How to Run X and LinkedIn With Grok Bot (Full Guide)
文章目录
We're at the point where a founder's X and LinkedIn can be run almost end to end with the right setup. Here's how to set Grok Bot up to build your voice, find what's working in your niche every morning, write the posts, hand you the engagement list and report what all of it turned into.
By the end of this guide you'll have:
-
A voice database built from how you actually talk
-
A daily scan of your niche with the winners flagged
-
A post queue every morning, in your voice, in an X version and a LinkedIn version
-
Visuals made for every post, ready to attach
-
A profile that turns reach into DMs and calls
-
A daily list of who to reply to and which buyer posts to comment on
-
Grok Bot running all of it and improving every week
How it basically works
The idea is simple. Grok Bot has its own computer, so it can read your X account and any public account, open your files and run routines on a schedule. You give it a setup prompt with everything about you and your buyer, and it sets itself up as your head of content.
It splits the work into separate jobs, like 1 job owning your voice, another doing research, another drafting, another building the engagement list, and so on. You can think of each job as a small bot with 1 task. This setup is for 1 account, and you copy it for each exec once the first 1 is working.
The part that makes this safe is how it connects. You connect X through the official sign-in, which gives Grok Bot read access to your account and the ability to hand you draft cards. Nothing goes out until you press post. That's the whole setup for X. For LinkedIn there's no official connection, so the bot works from your analytics export and anything you paste in, and it never logs in.
That's the rule for everything in this guide. The bot never posts, comments, DMs or sends connection requests on either platform. The platforms see an agent clicking around your account as an automation tool and they do not care why you did it, and your account getting restricted the week before a launch is a bad day.
Everything lives in 1 place. Either Grok Bot builds you a dashboard with everything in it, or you connect it to Notion and it runs everything from there.
The whole setup is built around whatever your conversion is, let's say a booked call or a trial, not impressions for their own sake. I think impressions are cool, but I care way more about what they turned into. 1 client account we run did 2.8M impressions in 100 days, and what they turned into was 30+ calls and $500K in pipeline.
What you need
-
Grok Bot with your X account connected through the official sign-in, which gives it your post history and public numbers on any account it looks at.
-
A folder of you talking, uploaded to the bot or shared as a Google Drive link. Fathom or Gong transcripts from sales calls, a podcast or 2, a 1 hour recording of you explaining your space to a friend, and your last 20 posts that performed. Slack exports are underrated here, since they're the least filtered version of how you write.
-
Your LinkedIn analytics export, dropped into that same folder every Friday, because LinkedIn doesn't show impressions publicly and the bot can't get them any other way. If you want it reading LinkedIn posts and buyer profiles too, connect a data provider like Crustdata or Apollo. Without 1, the LinkedIn side runs on what you paste in.
-
Notion or Airtable for the dashboard, or let the bot build 1.
-
A CRM the bot can connect to, or a Google Sheet it can read, where every call and every DM conversation on either platform gets logged with the post that brought it in. The bot never reads your inbox, so this log is how it knows what a post turned into. HubSpot, Attio or a sheet to start.
The setup prompt
This is the prompt that starts everything. It tells the bot who you are and who you sell to, sets up the dashboard and the jobs, and runs everything once as a test so you can see the output.
Fill in the brackets with as much as you know. The more it knows about you and your buyer, the better everything downstream gets.
You are my head of content for X and LinkedIn. Set up and run my accounts as a set of separate jobs, each with 1 task, that work together in 1 place. The goal is [a booked call, a trial, a demo, or whatever a conversion is for me], not impressions for their own sake.
About me:
- Who I am and what I do: [1 line]
- What I sell and what it costs: [offer and price]
- Who buys it: [title, company stage, industry, as specifically as you can]
- The 1 problem I want my name attached to: [the problem, in the buyer's words]
- The 2 to 3 topics I post about: [narrow enough that it sounds limiting]
- The things I'll say publicly that competitors won't: [opinions, or say "pull them from my transcripts"]
- My accounts: [X handle] and [LinkedIn URL]
- The 10 to 20 accounts my buyers already read: [handles, or say "find them"]
- Buyer handles on X: [a short list, or say "find them from who replies to my top posts and matches my buyer"]
- What I can't claim or talk about: [anything off limits]
Files (read them on your first run and save what you learn to memory):
- My transcripts and past posts, [uploaded here or at this Drive link]
- My LinkedIn analytics export, in the same folder. I'll drop a fresh 1 every Friday.
Tools:
- X through my connected account, for post history, public metrics and research. Read only.
- [Crustdata or Apollo, if connected] for LinkedIn posts and buyer profiles. If not connected, work from what I paste in and say when you're guessing.
- [Build me a dashboard, or use my Notion] as the 1 place everything lives, and this chat for approvals
- My CRM at [URL, or a Google Sheet you can read] for calls, DMs and pipeline. You never read my inbox on either platform, this log is your only source for DMs.
Step 1. Set up the dashboard with these sections: voice, swipe file, queue, engagement, tracking and rules, with a lessons page inside rules.
Step 2. Create these jobs, each as its own routine. Every job reads the voice page and the rules page before it does anything.
- Voice: build my voice database from the transcripts and past posts. Direct quotes, filler words, how I open and close a thought, and a kill list of phrases I'd never say. Update it every time I correct a draft.
- Research (daily, 7am Central): scan the accounts my buyers read and X search for my topics. Flag only posts that did 3x that account's usual numbers. Log hook, structure, format and why it worked to the swipe file. Note what's moving on X that I could bring to LinkedIn, and check it against LinkedIn only if a connector is on.
- Planner (Sunday night): turn the best swipe file entries and ideas into next week's queue, 5 LinkedIn posts for the week and 2 to 3 X posts a day. Show me here, and once I approve, save the approved queue to the queue page.
- Writer (daily, 7:30am Central): draft today's posts from the queue page, not from chat. An X version aimed at bookmarks and a LinkedIn version aimed at comments, with the visual made as a file for each, built from code if it has text or numbers on it. Score every draft against the rules page and rewrite anything under 7 before I see it.
- Engagement (daily, 8am Central): give me 5 X posts to reply to, 2 to 3 to quote post with my take, and 20 buyer posts to comment on. Buyers means the buyer handles above plus anyone replying to my top posts who matches my buyer, and you keep that list growing. X first, LinkedIn only from a connector or the buyer profiles I've pasted in. Every reply drafted. I post them myself.
- Profile (monthly): audit my X bio and pinned post as a buyer arriving from a post. For LinkedIn, ask me to paste in my headline, about and featured section or send a screenshot, since you can't open it. List every leak with the fix drafted.
- Report (Friday): read the fresh LinkedIn export I dropped in, then report what went out, what it earned by platform, which posts drove profile visits, and the DMs and calls logged in the CRM, then 3 to 5 plain lessons written to the rules page that every job follows.
If my situation needs a job that isn't here, or doesn't need 1 that is, say so and adjust.
Step 3. Rules
- You never post, comment, reply, DM, send connection requests or reach out to anyone on any platform. You write the text, I send it myself.
- You never invent a number, a client or a result. If you don't know, ask me here.
- Every draft runs through the kill list before I see it.
- Every morning send me 1 short message here: today's drafts, the engagement list, anything that broke out in my niche overnight, and the 1 thing you'd do next.
- When you need a decision from me, suggest the best option with 1 line on why, and ask me to confirm.
Step 4. Run every job once as a test, without publishing anything, and show me what each 1 made.
Start with your voice
The first thing the bot does is build your voice database, because every other job reads from it. If this page is weak you've just built a faster way to produce slop, and buyers can tell within a line.
Most people's version of training AI on their voice is pasting their website copy into a project. That's a company description. Your voice is your filler words, how you open a thought, how you trail off, the phrases you'd never say. The bot pulls that from recordings of you actually talking, quotes you directly and doesn't clean you up.
The kill list is the other half. Every AI tell you've ever cringed at goes on it. The 2 line hook into 3 bullets into a 1 liner, "it's not X, it's Y," "here's the thing," the word "genuinely." Every draft gets checked against it before you see it, and every correction you make gets added, so the bot on day 60 is sharper than day 1.
The bar for done is simple. You'd post a draft after changing 10 to 15 words.
Build my voice database.
- Read every transcript and past post I uploaded [here or at this Drive link]. The transcripts are me talking, the posts are what performed.
- Pull my filler words with direct quotes of me using them, how I open a thought, how I end 1, my rhythm when I'm explaining vs when I'm being direct, the words I reuse, and my opinions and stories with the quote attached
- Quote me directly. Never clean me up and never describe my "style."
- Build a kill list from the posts I've marked as flops or AI written, plus the standard tells: 2 line hook into 3 bullets into a 1 liner, "it's not X, it's Y," "here's the thing," "genuinely," em dashes, "stop X, start Y," motivational closers
- Write it all to the voice page. Then write 1 post with it and show me which lines came from which pattern
- From now on, every time I edit a draft, add what changed to the voice page or the kill list
Research what's already working
Every morning the bot checks the accounts your buyers actually read and pulls the posts that did way better than that account normally does, and turns them into a swipe file.
That "normally does" part matters. A 2,000 like post from someone who always gets 2,000 tells you nothing. A post that did 3x the account's usual numbers tells you the idea or the format broke out, and that's the only kind worth logging.
For each 1 it saves the hook, the structure, the format and why it worked, tagged by type (trend, playbook, proof, contrarian, story), so the writer job can grab a proven structure whenever it needs 1.
It also runs the relay. X is where things get argued about first, and LinkedIn is 2 to 3 weeks behind on almost everything, so a post that's a week old on X lands as brand new on LinkedIn. [My best LinkedIn post this year started as somebody's X post from almost a year earlier.] The research job flags what's moving on X today, and that's tomorrow's LinkedIn post with your opinion attached.
1 thing to know. The bot can read all of X, but it can't read LinkedIn without a connector like Crustdata, so "this hasn't hit LinkedIn yet" is its guess unless 1 is on. The relay still works without it, because the lag is reliable enough that you can just bring the winners over.
1 tip tho, have it sort by bookmarks and replies, not likes. Likes cost nothing so they're worth nothing.
Run research for my account.
- Every morning, pull the last 24 hours from [the accounts my buyers read] and X search for [my topics]
- Flag only posts that did 3x that account's median likes or bookmarks over the last 30 days, and ignore everything else
- For each flagged post, log to the swipe file: link, first line, hook type (number, list, claim, contrarian, story, question), structure, format, length, bookmarks, replies, and 1 line on why it worked
- Flag anything being argued about on X worth bringing to LinkedIn, with the 2 sides and which side is underrepresented. If a LinkedIn connector is on, check whether it's already there. If not, say it's unchecked.
- Once a week, re-rank the swipe file by what's still working and retire formats that have stopped
- Every entry needs a link
Plan the week and write the posts
Sunday night, the bot turns the best swipe file entries and ideas into next week's queue and asks you to approve it. A week is 5 LinkedIn posts, 7 max, and 2 to 3 a day on X. Posting 15 times a day on X splits your reach instead of multiplying it.
Once you approve, it saves the queue to the dashboard so the morning writer reads from there and not from a chat thread. Then it drafts each day's posts in your voice, as 2 separate pieces. On X it writes for bookmarks, so the post has to be something people come back to and use. On LinkedIn it writes for comments, so it's more of a story or an opinion and ends on something people want to argue with or add to. Same idea, 2 packagings, never the same text.
A couple of things from my own numbers that go in the rules page. Posts that open with a list or a how-to framework pulled several times the bookmarks of anything else I posted this year, and number hooks were right behind them. Question hooks and plain claims underperformed. The first line that works tells people exactly what they're getting.
Articles are still the most underrated format on X. My 2 biggest posts of the year were both long articles, and the quote post I put on top of the first 1 did almost as much as the article itself.
And lead magnet posts, "comment X and I'll send it," are still the highest engagement format on either platform. My best 1 pulled over 1,800 replies. They only work when the thing you're giving away is tied to what you sell.
The scoring at the bottom of this prompt is how the bot checks a draft before you see it. A draft that would only earn likes gets rewritten until it earns a save or a long comment.
Plan and write next week.
- Sunday night: turn the best swipe file entries and ideas into the queue. 5 LinkedIn posts for the week (Mon trend plus opinion, Tue playbook with a visual, Wed proof, Thu contrarian take, Fri behind the scenes) and 2 to 3 X posts a day. Show me here, wait for my approval, then save the approved queue to the queue page.
- Every morning: draft today's posts from the queue page. The X version first, aimed at bookmarks, then the LinkedIn version, aimed at comments. Same idea, different packaging, never the same text.
- First line rule: say what the reader gets. Prefer a number or a list structure. No questions, no teasers.
- Give every post 1 job: pipeline, authority, reach or a launch, and say which.
- Score each draft 1 to 10: save 10, DM send 8, comment of 10+ words 6, follow 5, read past the first 2 lines 5, repost with a take 4, like 1, short comment 1, fast scroll past -2, link in the body -5, engagement bait -10. Rewrite anything under 7 before I see it.
- Run everything through the voice page and the kill list. Never invent a number, a client or a result.
- When I send you a voice note, pull out every separate opinion and tell me how many posts are in there before you write any
Make the visuals
A visual takes up more of the feed, so people have to make a decision to scroll past you instead of drifting by. Most posts should have 1.
Before it makes anything, the bot writes a style note from the best visuals in the swipe file, so everything you put out looks like it came from 1 account. Then every post gets its visual: a chart built from your real numbers, a framework image, or a carousel PDF for LinkedIn. Carousels are still the highest dwell time format there, so 1 a week minimum.
Grok Bot makes the visual and the copy and drops the file in chat for you to download. The X draft card only takes text, so you attach the image yourself when you post, and LinkedIn is the same since you post there by hand anyway.
1 caveat. AI images are bad at small text, so anything with numbers or words on it, results charts, framework slides, carousels, should be built from code, not generated. Keep the AI image generator for simple graphics with no text.
Make the visuals for the approved posts.
- First write a style note from the best visuals in the swipe file: the look, the colors, the type, what's always shown and what's never shown
- For each approved post, make the visual and drop it in chat as a file: a chart from my real numbers, a framework image, or a carousel
- Anything with text or numbers on it gets built from code. Use image generation only for simple graphics with no text
- For carousels, write the full slide copy, a hook slide, 7 to 9 slides with 2 to 3 lines each, and a last slide with the next step, then build it as a PDF
- Never make a fake screenshot, result or customer
Fix the profile they land on
Most people click the profile before they do anything else, so the profile is what converts, not the CTA at the bottom of the post. If it's set up like a resume, every post you write is leaking.
So once a month the bot audits both profiles the way a buyer would, arriving from a post. It can open your X profile on its own. It can't open LinkedIn, since it never logs in and LinkedIn tends to throw a signup wall at anyone who isn't, so you paste in your headline, about and featured section or send it a screenshot.
On LinkedIn it checks:
-
Whether the headline says who you help and what they get
-
Whether the banner shows the result and proof
-
Whether the featured section gives 3 types of visitor somewhere to go (a link for buyers ready now, your best post for the curious, a lead magnet for everyone else)
-
Whether the about section says the outcome instead of your life story
On X it checks whether the bio states the outcome and the pinned post proves it.
It lists every leak with a fix and drafts the fixed copy, and writes what it learned to the rules page so the writer job stops sending people to a dead end.
Audit my profiles.
- Open my X profile as [my buyer] arriving from 1 of my posts. For LinkedIn, ask me to paste in my headline, about and featured section, or send a screenshot, and audit that.
- On each, check: does the headline or bio say who I help and what they get in under 5 seconds, is there proof near the top, is there a next step visible without scrolling, and does every type of visitor (ready now, curious, not ready) have somewhere to go
- List every leak with the fix, and draft the fixed copy: LinkedIn headline, banner text, about section, featured slots and link text; X bio and pinned post
- Write what you learned to the rules page as a lesson for every other job
- Publish nothing
Engage without getting banned
A post is decided in its first 30 to 60 minutes. The platform shows it to a small sample, watches what they do, and decides whether anyone else sees it. If you post and walk away to a meeting, you capped it.
The bot can't reply for you and it shouldn't, because that's exactly what gets accounts restricted. What it does is hand you a queue. Every morning it gives you 5 posts in your niche to reply to, 2 to 3 to quote post with your own take, and 20 posts from your buyers to comment on, with every reply drafted and ready to edit. You post them from your own account.
The jobs run on a schedule, so the bot doesn't know when your post goes up. Tell it, and it pulls the comments and drafts your replies, then checks again 30 minutes later. If you post at the same time every day, schedule that check an hour after.
This list is X first, because that's what the bot can read. It needs a starting point for who counts as a buyer, so give it the 10 to 20 accounts from the setup prompt plus a short list of buyer handles, or tell it to build 1 from who replies to your top posts and matches your ICP. It keeps that list growing as it goes. For LinkedIn it needs a connector or a list of buyer profiles you've pasted in, and then it drafts comments on their recent posts the same way.
- things about the comments. They have to be yours. AI comments are the lowest form of engagement and everyone can tell, so the draft is a starting point you rewrite in 2 sentences. And the ones that matter are on your buyers' posts, not other creators' posts. Read who's in the comment section first. If it's your ICP asking questions, be in there. If it's other founders telling each other "great post," that's reach from people who will never buy.
On my own posts this year, the ones I replied on inside the first hour ran ahead of the ones I didn't. That's a pattern, not a law, but it's held up long enough that I treat it as 1.
Build my engagement list.
- Every morning, give me:
- 5 X posts from the last 12 hours in my niche worth replying to, with my reply drafted in my voice, 2 sentences max, adding a take or a number, never just agreement
- 2 to 3 X posts worth quote posting, with my take drafted. Pick posts that did 3x the account's usual numbers or that my buyers are arguing in
- 20 posts from my buyers on X, with a comment drafted for each. Start from [the buyer handles] and [the 10 to 20 accounts my buyers read], add anyone replying to my top posts who matches [my buyer], and keep that list on the engagement page. Prioritize posts with buyers in the comments, not creators. For LinkedIn, use [the connector] or [the buyer profiles I've pasted in], and skip it if neither is there.
- When I tell you an X post is live, pull the comments and draft a reply to each that earns a reply back, then check again 30 minutes later and draft the new ones. I'll post them. For LinkedIn I'll paste comments in and you draft the same way.
- Log who I engaged with and whether they followed or replied. If a DM comes out of it, I'll log that in the CRM.
- You never post, comment or DM yourself. You write the text, I send it.
The weekly report and the learning loop
Everyone tracks impressions because it's the number sitting right there when you open the app. You've got 60 posts up and no idea which ones put a person in your DMs. The 1 that did 40,000 impressions usually made 0.
So every post gets logged with its platform, topic, hook type, format, bookmarks, comments from buyers, profile visits, DMs and any call or opt-in you can tie to it. Then every Friday the bot compares the posts that made conversations against the high impression posts that made nothing, and tells you what's different.
A few things about where the numbers come from:
-
X profile visits only exist for posts from the last 30 days, so the report logs them for this week's posts and doesn't look back.
-
LinkedIn numbers come from the export you drop in on Friday, so that's a step in your week, not a 1 time setup.
-
The bot never reads your inbox on either platform, so DMs and calls get logged in the CRM or the sheet by you, tagged with the post that started them. That's where Friday's DM count comes from.
It writes 3 to 5 plain lessons into the rules page, and every other job reads those lessons before it works. That's the loop. Research gets sharper about what to flag, the writer gets sharper about what to draft, and the whole setup gets a bit better every week.
Say out loud that half of this isn't trackable. The person who finally answers your cold email 3 weeks later because they recognized the name never shows up in attribution. Track what you can and treat it as the floor.
Write the weekly report.
- Every Friday, pull every post from the week into the tracking page: platform, topic, hook type, format, job, impressions, bookmarks or saves, comments from buyers (read the commenters' bios on X, and titles on LinkedIn if a connector is on), X profile visits for this week's posts, and any DM, opt-in or call tagged to it in [CRM or sheet]. Count DMs only from that log, never from my inbox.
- For LinkedIn, read the export I dropped in this week. If there isn't a new 1, tell me and skip the LinkedIn numbers. Never estimate them.
- Rank the week by DMs and calls, not impressions. Compare the posts that made conversations against the high impression posts that made nothing and tell me what's different.
- Tell me which post type to do more of and which to kill, with the evidence
- Write 3 to 5 plain lessons into the rules page that every job follows, and say what changed after each 1
- Use only real numbers. Mark anything estimated or not yet available as such.
The weekly schedule
Every morning you get 1 short message: today's drafts, the engagement list, and anything that broke out in your niche overnight. You edit, attach the image, post, tell the bot it's live, work the reply queue, log any DM conversations in the CRM, and sit in your own comments for the first hour.
Sunday night the plan for the week shows up for you to approve. Friday you drop in the LinkedIn export, and the report and the lessons come back. The profile audit runs once a month.
Your time is about an hour a day, plus 1 voice note a week where you ramble about your space for 10 minutes. The bot turns that into most of next week's queue.
Lmk if you have any questions 👇
沿着这个主题继续读
从一篇文章,找到相关的教程、经验和阅读路径。


