M

start ten businesses on autopilot with grok bot (full guide)

20 min readView source ↗

Cover image

grok bot has its own computer with a browser, a terminal and a filesystem, which means it can run a business fully autonomously

here are ten businesses you can start with it right now, you don't need to have an audience or a product

by the end of this guide you'll know:

  • how to pick a niche where people actually pay
  • how to run ten channels without repeating yourself
  • how to build the product the audience asks for
  • how to get every video made with an ai actor
  • how to sell to businesses already spending on ads
  • how to charge a retainer for something the bot does

Article image

every one of these needs images or video at some point, and that part runs on arcads across all ten.

a couple rules everything follows

the first one decides whether you make money or just make content.

the buyer has to be someone who wants a problem gone, and not someone who is just curious. for example, a story channel with adsense pays a few dollars per thousand views.

the same channel pointed at people who are stuck with something they deal with sells them the fix, which is completely different.

every channel in this guide sells a product. the merch systems which we will get into are the exception, those buyers are keen rather than stuck, and that's fine because the margin is in the product.

the second rule is that the bot never repeats itself. every system keeps a log in an obsidian vault of what it has covered, built or listed, and reads it before doing the next one.

the setup everything shares

grok bot has a persistent cloud computer, shared across every bot on your account with a screen for each one.

that's why it can use tools that have no api by clicking through them like a person.

set this up once and all ten systems use it.

use whop for collecting money. the cli runs from the bot's terminal, so it creates the products, sets the prices and generates the checkout links by itself

printful for anything printed. the bot generates artwork to their spec, lists the products, and orders route through when someone pays.

arcads for instances of image and video. i use it because everything is in one place and because the bot can reach most of it through the arcads mcp at mcp.arcads.ai and the video and lip-sync apis. there are over a thousand ai actors plus custom avatars, several video and image models.

a scheduling tool like post bridge for getting posts out. it publishes through each platform's official api rather than automating a browser, covers most of the places you'd post, and takes an api key so the bot can queue everything itself.

obsidian synced to the bot's computer, one folder per system, as the log and the source of truth.

paste this to get it all running:

> set up everything you'll need to run the systems i approve later.

> - install the whop cli and sign in from your terminal. run whop --llms and tell me in your own words what you can and can't do with this login

> - connect the printful api. read their docs and tell me the exact artwork spec, and generate everything to it from now on

> - connect the arcads mcp at mcp.arcads.ai and tell me which models and templates you can reach through it and which need the browser

> - here's the scheduling api key: [key]. list which of my accounts are connected and which platforms are missing

> - set up the obsidian vault with one folder per system, and a log file in each that you read before doing anything

> - set the spend cap at each provider's end, not in this chat

> - before any of this runs for real, do one test of each: one product, one print order to my own address, one video, one post to a private account

> - tell me now what you can't do with the access i've given you, rather than working around it later

Article image

1. answer channels in ten niches

people usually search youtube for specific questions. how do i get my deposit back, what do i do if my dog eats chocolate etc.,

those searches never stop and the videos answering them keep earning views for years. these are called search based videos.

the bot reads reddit, quora and facebook groups across ten niches every day for the questions getting the most engagement, writes the script, generates the presenter video in arcads with an ai actor, makes the thumbnail, and queues it to youtube with the short form version going to tiktok and instagram.

picking the ten niches.

> find me ten niches for question-answering channels.

> - look for topics where people search a specific question rather than browse, and where the person asking has a problem costing them money or time

> - for each niche give me five real questions people are asking right now, with links

> - tell me what the person asking would pay to have solved, and roughly what that's worth to them

> - skip anything where the audience is curious rather than stuck

> - skip anything medical, legal or financial where wrong information could hurt someone, unless i tell you i know that field

> - tell me which of these nobody would pay for and drop them

the daily run.

> every morning, for each niche in the vault, find the question worth answering today.

> - read the subreddits, quora topics and facebook groups in the config file for the last 24 hours

> - rank questions by how many people are asking a version of the same thing, not by upvotes on one post

> - check the log before you pick anything. if we've covered it or anything close to it, skip it and pick the next

> - write a script that answers it properly in under three minutes, in plain words, no intro and no sign-off

> - generate the presenter video in arcads using the actor already assigned to that niche, and the thumbnail with an image model

> - queue it for youtube, with a 45 second cut for tiktok and instagram

> - log the question, the date and the link

> - if nothing today is worth a video in a niche, post nothing there and tell me why

the money.

after a couple weeks the bot has a log of which questions keep coming back in each niche, and that log is the product outline.

> look at the log for [niche] and tell me what to sell.

> - group the questions from the last month into the problems underneath them

> - tell me which group comes up most, and how many separate people asked a version of it

> - write the outline for a guide that answers that group completely

> - create the product on whop with the cli, set the price, and generate the checkout link

> - make the sales video in arcads with the same actor as the channel

> - put the link in every description and pinned comment for videos in that group

> - if the questions in this niche are too scattered to make one product, say so rather than making a broad one

a broad product in a scattered niche is the most common way this fails. it works because each video answers one question for free and the product answers that whole group properly.

Article image

2. story channels for people who are stuck

reddit is full of people who are stuck. deposit withheld, wages unpaid, contractor took the money and disappeared, bank froze the account, dealer added fees after signing.

those threads get thousands of replies every time because everyone reading has either been there or is scared of it.

this format already does enormous numbers on youtube and tiktok as a robot voice over game footage.

> read these subreddits daily and build me one story video: [list].

> - pick a situation where the poster is stuck on something specific, not venting generally

> - rewrite it completely so nothing is copied word for word and nothing identifies the person

> - the script tells what happened and then what actually fixed it, with the fix taking half the runtime

> - generate it in arcads with the channel's actor and queue the short cut

> - log the situation type so we cover different ones

> - never contact anyone whose post you used, and never name a company that hasn't been named publicly in the thread

nobody is watching because someone else's deposit got withheld.

they're watching because theirs is, or might be, and they're waiting for what to do about it.

the money is a pack per situation, $29 to $49, with the demand letter, the small claims steps and the complaint template in it.

the same handful of situations keep coming back, and each pack sits under every story about that one.

an affiliate link that matches the story works as the easier version if you don't want to build the pack. legal and finance programs pay well and there's nothing to write.

only do this in an area you actually understand, because the people watching are in trouble and a wrong answer costs them more trouble.

Article image

3. ad remakes for businesses already spending

the only service in the list that's selling to businesses. these are people already paying for ads, so there's no convincing them the category works.

the bot scrapes who is running ads in a niche from the meta ad library and tiktok creative center. then mark, the ad agent inside arcads, reads the competitor ads that are working and recreates a version for your prospect's brand.

mark has no api yet, so the bot drives app.arcads.ai through the browser on its own computer.

basically mark finds good ads from competitor brands and rebuilds them for yours. it does not find bad ads.

> build me the prospect list and the pitch asset.

> - pull businesses running ads in [niche] from the meta ad library and tiktok creative center, with how long each has been running

> - prefer the ones running the same creative for months, they have budget and nothing new to say

> - for each one, run their site through arcads agent mode to build brand memory, then have mark break down what's working in the competitor ads

> - generate one finished vertical ad for their brand, and three alternate hooks across three aspect ratios

> - prepare everything and stop. i send the pitch myself

> - tell me which of these prospects looks too small to afford a retainer, rather than letting me find out on the call

you charge a retainer, and the number depends on your market.

Article image

4. the postcard retainer

this is the lead gen system from my christmas lights article and it's the one with a sense of urgency on it, because installers book their season in september and october.

the bot builds the installer list from google maps, finds the streets that already decorate, renders each house with lights on it, and mails a card to every address.

the card front is that house. the back has one line and a qr to a short clip of the installer's offer.

> build the installer list and the campaign.

> - pull christmas light installers from google maps in [region], keep the ones with 20 to 200 reviews, drop franchises

> - flag who offers permanent lighting, that changes what we sell them

> - find the streets that already decorate from crowdsourced light maps, local news roundups and local groups, and take the ones appearing in more than one

> - pull street view frames and satellite imagery for every house on those streets

> - render each house with lights on the roofline using an arcads image model as an edit, so the house stays exactly as it is

> - generate a short ugc clip of the installer's offer for the qr to point at

> - leave the qr corner blank and tell me, i composite a real one in

> - prepare the batch and stop. i approve before anything prints

on pricing, charge a monthly retainer for exclusivity in a set of postcodes rather than a percentage of closed jobs, because you can't really see what closed.

the number comes from what installers already pay lead platforms, and contractors have written about paying $188 for eight leads and getting no jobs.

Article image

5. the vet bill channel

someone posts that their dog needs a $4,000 surgery and they can't pay it, and that thread fills up every time. everyone with a pet reads it and thinks about their own.

> read the pet subreddits for posts about a bill someone can't afford.

> - pick ones where the outcome is known, so the video can say what happened

> - the script covers the situation, the options that existed, and what it cost in the end

> - generate it in arcads and queue it

> - never include the person's username, location or the clinic's name

> - if the outcome was that the animal was put down for money reasons, do not make that video

this works better than most affiliate channels because of the timing gap. someone watching a video about a $4,000 emergency does not have that emergency right now.

they have a healthy pet with a clear picture of what happens if that changes, which is the why anyone ever buys pet insurance.

it also means the back catalogue keeps working. a video from a few months ago is still finding people who just got a puppy, and the link under it is still the same link.

the money is pet insurance affiliate and the vet payment plan providers, both of which pay per signup.

6. fandom merch

fans have their own language. inside jokes, phrases they use with each other, references that mean people outside of the fanbase don't really understand.

> watch these fandom subreddits and tell me what to print: [list].

> - collect the phrases, jokes and references that fans use with each other, with how often each one appears

> - skip anything that is a lyric, a title, an artist or character name, or an official slogan

> - design around the phrase itself, text-led, nothing that copies an official design

> - list on printful with a mockup, and cut three short clips in arcads for the store

> - log every design so you never repeat one

> - tell me which of these you think crosses into someone's intellectual property, even if i asked for it

song lyrics are copyrighted, artist teams send takedowns, printful's own content policy rejects it. anything with a real person's name or likeness on it is a problem.

what fans say to each other makes better merch than a lyric anyway, because it's something only they understand.

so you can have grokbot download and repurpose the most viral clips in that niche and plug your store in the comments.

Article image

7. trend product stores

if something starts climbing on tiktok creative center and in the product subreddits weeks before it's everywhere, that gap is the opportunity.

> watch tiktok creative center and these subreddits for products climbing: [list].

> - only flag things where the trend is rising, not peaked, and show me the traction with dates attached

> - for anything i approve, set up the store on whop and generate the ugc in arcads using the short form ad generator and ugc product showcase templates

> - give me three hooks across three aspect ratios from one script

> - queue nothing until i approve the batch

> - set the kill metric before any traffic goes near it and tell me what number means this is dead

what matters is how fast that number is growing rather than how big it is, which is why the prompt asks for dates attached to the traction.

on the orders, tell it to use the address on the payment whenever anything disagrees, the person paying is often not the person receiving, and guessing wrong on a physical order costs you the product and the postage.

Article image

8. hobby print on demand

the same idea as fandom merch, pointed at hobbies where people build their identity around it. fishing, pickleball, homebrewing, woodworking. far less takedown risk than a fandom, and the buyers are older with more money.

> read these hobby subreddits and forums for what to print: [list].

> - collect the phrases, in-jokes and complaints that only people in this hobby would understand

> - design text-led around those, nothing using a brand, a tournament or a manufacturer's name

> - list on printful and cut three clips in arcads for the store

> - log everything so nothing repeats

> - tell me which of these would only make sense to someone already deep in the hobby, because those are the ones worth printing

a shirt anyone would understand doesn't sell, because it doesn't tell other people in the hobby that you're one of them.

what makes hobbies better than fandoms is the buyer. someone multiple years into fishing has spent real money on it and thinks of it as part of who they are.

they buy for themselves and for the other people in their club, which is where the volume comes from.

the complaints are the best source. every hobby has a thing everyone complains about, the gear that always breaks, the rule everyone argues over.

those are already phrased for a shirt because they've been repeated a thousand times.

nobody buys a shirt that makes them feel bad. so ask it which read as funny and which read as bitter, and skip anything aimed at a specific brand.

9. etsy digital products

> read etsy's trending searches and the review threads in those categories.

> - find what buyers say is missing or wrong with what's already listed

> - generate the designs through arcads image models to etsy's size and format spec

> - write the listing in the words buyers used in those reviews

> - mark every listing as made with ai tools where etsy requires it

> - log every design and never list a variation of one we've already done

> - tell me which of these categories is already saturated rather than listing into it anyway

the reviews are the reason to bother with this one rather than listing whatever the bot invents.

every saturated category has a bunch of two star reviews saying the same things, the file wouldn't print at the size promised, the colours came out different, there was no version for the paper size they use.

so basically, that's the product brief, written by people who already paid for the bad version.

start with ten listings and look at the data before adding more.

10. clip channels from public proceedings

council meetings, court proceedings, government hearings and other public records. these are usable in a way that conference talks and podcasts are not.

that distinction is worth getting right. a conference talk is owned by the speaker or the organiser and most conferences license the recordings tightly. an earnings call recording is the company's, though the facts said on it are free to report. government proceedings in most places are public record. taking the format of a clip channel is fine, clipping copyrighted shows is not, and the channels doing that get struck eventually.

> pull the transcripts from these public meeting sources and find what's worth clipping: [list].

> - find the moments where something happens that a normal person would care about, not procedural business

> - confirm the source is public record and tell me which rule makes it so before you use anything

> - cut the clip, add captions, and generate a short presenter intro in arcads that sets up what we're about to see

> - queue it

> - if you can't confirm the licensing on a source, skip it and tell me rather than using it anyway

council meetings work because almost nobody watches them and things happen in them constantly. a planning decision that changes what someone can build on their street, a budget line that raises a fee, an argument between two officials that goes somewhere. all of it is on record, all of it is legally usable, and none of it is being clipped.

the audience also self-selects. someone watching clips of planning meetings is a homeowner, a developer or a local business, and all three are worth something to anyone selling in that space.

the money is an affiliate or a whop product matched to whoever watches, not adsense. a channel clipping planning decisions sells a guide to getting permission through.

what it's allowed to do without asking

set this before you turn anything on.

> these are your permissions and they don't change without me saying so.

> - you can do these on your own: read sources, write scripts, generate images and video, draft listings, build lists, log everything

> - you prepare and stop for anything that publishes, prints, mails, contacts a real person or spends money

> - never refund, change a price or move money on whop

> - never reply to a comment or a dm, i do that

> - run a maximum of three systems live at once and queue the rest

> - if you hit a password, a captcha or a payment screen, hand it back to me rather than trying to get past it

> - when i approve something, that approves the thing i approved and nothing after it

the schedule

> from now on run this on a schedule.

> - morning: each live system collects and prepares, and queues what's ready for my review

> - midday: publish whatever i approved

> - night: update every log, and tell me what each system produced and what it cost

> - weekly: which systems are producing nothing and whether to drop them

> - monthly: revenue per system from whop, spend per system, and what to kill

> - if nothing moved, say that in one line rather than filling the report

where to start

pick two niches rather than multiple. get the daily run producing something you'd actually publish, watch the outputs yourself before any go out, and fix the script prompt with what you learn.

thanks to arcads for sponsoring the image and video sections. it's what i use for all of that work, and you can try it at https://www.arcads.ai/?comet_custom=everestchris6

Related articles