Someone published a setup guide for a bot called The Expert Simulator. The subtitle says it out loud: turning you into a LinkedIn thought leader without ever having an original thought. There is a disclaimer on page one about entertainment, education, and platform terms of service. It is a joke.
The joke is a fully wired n8n workflow.
Scrape the top viral posts in your niche every morning. Rewrite them so they read as yours. Score each one for how much it will spark a fight. Stamp a hook on top. Generate a whiteboard image. Post it. Then auto-reply to every comment and fire a humble-brag follow-up the second you cross a hundred likes. It runs on four services and one workflow file. Total cost is about twenty to fifty-five dollars a month.
You should not run the whole thing. Parts of it are spam and parts of it break LinkedIn's rules. But the shape underneath is the real lesson: a content pipeline that watches a market, drafts, ranks, and ships on a schedule. Read it as a build, keep the honest half, and you have something worth owning.
What the system actually is
Three loops running on their own timers.
The main loop fires once a day. It pulls posts, rewrites, scores, decorates, and publishes. That is the part with real teaching in it.
The comment loop runs every few minutes. It watches for new comments and answers each one with a canned line like "Great point" or "Couldn't agree more."
The alert loop runs every fifteen minutes. It checks the like count and, past a threshold, posts a follow-up bragging about the numbers.
The first loop is a legitimate pattern. The other two are the part that gets your account flagged. Hold that split in your head while you read the build.
The stack: four services, one workflow
Everything hangs off four accounts wired into n8n. Nothing here is exotic.
| Service | Job | Cost |
|---|---|---|
| Apify | Scrape LinkedIn posts | About $5 per 1,000 posts. Free tier covers 1,000 a month |
| Anthropic (Claude) | Rewrite, score, and add hooks | $10 to $30 a month at daily use |
| Google Gemini (Imagen) | Generate the images | $0.02 to $0.04 per image |
| Unipile | Read and post to LinkedIn | Varies by plan |
The scraping runs on Apify's curious_coder/linkedin-post-search-scraper actor. Claude does the language work. Imagen makes the picture. Unipile is the bridge to LinkedIn, since you cannot post to LinkedIn from a raw API key alone. n8n holds it together and runs the schedule.
One thing to notice before you build anything: the intelligence is cheap. The bulk of the monthly bill is scraping and images, not the model. When people tell you AI content is expensive, this is the counterexample sitting in a parody PDF.
Step one: wire the four credentials
Each service becomes a credential inside n8n. The guide walks each one, but the pattern is identical every time. Get a key, drop it into an n8n credential of the right type, attach it to the nodes that need it.
Apify and Unipile use HTTP auth credentials. Apify goes in as an HTTP Query Auth credential with the token as the token parameter. Unipile goes in as an HTTP Header Auth credential with your key under the X-API-KEY header. Anthropic and Gemini each have their own native credential type in n8n, so you paste the key and you are done.
Your first real run: set up only the Apify credential, then execute the scraping node by itself. If posts come back with engagement numbers attached, that credential is live. Do that before you touch anything else.
The mistake that makes it fail: connecting Unipile without actually linking your LinkedIn account inside Unipile first. The credential will look valid and every post will fail with "account not found," because the key works but there is no LinkedIn session behind it. Connect LinkedIn in Unipile, copy the account ID it gives you, and keep that ID handy for the next step.
Step two: set the configuration node
The whole workflow reads from one node called Set Configuration. Four values live there.
Copy this.
{
"industry_keywords": "AI, automation, productivity, leadership",
"apify_actor_id": "curious_coder/linkedin-post-search-scraper",
"unipile_account_id": "YOUR_ACTUAL_ACCOUNT_ID",
"humble_brag_threshold": 100
}
The keywords decide what gets scraped. Match them to the room you actually want to be seen in.
| Niche | Keywords |
|---|---|
| Tech and AI | AI, machine learning, automation, tech leadership |
| Marketing | marketing, growth, branding, content strategy |
| Sales | sales, B2B, revenue, pipeline, closing |
| Startup | startup, entrepreneurship, founder, VC, growth |
The unipile_account_id is the ID from step one. The humble_brag_threshold is the like count that triggers the brag loop. Leave that one alone, or better, delete the loop it feeds. More on that below.
The mistake that makes it fail: keywords too broad. "Business" or "leadership" on its own pulls a wall of generic posts and the whole pipeline drafts mush. Three or four tight terms beat one big one.
Step three: the daily loop, stage by stage
This is the part worth studying. Six stages, in order, every morning.
Scrape. The schedule fires at nine, sends your keywords to Apify, waits sixty seconds, then sorts what comes back by engagement and keeps the top fifty. This is the input layer. It is watching a live market instead of guessing what to write.
Rewrite. Each post goes to Claude with one instruction. The prompt in the build is this:
Rewrite this post completely so it sounds like YOUR original thought. Requirements:
- Keep the core insight but change everything else
- Use different structure, examples, and metaphors
- Make it sound authentic and personal
- Keep it to 1200 characters max
Read that prompt as the tell. It runs at a high creativity setting and its entire job is to keep someone else's idea while hiding where it came from. That is the line. We come back to it.
Score. Claude reads each rewrite and rates it zero to one hundred on five things.
| Metric | What it measures |
|---|---|
| controversial_potential | Will it start an argument |
| humble_brag_density | How much quiet bragging is baked in |
| fake_support_likelihood | Chance of generic "so true" replies |
| hook_strength | Strength of the opening line |
| shareability | Odds someone reposts it |
Look at what this scorer optimizes for. Not whether the post is true or useful. Whether it will provoke and spread. That is what a metrics-chasing feed rewards, and it is the wrong target if you want to be worth reading.
Add a hook. The winning post gets one of a handful of stock openers stapled on: "Here's what nobody tells you," "Unpopular opinion," "Most people get this wrong," "After 10 years in the industry." You have read all of these a thousand times. That is the point, and also the problem.
Generate an image. Imagen makes a hand-drawn whiteboard-style square to sit above the text.
Post. Unipile pushes the text and image to your LinkedIn.
That six-stage spine, watch a market, draft, rank, decorate, publish, is a real and reusable thing. What poisons this version is what fills each stage: stolen ideas in the rewrite, provocation as the scoring target, canned hooks on top.
Step four: the two loops to delete
The guide has two more loops. Both of them are where a content tool turns into a bot that gets you banned.
The comment loop watches your posts and auto-replies to every new comment within about four minutes, using lines like "Love this take" and "This resonates." Someone takes the time to write you a real reply and a machine answers with a stock phrase. That is not engagement. It is a way to insult everyone who bothered to respond.
The alert loop checks your like count every fifteen minutes and, once you pass a hundred, posts a follow-up drawing attention to the number. It manufactures a brag on a timer.
Build the workflow if you want to learn n8n from it. Leave these two loops switched off. There is no honest version of a machine that fakes your replies.
Where this crosses the line
Be plain about it, because the PDF is not.
Auto-rewriting other people's posts to pass them off as your own is the core of this thing, and it is theft dressed up as a feature. The rewrite prompt exists to launder the source. No prompt engineering makes that okay.
Auto-replying to comments with canned lines is a bot pretending to be you. Auto-bragging about your own metrics is the same. Both are the kind of automated behavior LinkedIn's terms are written to stop, and the disclaimer on page one already admits it.
And the rules are not even the worst of it. The scoring model is tuned for controversy and reposts, not for being right. Run this and you become a slightly louder echo of whatever was already loud in your niche this week. That is the opposite of standing out.
So the answer to "should I run the Expert Simulator" is no. Run the parts that teach you, keep none of the parts that fake you.
What to keep
The wiring is genuinely useful once you change what flows through it.
Keep the market-watch stage. Scraping the top posts in your niche each morning is a real research tool. Read them yourself. See what your buyers are reacting to this week. That input is worth having whether or not a bot ever writes a word.
Keep the schedule and the publish step. A workflow that drafts from your own notes and posts on a fixed cadence solves a real problem: consistency. The failure mode of most owners on LinkedIn is not bad writing. It is silence for three weeks, then a burst, then silence again.
Kill the laundering prompt. Replace it with your own take on what you read in the scrape. The machine handles the treadmill. You supply the thought. That is the split that keeps it honest, and it is the same wiring, minus the theft.
If you want the version of this that starts from your ideas and pitches to a real list instead of scraping strangers, that is a different build. I walk through one in the LinkedIn lead-magnet workflow, where the automation writes to people who already raised a hand rather than faking a crowd.
If you do one thing this week
Build the scrape node and nothing else. Point it at four tight keywords for your niche, run it once by hand, and read the fifty posts it pulls. No rewriting, no posting, no loops.
That single node is the honest core of the whole system. It shows you what your market is talking about today, which is worth more than any auto-posted opinion. If you never wire up the rest, you have still gained the one part with real value. Twenty minutes, one credential, and you keep only the piece that does not lie for you.