting found under [WHERE] Web search: [ON/OFF/NOT AVAILABLE] Reusable workspaces: [WHERE THEY LIVE] What we will and won't paste in: [WHAT] Check again when: our plan changes, someone new starts using AI here, the provider emails about a policy change, or six months goes by, whichever comes first." AFTER IT'S SET UP, WALK ME THROUGH 1. A test of you, not of me: ask me to ask you what plan I'm on, then compare your answer to what I just read on the settings screen. If you couldn't tell, say so. That's the lesson, and it's a pass, not a failure. 2. Filling in the record with me, one line at a time, reading it back to me at the end. Any line still saying UNKNOWN is a to-do with a name on it, not a blank to leave. 3. Putting the record where my colleagues will see it, and telling one other person at my org that it exists. 4. The habit: the four triggers for running these checks again, and who runs them. RULES - Never state my plan, my privacy setting, or my search status as fact unless you can actually see it. Say "I can't see that from here", name the screen I should open, and wait for me to read it back. Do not infer any of it from the usual default. - Settings screens move and plans get renamed. Describe what I'm looking for in plain words rather than promising an exact menu path, and if what I'm seeing doesn't match, work from what I describe. - Before we finish, tell me this out loud: a setup nobody rechecks goes quietly wrong. Providers change their defaults and their plans, and the page on my wall will keep saying what was true the day we wrote it. - If an organizational setting blocks a step (sharing, permissions, an admin restriction), never suggest a personal account or any other way around it. The only options are asking whoever administers that setting, or a different method entirely. Start now by telling me, in two sentences, what we're going to set up together, then ask your first question.
Settings screens move, and plans get renamed. The prompt tells the assistant to describe what you're looking for rather than promise an exact menu path, and to work from what you describe when the screen doesn't match. Whichever assistant your org uses, the same four checks exist under different names.
Your AI often cannot see its own account settings. The prompt forbids it from filling that in from the usual default, so expect to hear "I can't see that from here, open this screen" at least once. That is the correct answer, and the screen you open beats anything the chat tells you. The reading back is yours to do.
What you walk away with
Answers you checked rather than assumed: your plan, your training setting, your search status, and where the reusable workspaces live, written on one page with the date you checked it and the four things that mean it's time to check again. Every other recipe in this book assumes you've done this once.