Read the Notes Your AI Keeps on You

Wednesday, 7:40 a.m. Coffee in one hand, phone in the other, you ask your AI assistant to plan a quick week of dinners. It cheerfully suggests meals for a family of three, built around a diet you tried for two weeks last spring, scheduled around a job you left in June. You laugh. Then you pause. Wait, how does it know that? And what else is in there?

AI adoption doesn’t stall because your team can’t use a chat assistant. It stalls when people don’t trust what the assistant knows about them, or when its notes are so out of date that the answers stop fitting. For SMEs building generative AI for business workflows, a five-minute monthly memory check is an easy first rep: ask your AI “What do you remember about me?”, delete anything wrong or private, then add three things that matter now. Better answers. No creepy surprises.

This is Tip 28, a life-first framing of the “what does my AI know about me?” conversation people are having on X right now. It’s distinct from our standing-instructions post (you write the setup once), Tip 27 (a voice card for how drafts sound), and Tip 16 (voice dump → evening free). Tip 28 isn’t about instructions you write or the style of your drafts. It’s about the notes the AI keeps on its own: ask what it remembers → delete wrong or private → add three current truths → repeat monthly.

Stale AI memory doesn’t just give worse answers. It erodes trust.

Many AI assistants now offer some form of memory or personalization: details you mentioned once, preferences inferred from past chats, facts carried from one conversation to the next. When those notes are right, answers get noticeably better. When they’re wrong, outdated, or more personal than you expected, two things happen. The suggestions drift away from your real life, and you start to feel uneasy about using the tool at all.

Here’s what actually happens. You never check the notes, because nobody told you to and the setting lives somewhere you’ve never opened. The old job, the abandoned diet, the client you mentioned once, and a detail about your kid’s school all stay in there. One day an answer surfaces something you forgot you ever typed, and the whole tool feels a little less safe.

Tip 28 isn’t a new privacy app. It’s a five-minute monthly habit, using the AI assistant you already have, and you stay the person who decides what it keeps.

Why the “what does my AI know?” stories matter this week

Two trending stories put the question in front of everyone at once. A TIME investigation, as summarized in an X trending story, reports that Meta’s Muse AI agent maps relationships and tensions from users’ chats, messages, and emails, and even builds profiles of non-users mentioned by others. Meta says the data stays isolated and that users have controls to inspect, edit, or delete memories (X trending: Meta’s Muse AI builds secret dossiers on users and contacts). That’s the reporting and Meta’s response as summarized on X; the takeaway for you is simply that assistant memory is real, and the controls only help if you use them.

The lighter version of the same story: Amazon’s new “About You” shopper profiles, tied to Alexa for Shopping, are built from purchase history and searches, and users are sharing screenshots of spot-on guesses mixed with wildly wrong ones, like one shopper who buys for kids being profiled as four foot eleven. Users can view, edit, or delete those details (X trending: Amazon’s AI shopper profiles deliver laughs with wild guesses). Funny, and also the point of this tip: AI notes are often wrong, and you can fix them.

And one practical post sums up the habit. A software company’s account lists “3 AI memory settings you should check today,” starting with reviewing your AI app’s memory (view, edit, delete) and using temporary or incognito chats for private stuff (@thenanosoft). That’s their advice; Tip 28 turns the first two into a monthly routine.

You don’t need to stop using AI or read a privacy policy end to end. You need five minutes a month to read the notes, clear what shouldn’t be there, and add what should.

The three-step habit (ask · delete · add three)

Here’s the version I teach in corporate AI training. It works with ChatGPT, Claude, Gemini, Copilot, or whatever assistant your team already uses. One honest caveat first: memory features, what they’re called, and where they live differ by tool, plan, and company setup, and they change often. Some assistants don’t keep memory at all. So look for your AI’s memory or personalization settings, or search your tool’s help page for “memory,” rather than trusting a menu path from a blog post.

1. Ask: what do you remember about me?

Open a fresh chat and ask:

“What do you remember about me? List every saved detail or preference you’re using to personalize your answers, grouped into work, family, health, money, and other. Include anything you’ve inferred, not just things I told you directly. Don’t guess or add new details.”

Then open the memory or personalization settings and compare. The settings page is usually the more complete list; the chat answer is a quick, readable summary, not a guarantee.

2. Delete anything wrong, outdated, or private

Go through the list and remove three kinds of notes: wrong (the job you left, the diet you quit), outdated (last quarter’s project, an old schedule), and private (health details, money, account or client information, and anything about other people who never agreed to be in your AI’s notes, like a colleague’s situation or your kid’s school). If your company has an approved AI tool or a rule about what assistants may store, follow it. Deleting a memory doesn’t always delete past chats, so check your tool’s help page if that matters to you.

3. Add three things that matter now

Now give it three current truths that make answers fit your real life, in plain sentences: your working hours (“I work 8 to 4 and keep evenings free”), your context (“I run a 12-person accounting firm; my clients are small businesses”), and this quarter’s goal (“My Q4 focus is hiring one bookkeeper”). Add them in memory settings if your tool allows it, or tell the assistant directly and ask it to remember them. Three is enough. You’re correcting the picture, not writing a biography.

Step	What you do	Why it helps
Ask	“What do you remember about me?” plus a look at memory settings	You see what it’s actually working from
Delete	Remove wrong, outdated, and private notes	Fewer odd answers; nothing sensitive lingering
Add three	Hours, context, this quarter’s goal	Answers fit the real you right now

Never-checked memory	Monthly memory check
Answers built on an old job and old habits	Answers built on this month’s reality
“Wait, how does it know that?”	You’ve already seen the list
Private details sit there indefinitely	Private notes deleted on a schedule
Other people’s details stored by accident	Only your own context, kept on purpose
Quiet unease about using the tool	Confidence to use it more

Two habits make the check easier next month. Use a temporary or incognito chat, if your tool offers one, for private one-offs you don’t want remembered. And when you switch assistants or turn on a memory import, review what came across before you rely on it.

Team version (a shared page on what the company AI should and shouldn’t remember)

If you run a team, write one short page together: what our AI should remember (roles, working hours, approved client-facing tone, current priorities) and what it shouldn’t (client names and confidential details, health or HR information, passwords and account numbers, anything about people outside the conversation). Keep it next to your approved-tools list, and revisit it when you change tools or plans.

Then run a ten-minute memory check in your next AI training session: everyone asks the question, reads their list, deletes against the page, and adds their three truths. People trust AI when they can see what it knows, and that trust is what turns a pilot into daily use. It’s also one of the simplest AI-fluency habits there is: knowing what your assistant holds and how to correct it.

Why SMEs win when people can see what the AI knows

In a small business, one uneasy moment can stall adoption for a whole team. Someone sees an answer that references something it shouldn’t, mentions it at lunch, and suddenly nobody wants to use the tool. Tip 28 doesn’t need a privacy officer or a new platform. It needs five minutes a month and a shared page that says what belongs in the notes.

Deleting without adding leaves the AI guessing. Adding without deleting piles fresh context on top of stale and private notes. Put them together and AI adoption becomes something you can feel within a month: answers that fit your real hours and goals, fewer surprises, and a team that trusts the tool enough to use it. You’re in charge of the notes. The habit is simply ask → delete wrong or private → add three → repeat monthly.

A 30-day memory-check sprint

Week 1: Your own check. Ask the question, compare with your memory settings, delete what’s wrong or private, add your three truths. Note one answer that got better.

Week 2: Try a temporary chat. Use a temporary or incognito chat, if your tool has one, for one private question. Get comfortable with when to use it.

Week 3: Write the team page. Draft “what our AI should and shouldn’t remember” on one page. Keep it short enough to read in two minutes.

Week 4: Teach it once. Use corporate AI training to run the ten-minute check with the team against the page. Put a monthly reminder on the calendar (first Monday works well); drop anything that felt like busywork.

Soft next step

If your team has AI tools but people are uneasy about what those tools know, or the answers keep missing the real business, the gap is usually the habit (visible memory, a clear page on what belongs, a human who checks monthly), not another app. I’m Nathan Graham, founder of Synthetic Echo, Toronto-based · serving Canada & North America. Book a free discovery call and we’ll work out where AI memory and personalization help your business most, what should never be stored, and how training can make AI adoption stick. It’s a prioritization conversation, not a free full audit.