FixKit AI

Learn with Kit / Personal Assistant

Your AI personal assistant

Kit is wearing a tiny headset and carrying a clipboard, handing out jobs to three smaller Kits. 'I'm the chief of staff. Chef Kit, you're on dinner.'

Turn AI into your own chief of staff: one boss bot, a few small helpers with one job each, a 5-line morning note that starts with what changed, receipts for everything, and nothing happens without your yes

  1. Build your first helper in 10 minutes

    Open an AI app you already have (Grok, ChatGPT, Gemini, Claude, Perplexity, Muse or Hark). Step 1: type a name and one job, like "Your name is Sunny. Tell me the weather and what to wear." Step 2: ask it to do that every day at a time you pick, and tap Allow for notifications. Step 3: test it right now and tell it what to change. Kids: build it with a grown-up.

  2. One chief of staff, a few small helpers

    Give each helper a name and one job: Chef for meals, Encore for shows, Scout for sports, Penny for bills. A chief of staff bot hands out the work and gives you one short report. Grok Bot can run several bots as a team (on paid plans; plain Grok chat is free); in other apps, use separate projects, Gems or labeled jobs.

  3. Morning notes that lead with what changed

    Things change every day, so line 1 of every note is "Changed since yesterday." Keep it to about 5 lines, a link on every line, at a time you actually look at your phone. Use ChatGPT Tasks, Gemini scheduled actions, Grok Automations or Grok Bot routines, Perplexity Scheduled Tasks or Claude Cowork.

  4. No receipt, no done. No yes, no go.

    Helpers show proof for everything: a link, a quote, a screenshot or a call transcript. They never buy, book, pay, send or post until they read it back to you and you say yes to that exact thing.

  5. Calls, texts and staying safe

    With a phone number (Speko is a ready-made Grok Bot plugin; pick your local area code) a bot can call to book tables and appointments, one call at a time, after your OK. Bot texting in the US needs 10DLC sign-up, or it can draft texts from your own number on a Mac. Never paste passwords, card, account or Social Security numbers, and know how to turn routines off.

Your one-minute AI challenge

Open any AI assistant (Grok Bot, Claude, ChatGPT, Grok or Gemini) and paste this. Fill in the [brackets].

You are my chief of staff. My helpers are: [Chef: dinners and grocery lists], [Penny: bills and due dates], [Juniper: the family calendar]. Every day at [7:00 am], send me one note of 5 lines or less. Line 1: what changed since yesterday. Put a link or receipt on every line. My town is [town, state]. Rules: never buy, book, pay, send or post anything until you read it back and I say yes to that exact thing; never ask for passwords, card, bank or Social Security numbers. Start by doing a test note right now, then tell me exactly where to pause or delete this routine.

The full guide

This guide shows you how to turn AI into your own personal assistant, a bit like having a helpful chief of staff (the person who keeps a busy boss organized). You build one main helper bot that is in charge, plus a few small helper bots that each do just one job, like meals, music and events, sports, or bills. Each one sends you a short morning note that starts with what changed since yesterday, shows proof for everything it says, and never buys, books, pays, sends or posts until you say yes. Grok Bot is the worked example because it is what this site's owner uses every day, but every step also shows how to do it with ChatGPT, Gemini, Claude, Perplexity, Muse or Hark Pro; this is a how-to, not advice, and new to AI? Start with Your AI toolkit (fixkitai.com/learn/ai/).

๐Ÿ‘ต New to this? Start here

  1. Pick one AI app you already have (Grok, ChatGPT, Gemini, Claude, Perplexity, Muse or Hark) and open it on your phone.
  2. Type: "Be my morning helper. Every day at 7:30 am, send me 5 short lines: the weather, what's on my calendar, and anything that changed since yesterday. Put a link next to every fact."
  3. Give your helper a name and one job only. A name makes it easy to say "Chef, what's for dinner?" or "Penny, which bills are due?"
  4. Make one rule and write it into the bot's instructions: "Never buy, book, pay, send or post anything until I say yes to that exact thing."
  5. Ask for receipts: "Show me a link, a quote or a screenshot for anything you tell me." No receipt, no done.
  6. Never paste a password, card number, bank account number or Social Security number into any AI chat. Kids: build your bot with a grown-up.
๐Ÿ› ๏ธ Fix-it-yourself guides8 guides
๐ŸŽฌ Real-life scenarios10 stories

Real situations โ€” many from my own life, names removed โ€” showing exactly what was asked and which AI did what.

Atlas, the chief of staff for a busy family

๐Ÿ‘ค A parent of three with a full-time job

The problem: Every week something slipped: a form, a bill, a birthday. Five apps and three group chats, and nobody had the whole picture.

See what they did
  1. Set up Grok Bot and made a chief of staff bot named Atlas.
  2. Added three helpers, each with one job: Chef (meals), Penny (bills), Juniper (family calendar).
  3. Wrote the rules once: lead with what changed, 5 lines, a receipt for every fact, nothing bought or sent without a yes.
  4. Gave Atlas a 7:00 am daily summary, and gave the helpers weekly times.
  5. Every Sunday asked Atlas: 'Did any helper break a rule this week?'

What happened: After a month, one 5-line note each morning replaced checking five apps. Two missed forms and one late fee were caught early. Every purchase was still approved by the parent.

Grok BotGoogle CalendarEmail connector

๐Ÿ’ก One boss bot, small helpers with one job each, and one short report. You stay the one who says yes.

Grandma builds Sunny, her morning helper

๐Ÿ‘ค A 67-year-old grandmother who had never set up an app routine

The problem: She wanted to know the weather, what to wear, and whether anything was on her calendar, without reading a whole news site.

See what they did
  1. Opened the free Gemini app her grandson had put on her phone.
  2. Typed: 'Your name is Sunny. Your only job is the weather, what to wear, and my appointments.'
  3. Asked: 'Every day at 8 am, send me 3 lines in big simple words.' Tapped Allow when the phone asked about notifications.
  4. Ran a test right away, then said: 'Shorter. Weather first.'
  5. Wrote the steps on an index card and kept it by her phone.

What happened: Sunny's note arrives every morning at 8. It once said 'Changed: your eye doctor moved to Thursday,' which she had forgotten. She showed her friend from church how to make her own.

GeminiGemini scheduled actionsGoogle Calendar

๐Ÿ’ก If you can send a text, you can build a helper. Start with one job and three lines.

Grandpa and Ivy build Professor Paws

๐Ÿ‘ค A grandpa and his 7-year-old granddaughter, building together

The problem: Ivy loves animals and kept asking for 'one more fact.' Grandpa wanted something fun and safe they could make together.

See what they did
  1. Grandpa opened ChatGPT on his own phone, with Ivy sitting next to him.
  2. Ivy picked the name: Professor Paws. Its only job: one animal fact and one silly question every morning.
  3. Grandpa set the routine for 7:15 am on school days and checked there were no connections to email, calendar or shopping.
  4. They tested it together and Ivy said 'more jokes, please!' so Grandpa typed that in.
  5. Grandpa showed Ivy how to say 'Professor Paws, stop' and how he can pause it in Tasks.

What happened: Every school morning, Grandpa reads Ivy the fact on a video call. She has learned 40 animal facts and taught her class about octopus hearts.

ChatGPTChatGPT TasksGrandpa's phone

๐Ÿ’ก Kids build with a grown-up, on a grown-up's account, with fun jobs and no connections.

Leo's homework helper, with Mom in charge

๐Ÿ‘ค A 10-year-old and his mom

The problem: Leo forgot what homework was due and which days he needed gym shoes. Mom didn't want him alone in an AI app.

See what they did
  1. Mom made a helper called Captain Clock in Claude on her account, with Leo choosing the name.
  2. Its job: each school night at 6:30 pm, list tomorrow's homework and what to pack, from a list Mom pastes in on Sundays.
  3. Rule: help Leo plan and quiz him, but never do the homework for him.
  4. Mom reads the note first, then shows Leo.
  5. On Fridays, Captain Clock gives Leo a 5-question spelling quiz.

What happened: Leo stopped forgetting gym shoes. Mom liked that the helper quizzed him instead of giving him answers.

Claude ProjectsMom's calendar

๐Ÿ’ก A kid's helper should teach and remind, not do the work. A grown-up reads it first.

Hal's bot books the barber by phone

๐Ÿ‘ค A retired man who dislikes being put on hold

The problem: Booking a haircut meant calling during work hours and waiting on hold. He wanted the bot to call for him.

See what they did
  1. Asked his Grok Bot to connect Speko, signed in to Speko himself and picked a number with his own area code.
  2. Had the bot call his cell phone as a test and read the transcript.
  3. Pasted the 8 phone rules into the bot's instructions.
  4. Asked it to call the barber for 'any time Tuesday or Wednesday morning,' read back the plan, and said yes.
  5. Got a report with the transcript: 'Wednesday 10:30 am, booked under Hal.' Then said yes to adding it to his calendar.

What happened: The haircut was booked in 4 minutes without Hal on hold. When the shop asked for a card to hold the spot, the bot said 'He'll call back,' and Hal did.

Grok BotSpekoGoogle Calendar

๐Ÿ’ก Test on your own phone first, give the bot clear phone rules, and keep payment details for yourself.

Penny watches the bills, Sam makes the call

๐Ÿ‘ค A young couple watching their spending

The problem: Their internet and phone bills crept up, and they never found time to call.

See what they did
  1. Made a helper called Penny in ChatGPT with one job: a Monday note of bills due this week and any price increases.
  2. Penny flagged: 'Changed: internet bill is higher than last month (link to the bill email).'
  3. Asked Penny to find current new-customer deals and two competitors' prices, with links.
  4. Penny wrote a short script asking about promos and retention offers, and a stop rule for account verification.
  5. Sam made the call himself with the script.

What happened: The company offered a lower price for 12 months. Sam accepted it himself and saved the confirmation email as the receipt.

ChatGPTChatGPT TasksEmail connector

๐Ÿ’ก Let the AI do the homework and the script. You say yes to any new deal.

Encore finds shows, Muse handles the errand

๐Ÿ‘ค A music lover who uses WhatsApp all day

The problem: She kept missing small local shows and only heard about them after they sold out.

See what they did
  1. Messaged Muse inside WhatsApp: 'You're Encore. Every Thursday at 6 pm, send me 5 shows near me this weekend under my budget, with links.'
  2. Added: 'Line 1 is what changed: new shows, sold out, or price changes.'
  3. When she liked a show, asked Muse to get the tickets to the checkout page and stop.
  4. Muse showed the read-back: show, date, seats, total with fees. She approved it herself.

What happened: She went to 3 shows in a month she would have missed. Every purchase waited for her yes.

MuseWhatsAppOfficial venue sites

๐Ÿ’ก Use the app you already live in. The helper finds and fills; you approve the total.

Mia's soccer-team texts from her own number

๐Ÿ‘ค A volunteer team parent

The problem: Practice times kept changing and she spent evenings texting 14 parents one by one.

See what they did
  1. Signed in to Messages on her Mac with her own phone number.
  2. Connected her assistant to the Mac and pasted in the texting rules: only people I name, show me every text first.
  3. When practice moved, asked: 'Draft a text to the team parents saying practice moved to 5 pm.'
  4. Read the draft, changed one word, and said yes.

What happened: Updates went out in 1 minute instead of 20. No text went out without her reading it first.

Grok Bot desktop appApple Messages on Mac

๐Ÿ’ก Drafts from your own number are the simplest way to text. You see every word first.

Ana sorts a 200-email inbox

๐Ÿ‘ค A small-business owner buried in email

The problem: Important emails were hiding under ads and newsletters, and she was missing replies.

See what they did
  1. Connected Gmail to Gemini (she signed in herself).
  2. Asked: 'Look at the last 2 days. The 5 that need me, in 5 lines, with links.'
  3. Asked for short draft replies, and sent them herself after fixing one.
  4. Asked Gemini to help her make a Gmail filter that moves ads into a folder.
  5. Set a scheduled action for 8 am weekdays: 'what needs me today.'

What happened: Her morning email time dropped from 45 minutes to about 10. She checks the filter folder on Fridays.

GeminiGmailGmail filters

๐Ÿ’ก Let the helper sort and draft. You send.

The off-switch drill

๐Ÿ‘ค A family that set up several helpers

The problem: They had 6 routines running and weren't sure how to stop them if something went wrong.

See what they did
  1. Each family member asked their assistant: 'Tell me exactly how to pause, delete and stop all your routines.'
  2. Practiced pausing one routine and turning it back on.
  3. Opened Settings and turned off connections helpers didn't need.
  4. Added a monthly calendar reminder: 'Check AI connections and routines.'

What happened: They turned off 2 routines nobody read and removed email access from a music helper. Everyone knows where the off switch is.

ChatGPT TasksGemini scheduled actionsGrok BotPerplexity Scheduled Tasks

๐Ÿ’ก Know the off switch before you need it. Fewer, simpler helpers are safer and more useful.

Build your first helper bot in 10 minutes3 topics

You don't need to be good with computers for this. If you can send a text message, you can build a helper bot. You will pick an app, give your helper a name, give it one job, pick a time, test it, and then fix anything you don't like. Set a kitchen timer for 10 minutes and let's go.

Minutes 1 to 3: pick an app and give your helper a name

Open an AI app you already have. Tell it a name and one job. That's your first helper bot.

๐Ÿ“– Example

June, 66, had the free Gemini app on her phone. She typed: 'Your name is Sunny. Your only job is to tell me the weather and what to wear each morning.' Sunny answered, 'Hi June, I'm Sunny!' It took her 2 minutes.

More + copy-ready prompt

Step 1: Open the AI app on your phone. It looks like a chat screen with a box at the bottom that says something like "Ask anything" or "Message." If you don't have one yet, the free Grok, ChatGPT, Gemini, Claude, Perplexity or Hark apps all work for this first bot. Step 2: Tap the box at the bottom. Your keyboard pops up. Step 3: Type a name and one job, like: "Your name is Sunny. Your only job is to tell me the weather and what to wear." Step 4: Tap the arrow or Send button. The AI writes back and says hi. Step 5 (optional, for keeping it): Many apps let you save a helper so you can find it again. In ChatGPT these are called projects or GPTs (custom helpers). In Gemini they're called Gems (saved helpers with your instructions). In Claude they're called Projects. In Grok Bot (a paid add-on to Grok, different from the free Grok chat), you tap "New Bot," give it a name, and it stays on all the time. Muse and Hark remember you on their own. Look for a button that says New, Create or + near the top of the screen. If you can't find it, just ask the AI: "How do I save you as a helper so I can find you tomorrow?"

๐Ÿค– How AI helps: Copy the prompt, change the name and the job, and tap Send.

โœ… Your first step: Open your AI app right now and type just this: "Your name is Sunny. Say hi."

Your name is [Sunny]. You are my helper. Your only job is to [tell me the weather in my town and what to wear]. My town is [town, state]. Talk to me in short, simple sentences. If you are not sure about something, say so. Never buy, book, pay, send or post anything. Say hi and tell me in 2 lines what you will do for me.

Grok (free chat) โ†— ยท xAI: Introducing Grok Bot โ†— ยท ChatGPT โ†— ยท Gemini โ†— ยท Claude (Anthropic) โ†— ยท Perplexity โ†— ยท Muse (personal AI agent) โ†— ยท Hark โ†—

Minutes 4 to 6: set a time so it comes to you

A real assistant comes to you. Pick a time, and your helper sends you its note every day without you asking.

๐Ÿ“– Example

Leo, 9, and his dad set up a helper called Captain Clock in ChatGPT. Every school day at 7:00 am it sends: 'Good morning Leo! It's rainy, bring your coat. Today is library day, bring your book.' His dad checks it each morning too.

More + copy-ready prompt

Step 1: In the same chat, type: "Every day at 7:30 am, send me the weather and what to wear." Step 2: Tap Send. Many apps will now make a scheduled routine (a job that runs by itself at the same time every day). You'll usually see a small box or card that says something like "Task created," "Scheduled" or "Automation," with the time on it. Step 3: Check where routines live in your app. ChatGPT: Tasks (in the menu, or ask "show my tasks"). Gemini: scheduled actions (Google's help calls them Schedules). Grok: Automations for the free chat, and routines inside each Grok Bot on paid plans. Perplexity: Scheduled Tasks. Claude: scheduled tasks in Cowork (Claude's helper that works on your computer, on paid plans). Muse and Hark: ask in plain words and check what it tells you back. Step 4: Make sure your phone will show it. When the app asks "Allow notifications?" tap Allow, or the note will arrive quietly and you'll miss it. If your app can't schedule things, that's okay: save the prompt in your Notes app and paste it in each morning.

๐Ÿค– How AI helps: Ask your AI to make the routine, then ask it to tell you exactly where you can see it and turn it off.

โœ… Your first step: Pick your time now. Most people like 15 minutes after they wake up, with their coffee or breakfast.

Every day at [7:30 am], send me a short note. Keep it to 5 lines or less. Line 1: anything that changed since yesterday. Then the weather in [town] and what to wear. Put a link next to every fact. After you set this up, tell me: the exact time it will run, where I can see it in this app, and how to pause or delete it.

Grok Automations โ†— ยท xAI: Introducing Grok Bot โ†— ยท OpenAI help: Tasks in ChatGPT โ†— ยท Google help: Schedules in Gemini โ†— ยท Perplexity help: Scheduled Tasks โ†— ยท Claude Cowork โ†— ยท Muse (personal AI agent) โ†— ยท Hark โ†—

Minutes 7 to 10: test it, then change it

Don't wait until tomorrow. Ask your helper to do its job right now, read it, and tell it what to change.

๐Ÿ“– Example

June's first test from Sunny was 12 lines long. She typed: 'Too long. 3 lines. Bigger words first.' The next test was 3 lines: 'Cold, 41 degrees. Coat and gloves. Rain after 3 pm, bring an umbrella.' Perfect.

More + copy-ready prompt

Step 1: Type: "Do your job now so I can see a test." Tap Send. Step 2: Read what it sends. Ask yourself 3 questions: Is it short enough? Is it right (look out the window, check one fact)? Is there a link I can tap to check? Step 3: Tell it what to fix, in normal words. For example: "Shorter." "Start with what changed." "Use bigger, simpler words." "Put the link at the end of each line." Step 4: Ask for one more test. Keep going until you like it. Step 5: Ask it to save your changes: "Remember these changes for every morning note." Step 6: Tomorrow morning, check that the note really arrived. If it didn't, open the app's routine list (see the card above) and check the time and that it's turned on. Changing your mind later is easy: just tell the helper what you want now.

๐Ÿค– How AI helps: Run a test right away, then fix it with short, plain instructions.

โœ… Your first step: Run your first test before you put your phone down.

Do your job right now as a test. Then ask me 3 questions: Was it short enough? Was anything wrong? What should you change? Wait for my answers, make the changes, and show me the new version. Then remember the new version for every future note.

xAI: Introducing Grok Bot โ†— ยท ChatGPT โ†— ยท Gemini โ†— ยท Claude (Anthropic) โ†— ยท Perplexity โ†— ยท Muse (personal AI agent) โ†— ยท Hark โ†—

One chief of staff bot, many small helpers3 topics

One giant helper that does everything gets confused. A team works better: a few small helpers, each with a name and one job, and one chief of staff bot (the boss bot) that hands out work and gives you one short report. Start with 2 helpers. Add more only when the first ones work well.

Make a little team: one name, one job each

Write a short list: helper name, its one job, and what it should never do. Then build them one at a time.

๐Ÿ“– Example

Nina set up 4 helpers: Chef (dinners and grocery lists), Encore (concerts and shows near her), Scout (her favorite teams' games) and Penny (bills and due dates). Each one does one thing, so each one gets better at it.

More + copy-ready prompt

Step 1: Get a piece of paper or open your Notes app. Step 2: Write 2 to 4 lines like this: "Chef: plans dinners and makes the grocery list. Never orders food." Step 3: Build the first helper using the 10-minute steps above. Step 4: Build the next one the same way. In Grok Bot, each helper is its own bot: tap New Bot, name it, and paste its job. In ChatGPT, use a separate project or GPT (custom helper) for each. In Gemini, make a Gem (saved helper) for each. In Claude, make a Project for each. In Perplexity, use a Space (a saved folder with instructions) for each. Muse and Hark are one assistant each, so give them labeled jobs instead: "When I say 'Chef,' you're my dinner helper." Step 5: Give each helper a nickname you'll remember. A funny name helps kids and grandparents remember who does what.

๐Ÿค– How AI helps: Ask your AI to help you design your little team before you build it.

โœ… Your first step: Write your team list on paper: 2 names, 2 jobs. That's enough to start.

Help me plan a small team of AI helper bots. My busy areas are [meals, kids' school, bills, fun weekend plans]. Suggest 3 helpers. For each one give: a friendly name, its one job in one sentence, what time it should check in, and one thing it must never do. Keep it simple enough for a beginner. Don't set anything up yet.

xAI: Introducing Grok Bot โ†— ยท ChatGPT โ†— ยท Gemini โ†— ยท Claude (Anthropic) โ†— ยท Perplexity โ†— ยท Muse (personal AI agent) โ†— ยท Hark โ†—

Make one chief of staff bot that routes the work

The chief of staff is the boss bot. You talk to it; it sends each job to the right helper and reports back.

๐Ÿ“– Example

Omar told his chief of staff bot, Atlas: 'Saturday is my mom's birthday.' Atlas asked Chef for a dinner idea, Encore for a show nearby, and Penny whether it fits the budget. Then it sent Omar one short plan with links. Omar approved the dinner and skipped the show.

More + copy-ready prompt

Step 1: Build one more helper and name it something like Atlas, Boss or Captain. Step 2: Give it this job: "You are my chief of staff. You know my helpers and their jobs. When I ask for something, decide which helper should do it, and send me one short combined answer." Step 3: Paste your team list into its instructions so it knows who does what. Step 4: Test it: "Plan Friday night for 4 people under [my budget]." Step 5: Check the answer says which helper did each part. How each app does this: xAI says Grok Bot can run several bots at once, often with one chief of staff bot on top, and the bots share one cloud computer (a computer on the internet that your bots use). In ChatGPT, Gemini, Claude, Perplexity, Muse and Hark you usually have one assistant, so the "helpers" are just sections of one set of instructions, or separate saved projects that you visit yourself. That's fine. The idea is the same: one job per helper, one report for you.

๐Ÿค– How AI helps: Paste your team list into the chief of staff's instructions, then give it a small test job.

โœ… Your first step: Give your chief of staff one small test job today, like planning tomorrow's dinner.

You are my chief of staff. My helpers are: [Chef: dinners and grocery lists], [Encore: shows and events near me], [Scout: my teams' games], [Penny: bills and due dates]. When I ask for something: 1) say which helper is handling each part, 2) give me one short combined answer, 3) show a link or receipt for every fact, 4) list anything that needs my yes before it happens. Never buy, book, pay, send or post without my yes for that exact thing.

xAI: Introducing Grok Bot โ†— ยท ChatGPT โ†— ยท Gemini โ†— ยท Claude (Anthropic) โ†— ยท Claude Cowork โ†— ยท Perplexity โ†— ยท Muse (personal AI agent) โ†— ยท Hark โ†—

Give each helper a favorites file (its memory)

Each helper keeps a short running list of what you like and don't like, so you never have to repeat yourself.

๐Ÿ“– Example

Chef's favorites file says: 'Two kids, one won't eat mushrooms. Tacos on Tuesdays. Under 30 minutes on school nights.' After 2 weeks, Nina never had to explain her family again.

More + copy-ready prompt

Step 1: Tell each helper: "Keep a favorites file about me. Add to it when I tell you something I like or don't like." Step 2: Each time you react, say it out loud in the chat: "We loved this one, add it." "Never suggest that store again." Step 3: Once a week, ask: "Show me your favorites file." Read it and fix anything wrong. Step 4: Ask it to delete things you don't want saved. Where memory lives: Grok Bot remembers what you like on its own. ChatGPT has Memory (you can see and delete it in Settings, Personalization). Claude has memory and chat search on many plans (Settings). Gemini can remember things you tell it to (Settings). Muse and Hark remember your preferences too. If your app doesn't remember, keep a note on your phone called "Chef favorites" and paste it in at the start.

๐Ÿค– How AI helps: Ask your helper to show you what it remembers, and fix it.

โœ… Your first step: Ask one helper today: "What do you remember about me?" You might be surprised.

Show me everything you remember about me for your job, as a short bulleted list. Mark anything you guessed instead of something I told you. Then ask me what to change or delete. Never save passwords, card numbers, account numbers, ID numbers or medical details.

xAI: Introducing Grok Bot โ†— ยท Claude help: memory and chat search โ†— ยท ChatGPT โ†— ยท Gemini โ†— ยท Muse (personal AI agent) โ†— ยท Hark โ†—

Morning notes that start with what changed3 topics

Things change every day: a game time moves, a bill comes in, a show sells out. So each helper sends a short note at a time you pick, and the first line is always what changed since yesterday. If nothing changed, it says so in one line. That keeps it short enough to read with your coffee.

Lead with what changed since yesterday

Tell every helper: the first line of your note is what's new or different since yesterday. Everything else comes after.

๐Ÿ“– Example

Scout's Thursday note started: 'Changed: Saturday's game moved from 1 pm to 4 pm (link).' Rob would have shown up 3 hours early. The rest of the note he skipped, and that was fine.

More + copy-ready prompt

Step 1: Open each helper. Step 2: Type: "From now on, line 1 of every note is 'Changed since yesterday:' and then what's new. If nothing changed, say 'Nothing new since yesterday' and stop." Step 3: Tap Send. Step 4: Ask it to remember yesterday's note so it can compare. Grok Bot keeps this on its own. In other apps, ask the routine to "compare to your last note in this chat." If your app forgets between days, ask it to put the date at the top so you can spot the difference yourself. Step 5: Check tomorrow: does line 1 really tell you something new?

๐Ÿค– How AI helps: Add this rule to every helper's routine.

โœ… Your first step: Add the "Changed since yesterday" rule to your first helper today.

From now on, start every note with 'Changed since yesterday:' and list only what's new, moved, cancelled, due or sold out, each with a link. If nothing changed, write 'Nothing new since yesterday' and stop there. After that, give me at most 4 more lines.

xAI: Introducing Grok Bot โ†— ยท Grok Automations โ†— ยท OpenAI help: Tasks in ChatGPT โ†— ยท Google help: Schedules in Gemini โ†— ยท Perplexity help: Scheduled Tasks โ†— ยท Claude Cowork โ†—

Pick times so notes don't pile up

Spread your helpers out. One note at breakfast, one at lunch, one in the evening is plenty.

๐Ÿ“– Example

Nina's schedule: chief of staff at 7:00 am, Penny on Mondays at noon, Encore on Thursdays at 6 pm, Chef on Sunday at 4 pm for the week's meals. Four notes a week from three helpers, plus one daily summary.

More + copy-ready prompt

Step 1: Write down when you actually look at your phone (breakfast, lunch break, after dinner). Step 2: Give each helper one of those times. Not every helper needs to be daily: bills can be weekly, events can be twice a week. Step 3: Tell each helper its time: "Send your note every Monday at noon." Step 4: Ask your chief of staff to send one daily summary that combines the others. Step 5: Look at your app's routine list (ChatGPT Tasks, Gemini scheduled actions, Grok Automations or each Grok Bot's routines, Perplexity Scheduled Tasks, Claude Cowork scheduled tasks) and check the times. Step 6: Pick quiet hours: "Never message me before 7 am or after 9 pm."

๐Ÿค– How AI helps: Ask your chief of staff to plan your schedule of notes.

โœ… Your first step: Turn off any routine that you skipped reading 3 days in a row. Less is more.

Here are my helpers: [list]. Here's when I look at my phone: [breakfast 7:15, lunch 12:30, after dinner 7:30]. Suggest a schedule so I get no more than 2 notes a day. Some helpers can be weekly. Add quiet hours: nothing before [7 am] or after [9 pm]. Show it as a small table, then ask me before setting anything up.

Grok Automations โ†— ยท OpenAI help: Tasks in ChatGPT โ†— ยท Google help: Schedules in Gemini โ†— ยท Perplexity help: Scheduled Tasks โ†— ยท Claude Cowork โ†— ยท Muse (personal AI agent) โ†— ยท Hark โ†—

About 5 lines, and a link for everything

Short notes get read. Long notes get skipped. Ask for 5 lines or less, each with a link you can tap.

๐Ÿ“– Example

Encore's note: 'Changed: 2 new shows added this month. Fri 8 pm jazz at the library, free (link). Sat 2 pm kids' puppet show, $8 (link). Nothing sold out. Want me to hold a reminder?' Five lines. Done in 20 seconds.

More + copy-ready prompt

Step 1: Tell each helper: "5 lines or less. One fact per line. A link at the end of each line." Step 2: Ask for plain words: "Write like you're texting a friend. No fancy words." Step 3: Ask it to end with one question if it needs your OK: "Want me to book it?" Step 4: Tap a link or two each day to make sure they really go to the right page. If a link is broken or goes somewhere odd, tell the helper: "That link was wrong. Only use official pages." Step 5: If a note is too long, just reply "shorter" and it should learn.

๐Ÿค– How AI helps: Give your helper the format once; it should follow it every day.

โœ… Your first step: Tap one link in tomorrow's note to check it works.

Your notes must be 5 lines or less. Line 1: what changed since yesterday. One fact per line, in plain words a child could read. Put a link at the end of every line, from the official source when possible. If you couldn't confirm something, write 'not confirmed.' End with one yes-or-no question if you need my OK.

xAI: Introducing Grok Bot โ†— ยท ChatGPT โ†— ยท Gemini โ†— ยท Claude (Anthropic) โ†— ยท Perplexity โ†— ยท Muse (personal AI agent) โ†— ยท Hark โ†—

Receipts and approval rules: no receipt, no done3 topics

Two rules keep your helpers honest and you safe. Rule 1: receipts (proof). If a helper says it did something, it shows you a link, a quote, a screenshot or a call transcript. No receipt, no done. Rule 2: approvals. A helper never buys, books, pays, sends or posts until you say yes to that exact thing, after it reads it back to you.

Ask for receipts every time

A receipt is proof: a link, a short quote, a screenshot, a confirmation number or a call transcript (the written words of a call).

๐Ÿ“– Example

Penny said, 'Your water bill is paid.' Ana typed, 'Receipt?' Penny answered, 'Sorry, I'm not sure. I only saw the bill email, not a payment.' The bill was still due. Asking for the receipt saved a late fee.

More + copy-ready prompt

Step 1: Add this rule to every helper: "Whenever you say something is done, true, booked or cheaper, show me a receipt: a link, a quote, a screenshot, a confirmation number or a call transcript." Step 2: When a helper says "done," reply with one word: "Receipt?" Step 3: Look at the receipt. Does it really say what the helper said? Tap the link. Step 4: If there's no receipt, treat it as not done. Step 5: Save important receipts. Ask the helper: "Put the confirmation number and link in my favorites file." Grok Bot, Muse, Hark and agent modes in ChatGPT, Gemini and Claude often show screenshots or step lists of what they did. Ask to see them.

๐Ÿค– How AI helps: Add the receipt rule to every helper's instructions.

โœ… Your first step: The next time any AI says "done," reply "Receipt?"

New rule: no receipt, no done. Whenever you tell me something is done, booked, paid, true or cheaper, show me proof: a link to the official page, a short exact quote, a screenshot, a confirmation number, or a call transcript. If you don't have proof, say 'not confirmed' instead of 'done.'

xAI: Introducing Grok Bot โ†— ยท OpenAI help: ChatGPT agent โ†— ยท Gemini agent โ†— ยท Anthropic: Claude in Chrome โ†— ยท Muse (personal AI agent) โ†— ยท Hark โ†—

The yes rule: nothing happens without your OK

Your helpers find, plan and draft. You press the button. Every buy, booking, payment, message or post needs your yes for that one thing.

๐Ÿ“– Example

Chef had a grocery cart ready: 'Cart total $52.10 at the usual store, pickup 5 pm. Say yes to place it.' Mia saw it had 3 bags of chips. She removed 2, then said yes.

More + copy-ready prompt

Step 1: Add this rule to every helper: "Never buy, book, pay, send, post, delete or sign up for anything until I say yes to that exact thing." Step 2: Add: "Before you ask for my yes, read it back to me: what, where, when, how much in total, and who it goes to." Step 3: Only say yes to the exact thing in front of you. A yes today is not a yes for tomorrow. Step 4: If the read-back is wrong, say "No, change it" and explain. Step 5: Check the app's own safety settings too. Muse comes back before it sends an email or makes a purchase. Hark asks before anything that needs your approval. Claude in Chrome and ChatGPT's agent ask before important actions. Grok Bot comes back when it needs your OK. Still, your own written rule is the best backup.

๐Ÿค– How AI helps: Add the yes rule and the read-back rule to every helper.

โœ… Your first step: Write "I press the button" on a sticky note by your computer.

Rules for you, always: 1) Never buy, book, pay, send, post, delete or sign up for anything until I say yes to that exact action. 2) Before asking, read it back: what, where, when, total cost with fees, and who it goes to. 3) A yes is for one action only. 4) If anything changes after my yes (price, time, place), stop and ask again.

xAI: Introducing Grok Bot โ†— ยท Meta: Introducing Muse โ†— ยท Hark โ†— ยท Anthropic: Claude in Chrome โ†— ยท OpenAI help: ChatGPT agent โ†— ยท Simon Willison: the lethal trifecta for AI agents โ†—

Let the chief of staff check the helpers' work

Before a helper's work reaches you, the chief of staff checks it: are there receipts, did it follow the yes rule?

๐Ÿ“– Example

Atlas caught Encore saying 'tickets $20' with no link. Atlas sent it back: 'Receipt?' Encore found the real price was $34 with fees. Omar only saw the right number.

More + copy-ready prompt

Step 1: Tell your chief of staff: "Before you show me anything from a helper, check it has a receipt for every fact and that nothing was done without my yes." Step 2: Ask it to mark each item: checked, not confirmed, or needs your yes. Step 3: Once a week, ask: "Did any helper break a rule this week?" Step 4: If a helper keeps making the same mistake, change its instructions or turn it off. In apps with one assistant (ChatGPT, Gemini, Claude, Perplexity, Muse, Hark), you can still ask: "Before you answer, check your own work and show me receipts." It's not perfect, so you still look at the receipts yourself.

๐Ÿค– How AI helps: Make checking part of the chief of staff's job.

โœ… Your first step: Ask your chief of staff on Sunday: "Did anyone break a rule this week?"

Before you show me anything from my helpers, check each item: Does it have a receipt (link, quote, screenshot, transcript)? Did anyone do something without my yes? Mark each line as 'checked,' 'not confirmed' or 'needs your yes.' Every Sunday, tell me if any helper broke a rule.

xAI: Introducing Grok Bot โ†— ยท NIST AI Risk Management Framework โ†— ยท Simon Willison: the lethal trifecta for AI agents โ†—

Give your bot a phone number so it can call for you3 topics

Some chores still need a phone call: booking a table, a haircut, a doctor's appointment, or asking 3 shops for a price. A helper bot can make those calls if it has its own phone number. With Grok Bot, the easiest way is Speko, a ready-made plugin (an add-on you switch on with a tap). Other assistants have their own ways, and some can't call yet.

Connect a phone number to your bot (Speko in Grok Bot)

Your bot needs its own phone number to make calls. Speko is a ready-made Grok Bot plugin: tap Connect, sign in, pick a number with your local area code.

๐Ÿ“– Example

Hal, 68, wanted his bot to book his barber. He asked his Grok Bot, 'Connect Speko so you can make phone calls.' He signed in, picked a number with his own area code, and had the bot call his cell phone as a test. It said, 'Hi Hal, this is your assistant testing the line.'

More + copy-ready prompt

Step 1: Open Grok Bot (this is the paid helper part of Grok, not the free chat). Step 2: Open your bot's plugins or connectors (add-ons that give your bot new powers). Find Speko and tap Connect. Or just type: "Connect Speko so you can make phone calls." Step 3: A Speko web page opens. Sign in or make your Speko account yourself. You set up and pay for this account on your own; the bot should never type your password or card for you. Step 4: Pick a phone number. Choose your own area code (the first 3 numbers, like your home phone) so local shops are more likely to pick up. Step 5: Go back to Grok Bot. Type: "Call my cell phone at [your number] as a test and say hello." Your phone rings. Answer it. Step 6: Ask for the transcript (the written words of the call) as your receipt. Why Speko: it already shows up as a ready-made plugin inside Grok Bot, so you skip the manual "MCP" setup (MCP, Model Context Protocol, is a techie way to plug tools into an AI by hand). Speko's docs also list a connector for Meta's Muse. Other assistants: some can call businesses for you in certain places and some can't yet. Check your app's help page or ask it: "Can you make phone calls for me? How?"

๐Ÿค– How AI helps: Have your bot call your own phone first. Never skip the test.

โœ… Your first step: Ask your bot: "Can you make phone calls? What do I need to set it up?"

Help me give you a phone number so you can call businesses for me. Tell me step by step what I need to tap and where, in very simple words. I will sign in and pay for any phone service myself; never type my password or card number. When it's connected, call my phone at [my number] as a test, say a short hello, and show me the transcript.

Speko โ†— ยท xAI: Introducing Grok Bot โ†— ยท Anthropic: Model Context Protocol (MCP) โ†— ยท Muse (personal AI agent) โ†— ยท Hark โ†—

Book a table, an appointment or a quote by phone

Once your bot has a number, it can call to book a table, a haircut, a tee time or a vet visit, or call around for prices.

๐Ÿ“– Example

Tess asked her bot to call 3 car shops for a price on new brakes. It came back with a table: Shop A $310, Shop B $260 (Friday), Shop C wouldn't quote by phone. Each line had the transcript. Tess called Shop B herself to book.

More + copy-ready prompt

Step 1: Tell your bot exactly what you want: who to call, what to ask for, your best times, and your limit. For example: "Call [restaurant] and book a table for 4 on Friday between 6 and 7:30 pm. Indoor. Name: [first name]." Step 2: Ask it to read the plan back before calling. Check it. Say yes. Step 3: Wait. The bot calls and then reports back. Step 4: Read the result and the transcript (receipt). Step 5: Ask it to add the booking to your calendar, after you say yes. Want to see a ready-made example? Someone built a public Grok Bot called "Call A Restaurant Reservation" (link below), and Brian Roemmele wrote a post about it. You can try it or copy its idea for your own bot. For quotes, ask it to call 2 or 3 places and put the answers in a small table. Muse and Hark can also run errands like bookings online; ask yours what it can do by phone.

๐Ÿค– How AI helps: Give the bot a clear script and a limit, then wait for the read-back.

โœ… Your first step: Try the public reservation bot once just to see how a bot call works.

Please call [business name] at [their phone number from their official site]. Goal: [book a table for 4 on Friday between 6 and 7:30 pm, indoors, under the name [first name]]. If that time is gone, ask for the closest time and don't accept anything yet. Before you call, read this plan back to me and wait for my yes. After the call, show me the result and the transcript. Don't give any card or account numbers.

Grok Bot: Call A Restaurant Reservation โ†— ยท Brian Roemmele's post about the reservation bot โ†— ยท Speko โ†— ยท xAI: Introducing Grok Bot โ†— ยท Muse (personal AI agent) โ†— ยท Hark โ†—

Phone safety rules for your bot

Bots on the phone need extra rules: read back, wait for OK, one call at a time, no voicemails unless you ask, and never share private numbers.

๐Ÿ“– Example

Ana's bot called a dentist. The office asked for her insurance member ID. The bot followed its rule: 'I can't share that. My owner will call you back.' Ana called back and gave it herself.

More + copy-ready prompt

Step 1: Paste these rules into your bot's instructions: read back the plan before every call and wait for my OK. Step 2: One call at a time. Report back after each one. Step 3: Say it's an AI assistant calling for you if asked, and always be polite. Step 4: Don't leave voicemails unless I asked. Step 5: Never give out a card number, bank account, Social Security number, password, PIN or insurance ID. If they need it, say "My owner will call back." Step 6: Only call businesses, at normal business hours. No calling friends or strangers who didn't ask for it, and never use it for sales or robocalls (automatic calls to lots of people). Step 7: Hang up and tell me if anyone asks for payment or seems pushy. Step 8: Show me the transcript of every call.

๐Ÿค– How AI helps: Paste the rules once; check the transcripts after each call.

โœ… Your first step: Paste the 8 phone rules into your bot before its first real call.

Phone rules for you: 1) Read back the plan before every call and wait for my OK. 2) One call at a time; report after each. 3) If asked, say you're an AI assistant calling for [first name]. 4) No voicemails unless I ask. 5) Never share card, bank, Social Security, password, PIN, or insurance numbers; say 'My owner will call back.' 6) Only call businesses during business hours. 7) Never agree to pay, buy or sign up for anything. 8) Show me every transcript.

Speko โ†— ยท FCC: Stop unwanted robocalls and texts โ†— ยท National Do Not Call Registry โ†— ยท CFPB: Fraud and scams โ†— ยท xAI: Introducing Grok Bot โ†—

Can my bot send texts too?3 topics

Yes, in two ways. Way 1: texts from the bot's own phone number. In the US, phone companies make you register first (it's called 10DLC). Way 2: drafts from your own number, using a Mac that's signed in to Messages, where you approve each text. Either way, you see every text before it goes out.

Way 1: texts from the bot's own number (10DLC)

Before a bot can text from a US local number, the phone companies need a one-time business sign-up called 10DLC.

๐Ÿ“– Example

Hal wanted his bot to text the barber, 'Running 10 minutes late.' Speko told him to finish the texting sign-up first. He filled in the form himself and waited a few days for the phone companies to approve it.

More + copy-ready prompt

Step 1: Know the word. 10DLC means "10-digit long code," a normal-looking local phone number used by an app. US phone companies require a sign-up before an app or bot sends texts from one. Step 2: Open Speko's SMS page (link below; SMS means plain text messages). It explains the steps. Speko says you can start the sign-up yourself from its dashboard (your account's control page), or ask their support. Step 3: Fill in the form yourself, never the bot. It asks for your business's legal name exactly as on your tax papers, a tax ID if you have one (sole proprietors, meaning one-person businesses, can often skip it), your address and a website. Step 4: Wait. Approval can take a few days. Step 5: Test by having the bot text your own phone first. Other assistants: most don't text from their own number yet. Muse lives inside WhatsApp, so you text it, and it can draft messages for you to approve.

๐Ÿค– How AI helps: Read Speko's SMS page first, fill the form in yourself, then test on your own phone.

โœ… Your first step: Open Speko's SMS page and read the first section.

I want you to be able to send texts from your phone number. Explain the 10DLC sign-up to me in very simple words, step by step, and tell me what information I'll need to fill in myself. Don't fill in any form for me. When it's approved, send a test text to my phone at [my number] and wait for my OK before texting anyone else.

Speko docs: SMS and 10DLC โ†— ยท Speko โ†— ยท The Campaign Registry (10DLC) โ†— ยท FCC: Stop unwanted robocalls and texts โ†—

Way 2: drafts from your own number on a Mac

If your bot is connected to a Mac that's signed in to Messages, it can write texts from your own number. You approve each one.

๐Ÿ“– Example

Mia asked her bot, 'Text the soccer group that practice moved to 5.' The bot wrote a draft and showed it: 'Hi all, practice moved to 5 pm today at the same field. See you there!' Mia changed one word and said yes.

More + copy-ready prompt

Step 1: On your Mac, open the Messages app and sign in with your Apple Account (the one your iPhone uses). Step 2: Make sure your phone number is turned on in Messages (Apple's help page below shows how). Step 3: Connect your bot to the Mac. Grok Bot has a Mac desktop app; Claude has a desktop app with Cowork; ask your assistant how it connects to your Mac. Step 4: Ask the bot to draft a text: "Draft a text to [name] saying [message]. Show me before sending." Step 5: Read the draft on screen. Change anything you want. Step 6: Say yes, or send it yourself with your own thumb. Speko can also turn on iMessage from the bot's own number as a separate add-on, if you ask them. Rule: you see every text before it goes.

๐Ÿค– How AI helps: Have your bot draft one text to yourself first.

โœ… Your first step: Sign in to Messages on your Mac if you haven't already.

Draft a text message to [person's first name] that says [message in your own words]. Keep it short and friendly. Show me the exact words and who it's going to. Don't send it until I say yes. If I say 'change it,' fix it and show me again.

Apple: Messages User Guide for Mac โ†— ยท Apple: Add your phone number in Messages โ†— ยท xAI: Introducing Grok Bot โ†— ยท Download Claude โ†— ยท Speko โ†—

Texting rules: only people who said yes

No texting people who didn't ask for it. No group blasts. You see every message first.

๐Ÿ“– Example

Rob's bot offered to text the whole neighborhood about a lost dog. Rob said no, and posted it in the neighborhood group himself. Texting lots of strangers can break the rules and annoy people.

More + copy-ready prompt

Step 1: Paste these rules into your bot: only text people I name, who already know me. Step 2: Never text a list of people at once unless I say yes for that exact list. Step 3: Never text before 8 am or after 9 pm. Step 4: Show me every text before it's sent, every time. Step 5: Never put card numbers, account numbers, codes or passwords in a text. Step 6: If someone texts back "stop," stop texting them and tell me. Step 7: Never click links in texts people send back; tell me about them instead. Scam texts are common (see the CISA and FCC links).

๐Ÿค– How AI helps: Add the texting rules before any real text.

โœ… Your first step: Text yourself a test from your bot before anyone else.

Texting rules for you: 1) Only text people I name who already know me. 2) No group texts without my yes for that exact list. 3) Only between 8 am and 9 pm. 4) Show me every text before sending. 5) Never include card, account, code or password information. 6) If anyone replies 'stop,' stop and tell me. 7) Never open links people text back; tell me about them.

FCC: Stop unwanted robocalls and texts โ†— ยท CISA: Recognize and report phishing โ†— ยท Speko docs: SMS and 10DLC โ†— ยท National Do Not Call Registry โ†—

Everyday chores your assistant can help with3 topics

Once your helpers work, give them the boring jobs: keeping the calendar, sorting email, planning errands, calling to lower bills, and keeping the family on the same page. Start with one chore. Add the next one when the first is running smoothly.

Calendar and inbox: sort it, don't send it

Let a helper look at your calendar and email, tell you what matters, and draft replies. You send them.

๐Ÿ“– Example

Ana's inbox had 214 unread emails. Her helper sorted them: 4 need an answer today, 11 are bills or school, the rest are ads. It drafted 4 short replies. Ana read them, fixed one, and sent them in 10 minutes.

More + copy-ready prompt

Step 1: Connect your calendar and email to your assistant. Each app does it a bit differently: Gemini works with Google apps like Gmail and Calendar (look in Settings for Apps or Connected apps). Claude uses connectors (add-ons that link to apps like Gmail or Google Calendar; Settings, Connectors). ChatGPT has connectors too (Settings). Perplexity has app connectors. Grok Bot, Muse and Hark can connect accounts or use their own secure browser. You sign in yourself; the bot never types your password. Step 2: Ask: "Look at my email from the last 2 days. Tell me the 5 that need me, in 5 lines." Step 3: Ask for draft replies: "Draft short replies to those. Don't send." Step 4: Read and send them yourself, or say yes to each one. Step 5: For the calendar, ask: "What's on my calendar this week? Any clashes?" Step 6: If you want, ask for a filter (a rule your email follows by itself) to move ads into a folder. Gmail's help page below shows how.

๐Ÿค– How AI helps: Connect only the accounts you need, then ask for a 5-line sort and drafts.

โœ… Your first step: Connect just your calendar first. Email can come next week.

Look at my email from the last [2 days] and my calendar for [this week]. Tell me, in 5 lines or less: what needs my answer today, any bills or school notes, and any calendar clashes. Draft short replies for the emails that need me, but don't send anything. Put a link to each email or event. Ignore ads.

Claude help: connectors โ†— ยท Google Calendar: get started โ†— ยท Gmail: create rules to filter email โ†— ยท Gemini โ†— ยท Perplexity app connectors โ†— ยท ChatGPT โ†— ยท xAI: Introducing Grok Bot โ†— ยท Muse (personal AI agent) โ†— ยท Hark โ†—

Calls to lower a bill (and when to stop)

A helper can prepare a script, or with a phone number make the call, to ask about lower prices, promos and retention offers (deals to keep you as a customer).

๐Ÿ“– Example

Sam's internet bill went up. His helper wrote a script: 'I've been a customer 6 years. A new customer gets a lower price. Do you have a promo or a loyalty offer?' The company offered a lower price for 12 months. Sam said yes to that himself.

More + copy-ready prompt

Step 1: Find your bill and see the price and the company's phone number (from the bill or official website). Step 2: Ask your helper to look up current deals for new customers and what competitors charge, with links. Step 3: Ask it to write a short, polite script asking about promos and retention offers. Step 4: Make the call yourself with the script, or, if your bot has a phone number, let it call with your OK. Step 5: Important stop rule: if the company asks for account verification (your account PIN, last 4 of your Social Security number, a security code), the bot must stop and say "My owner will call back." Then you call. Step 6: Never let the bot agree to a new plan. It reports the offer; you decide. Step 7: Ask for the transcript and the offer in writing (receipt).

๐Ÿค– How AI helps: Let the AI do the research and script; you say yes to any new deal.

โœ… Your first step: Pick one bill that went up this year. That's your first call.

Help me lower my [internet, phone, insurance] bill. My company is [company] and I pay about [amount] a month. 1) Find current new-customer deals and 2 competitors' prices, with links. 2) Write me a short, polite phone script asking about promos and loyalty or retention offers. 3) If you make the call, stop as soon as they ask for account verification, PINs or ID numbers, and say I'll call back. Never agree to any change; report the offer and wait for my yes.

Speko โ†— ยท CFPB: Fraud and scams โ†— ยท xAI: Introducing Grok Bot โ†— ยท ChatGPT โ†— ยท Gemini โ†— ยท Claude (Anthropic) โ†— ยท Muse (personal AI agent) โ†— ยท Hark โ†—

Family logistics, errands and reminders

One helper keeps the family plan: who drives where, what's due, what to buy, and reminders for everyone.

๐Ÿ“– Example

Rosa's helper, Juniper, sends a Sunday note: 'This week: Mon dentist 3:30 (Dad drives). Wed field trip form due. Fri grandma visits. Errands: pharmacy pickup, return library books, birthday card for Theo.' The whole family reads it at dinner.

More + copy-ready prompt

Step 1: Make a family helper with one job: "Keep our family plan for the week." Step 2: Each Sunday, tell it what's coming (or let it read your shared calendar, if connected). Step 3: Ask it for one weekly note: who, where, when, who drives, what's due, what to buy. Step 4: Ask it to plan errands in a smart order: "Put my errands in an order that saves driving." Step 5: Ask for reminders: "Remind me Wednesday at 7 am about the field trip form." ChatGPT, Gemini, Grok and Perplexity can set scheduled reminders; your phone's own Reminders or Calendar app works too. Step 6: Kids' info stays simple: first names only, no school addresses or photos in the chat. Step 7: Grown-ups approve anything that gets sent to teachers, coaches or other parents.

๐Ÿค– How AI helps: Give the family helper one weekly job and a Sunday time.

โœ… Your first step: Tell your family helper about next week tonight.

You are our family helper. Every Sunday at [4 pm], send one note for the week: each day's plans, who drives, what's due (forms, payments, gear), and a short errand list in an order that saves driving. Use first names only. Put a link to any calendar event or official page. Remind me about anything due the night before at [7 pm]. Never send anything to anyone but me.

Google Calendar: get started โ†— ยท xAI: Introducing Grok Bot โ†— ยท OpenAI help: Tasks in ChatGPT โ†— ยท Google help: Schedules in Gemini โ†— ยท Grok Automations โ†— ยท Perplexity help: Scheduled Tasks โ†— ยท Muse (personal AI agent) โ†— ยท Hark โ†—

Privacy, safety and turning it off3 topics

A helper bot is useful because it can see things and do things. That's also why you set limits. Give it only what it needs, keep secrets out of the chat, check what it can reach, and know exactly how to switch it off. Kids always build and use helper bots with a grown-up.

Never paste these into any AI

Some things never go into an AI chat: passwords, card numbers, bank account numbers, Social Security numbers, PINs, and security codes.

๐Ÿ“– Example

Theo's bot asked for a code that was texted to his phone 'to finish logging in.' He didn't paste it. He typed the code into the real website himself. Codes are like keys; you keep them.

More + copy-ready prompt

Step 1: Learn the never-paste list: passwords, PINs, security codes from texts, full card numbers, bank account numbers, Social Security numbers, driver's license or passport numbers, and medical record numbers. Step 2: If a website needs a password, sign in yourself on the real website. Step 3: If a bot asks you for any of those, stop. Close the chat. It might be a mistake or a trick. Step 4: Turn on two-step sign-in (MFA, multifactor authentication, meaning a code or tap on your phone as a second lock) for your email and money accounts; CISA's page shows how. Step 5: Use a password manager (an app that remembers strong passwords for you) instead of telling any AI your passwords. Step 6: When paying online, some agents like Muse can use a one-time card number so your real card stays hidden. Still, you approve every payment.

๐Ÿค– How AI helps: Add the never-share list to every helper's instructions.

โœ… Your first step: Turn on two-step sign-in for your email account today.

Rules about private info: Never ask me for passwords, PINs, security codes, full card numbers, bank account numbers, Social Security numbers, ID numbers or medical record numbers. If a website or person needs one, stop and tell me so I can do it myself. Never store any of these in your memory or favorites file.

CISA: Use strong passwords โ†— ยท CISA: Turn on multifactor authentication โ†— ยท CISA: Secure Our World โ†— ยท CFPB: Fraud and scams โ†— ยท Meta: Introducing Muse โ†—

Check what your bot can see and do

Every connection is a door. Open only the doors your helper needs, and look at the list once a month.

๐Ÿ“– Example

Nina's music helper had access to her email, which it didn't need. She turned that connection off. Now it can only see the web and her concert favorites list.

More + copy-ready prompt

Step 1: Open your AI app's Settings. Look for words like Connectors, Connected apps, Apps, Plugins, Integrations or Permissions. Step 2: Read the list. Each item is something the AI can reach (email, calendar, files, a phone number, a browser). Step 3: Turn off anything a helper doesn't need for its one job. Experts call this least privilege (only the access needed, nothing more). Step 4: Be extra careful with any helper that can see your private stuff, read outside stuff like websites or emails, and send messages out. Hidden text on a website can trick that kind of helper (Simon Willison calls this the lethal trifecta). Step 5: In Grok Bot, check each bot's plugins. In ChatGPT, Claude and Perplexity, check Connectors. In Gemini, check connected apps. In Muse and Hark, check their settings and security pages. Step 6: Put a monthly reminder on your calendar: "Check my AI connections."

๐Ÿค– How AI helps: Ask your AI to list everything it can reach, then compare with Settings.

โœ… Your first step: Set a monthly calendar reminder: "Check my AI connections."

List everything you can access for me right now: apps, accounts, files, phone numbers, browsers and memories. For each one, tell me in one line why you need it for your job, or say 'not needed.' Then tell me exactly where in Settings I can turn each one off.

Simon Willison: the lethal trifecta for AI agents โ†— ยท NIST AI Risk Management Framework โ†— ยท Claude help: connectors โ†— ยท Hark security โ†— ยท xAI: Introducing Grok Bot โ†— ยท Perplexity app connectors โ†—

Kids, grown-ups, and the off switch

Kids build helper bots with a grown-up, and everyone in the family knows how to pause or delete a routine.

๐Ÿ“– Example

Ivy, 7, built a helper called Professor Paws with her grandpa. Its only job: one animal fact every morning. Grandpa set it up on his phone, turned off all connections, and showed Ivy how to tell it 'stop.'

More + copy-ready prompt

Step 1 (kids): Always build and use a helper bot with a grown-up. Many AI apps are for ages 13 or 18 and up; check the app's rules, and use the grown-up's phone or account. Step 2 (grown-ups): Use your phone's parental controls (Apple Screen Time or Google Family Link, links below) to choose which apps kids can use. Step 3: Give kid helpers fun, safe jobs only: a fact of the day, a spelling quiz, a bedtime story. No connections to email, calendar, phone or shopping. Step 4: Learn the off switch: open the app's routine list (ChatGPT Tasks, Gemini scheduled actions, Grok Automations or each Grok Bot's settings, Perplexity Scheduled Tasks, Claude Cowork scheduled tasks) and tap Pause or Delete. Or just type to the helper: "Stop all your routines." Step 5: To remove a phone number or plugin, disconnect it in the bot's plugins and cancel it in the service's own account (for example Speko). Step 6: If something goes wrong (a scam, a charge you didn't make), stop the bot, call your bank using the number on your card, and report scams at reportfraud.ftc.gov.

๐Ÿค– How AI helps: Practice turning a routine off once, so you know how before you need it.

โœ… Your first step: Pause one routine and turn it back on, just to practice.

Tell me, step by step in very simple words, how to: 1) pause one of your routines, 2) delete it, 3) stop all your routines at once, 4) disconnect each app or plugin you use, and 5) delete what you remember about me. Use the exact button names I'll see on my screen.

Apple: parental controls on iPhone or iPad โ†— ยท Google Family Link โ†— ยท FTC: Report fraud โ†— ยท CFPB: Fraud and scams โ†— ยท Speko โ†— ยท xAI: Introducing Grok Bot โ†—

โš–๏ธ Your options at a glance10 compared
OptionWhat it isGood forWatch out
Grok Bot (xAI)Always-on helper bots that share a cloud computer, run routines, remember you and can work as a team with a chief of staff bot. Plugins like Speko add a phone number.A team of named helpers with daily routines; calls through ready-made plugins.Paid plans only (SuperGrok, SuperGrok Plus, SuperGrok Heavy, some Cursor plans) with its own usage; in beta. Plain Grok chat is free and different.
Grok (free chat) with AutomationsThe regular Grok chat app. Automations run a question on a schedule.A free first helper and a simple morning note.Doesn't do chores on its own like Grok Bot. Check what's free on the current plans page.
ChatGPT (OpenAI)Chat assistant with Tasks (scheduled reminders and notes), Memory, projects and custom GPTs, connectors, and an agent mode for multi-step jobs.Scheduled notes, drafting, planning, many app connections.Agent features and their names change (OpenAI now points longer jobs to ChatGPT Work). Some features are paid only.
Gemini (Google)Google's assistant with scheduled actions, Gems (saved helpers) and links to Gmail, Calendar, Drive and Maps.People who live in Google apps; calendar and email sorting.Some features need a paid Google plan; check what it can see in your Google account.
Claude (Anthropic)Chat assistant with Projects, memory, connectors to apps, Cowork for doing tasks on your computer, and Claude in Chrome for clicking around websites.Careful writing, sorting documents, desktop chores with approvals.Cowork and Claude in Chrome are on paid plans; keep it away from banking and health sites when browsing.
PerplexityAnswer engine with sources, Spaces, Scheduled Tasks, app connectors and the Comet browser assistant.Morning notes with links to sources; research routines.Some features are paid; the browser assistant can click on websites, so keep approvals on.
Muse (Meta)Personal AI agent in the Muse app and inside WhatsApp. Works on its own secure computer, books, shops and fills forms, and comes back for your OK.People who already text in WhatsApp; errands that run while you're away.New (launched September 2026); check what it can access and keep approvals on for every purchase.
Hark Pro (Hark)Personal assistant on web, iPhone and Android with a secure browser (Handoff), memory, one-tap Action Buttons and custom dashboards.A free start, simple one-tap actions, beginners.New (October 2026); read its security page and check which features cost money.
Your phone's built-in assistant (Siri, Google Assistant)Built into your phone. Sets reminders, timers, alarms and calendar events by voice.Quick reminders and alarms with no new app.Not built for daily routines with receipts; pair it with one of the assistants above.
Phone and text add-ons (like Speko)Services that give a helper bot its own phone number for calls and, after 10DLC sign-up, texts.Bookings, appointments and price quotes by phone.You set up and pay for the account yourself; US texting needs carrier registration; never let it share private numbers.
๐Ÿšฉ Red flags10
  • A bot (or anyone) asks you for a password, PIN, security code, card number, bank account number or Social Security number.
  • A helper says "done" but can't show a receipt: no link, quote, screenshot, confirmation number or transcript.
  • A helper bought, booked, paid, sent or posted something without your yes for that exact thing.
  • The total price changed after you said yes, and the helper didn't stop to ask again.
  • A helper has access to apps it doesn't need for its one job (like a music bot that can read your email).
  • Your morning notes are getting long, or line 1 isn't really what changed since yesterday.
  • A bot on the phone gave out account or insurance details, left voicemails you didn't ask for, or kept calling.
  • Texts going to people who didn't ask for them, or group texts you didn't approve.
  • A website, email or text tells your AI to do something new (hidden instructions are a known trick).
  • A child is using a helper bot alone, or a kid's helper is connected to email, shopping or a phone number.
๐Ÿ™‹ Questions to ask a pro8
  • (Your bank or card company) Can I get a virtual or one-time card number, and can I set alerts for every charge?
  • (Your phone carrier) Can I block unknown numbers or spam texts, and does my plan have call-filter tools?
  • (The AI company's help center) What exactly can this assistant see and do in my account, and how do I delete what it remembers?
  • (Speko or your phone add-on's support) How do I turn on texting, what does 10DLC need from me, and how do I cancel the number?
  • (A tech-savvy friend or library tech help desk) Can you sit with me while I connect my calendar and check my settings?
  • (Your kids' school or pediatrician) Which apps and ages do you recommend for kids using AI?
  • (The business your bot calls) Is it OK if my AI assistant calls to book, and what info do you need from me directly?
  • (An accountant or financial pro) Before I act on a bill or money idea my AI found, is this right for me?
๐Ÿ“– Plain-English glossary20 words
Personal assistant (AI)
An AI helper that keeps track of your stuff, reminds you, plans and drafts things, and asks before doing anything big.
Chief of staff bot
The boss helper bot. You talk to it; it hands jobs to the right small helpers and gives you one short report.
Helper bot
A small AI helper with a name and one job, like meals, events or bills.
Agent
An AI that can take steps on its own, like browsing websites or filling forms, until the job is done or it needs your OK.
Routine
A job your helper does by itself at the same time every day or week, like a morning note.
Scheduled task
Another name for a routine. ChatGPT calls them Tasks, Perplexity calls them Scheduled Tasks, Gemini calls them scheduled actions, Grok calls them Automations.
Morning note
A short daily message from your helper. Line 1 is always what changed since yesterday.
Receipt
Proof that something is true or done: a link, a short quote, a screenshot, a confirmation number or a call transcript.
Read-back
When your helper repeats the plan to you (what, where, when, how much, who) before asking for your yes.
Approval (the yes rule)
Your OK for one exact action. Nothing gets bought, booked, paid, sent or posted without it.
Plugin
An add-on you switch on to give your bot a new power, like a phone number. Speko is a plugin for Grok Bot.
Connector
A link between your AI and another app, like your email or calendar. You sign in yourself.
MCP (Model Context Protocol)
A techie, standard way to plug tools into an AI by hand. Ready-made plugins let you skip it.
Transcript
The written-down words of a phone call, so you can read exactly what was said.
10DLC
Short for 10-digit long code. A US phone company sign-up that apps and bots need before texting from a normal local number.
SMS
A plain text message on a phone.
Area code
The first 3 numbers of a US phone number. Picking your local one helps local shops answer.
Favorites file (memory)
A running list your helper keeps of what you like and don't like, so you don't repeat yourself.
Least privilege
Giving a helper only the access it needs for its job, nothing more.
MFA (two-step sign-in)
A second lock on your account, like a code on your phone, on top of your password.
๐Ÿ”— Sources48
๐ŸŽ™๏ธ People I learn from

Listed because I've learned from them โ€” not endorsements, not sponsors, and they don't know FixKit AI exists (yet).