LinkedIn and X: from posts to clients

Courses sell “the algorithm”. With X you can simply read it: the code is on GitHub. In March 2026, LinkedIn explained on its engineering blog how the feed works. This playbook builds on those two, plus honestly labelled data.

Evidence labels: ✅ primary source (code, platform page). ⚠ secondary, vendor or self-reported. Weights and thresholds reflect the X code as of 31 October 2026 (commit 9106e9f). X says many values come from a configuration system and that the code shows the production values; so they can change.

How it really works

• X predicts, per viewer, the probability of ~25 actions and adds them up weighted. A reply weighs 5, a like 0.66, “share via copy link” 20 (the highest). Mute, block and report weigh heavily negative. ✅ Note: these are weights on probabilities, not “1 reply = 7 likes”.
• X gives small accounts a cold-start boost. Original posts from accounts with up to 50,000 followers, less than 2 hours old and with fewer than 200 impressions: one of these per feed refresh can be moved to a fixed slot (around position 15). A small chance at extra impressions, not a guarantee. After 48 hours a post drops out of For You. ✅
• X rewards mutual follows. Since July 2026, the reply probability gets +15 extra weight if viewer and author follow each other, only for original posts. ✅
• LinkedIn matches on topic via a language model. The prompt per post contains the format, your name, headline, company and industry, the engagement as a percentile and the post text. ✅ LinkedIn now also fetches many posts from people outside your network based on topic (our reading of the blog).
• Scrolling past counts negatively. LinkedIn trains on posts you saw but ignored; X has a “not dwelled” signal (small weight, −0.02). ✅
• Both hunt AI slop and pods. LinkedIn says it detects AI slop outside your direct network with 94% precision, and earlier reported 40% fewer views for slop. ⚠ (LinkedIn exec via Social Media Today)

flowchart LR
  P["Your post"] --> L["LinkedIn: headline + topic<br/>in the prompt"]
  L --> M["Match with interests<br/>of strangers"]
  P --> X["X: chance of reply, quote,<br/>share per viewer"]
  X --> K["Small account?<br/>Chance of extra slot in first 2 hours"]
  M --> R["Reach with your ICP"]
  K --> R

Who does it perfectly

Marc Lou (X)

• Does: build in public. Every launch, revenue figure and failure in public. Sells to makers, so his audience is his customer.
• Result: 20 products, over $3.2 million total revenue and ~420,000 X followers (TrustMRR, connected to payment providers, November 2026). ⚠ TrustMRR is his own platform.
• Steal this: share real numbers with proof. Verifiability stands out among all the AI slop.

Ruben Hassid (LinkedIn)

• Does: explains one AI topic every day with visuals and carousels, with a Substack newsletter as the end point.
• Result: hundreds of thousands of LinkedIn followers; exact figures differ per source and are not verified. ⚠
• Steal this: one topic, every time. LinkedIn’s language model then knows exactly where you belong.

Richard van der Blom (LinkedIn) 🇳🇱

• Does: publishes a report on the LinkedIn algorithm every year. That’s his lead magnet, proof of expertise and entry point for trainings.
• Result: by his own account, the 2026 report analyses ~1.3 million posts and is cited across the industry. ⚠
• Steal this: publish one dataset or benchmark of your own from your niche. That’s content others spread for you.

Honestly: we found no Dutch case

We looked for a Dutch or Flemish AI service provider that verifiably shows how many clients came in via LinkedIn. Not found. If you document it honestly, you’ll be the first.

Do this

1. Make your headline a topic label

1. Pick one topic and one type of client (ICP).
2. Write your headline in your client’s words. For example (made up): “AI chatbots for installation companies | fewer phone calls, more appointments”.
3. Featured: at most three items (case, lead magnet, newsletter).

Let an AI assistant guess your topic

Your headline is literally in the prompt of every post you make. ✅ Van der Blom’s test ⚠: paste your last 20 posts into an AI assistant and ask “what are the two main topics?”. If the answer doesn’t match your headline, LinkedIn won’t match you well either.

2. Opening line: topic plus tension

1. Write five opening lines and pick the most specific one (number, situation, contrast).
2. Good: “We cut 40% of a plumber’s phone calls with one chatbot flow. This is the flow.” Bad: “I want to share something…”.
3. No clickbait: “not interested” costs −52 on X. ✅

TopicWho forWhat it deliversCommon mistake

Copy all

Put the hook on screen, as on-screen text and in your first sentence at the same time. That way it also works with the sound off.

Free · runs entirely in your browser · nothing is stored or sent

3. Choose a format that performs

Format (LinkedIn)	Median engagement	Note
Image + text	2.77%	Best median
Document/carousel	2.52%	Most comments
Video	2.13%	Most views, only if it fits the text
Text only	1.98%	Median 0 comments

Source: Taplio, 219,456 posts, March 2026. ⚠ Vendor data from Taplio users only. Length that scores best: 1,000-2,000 characters.

4. Comment in the right places

1. Make a list of 20-30 accounts your clients read (on X a private List).
2. Respond 5-10 times a day with 2-4 sentences: a data point, counterexample or mini-case.
3. Follow back people who respond.

Mutuals are your real reach on X

The +15 boost on replies between mutual followers ✅ means: every substantive comment that leads to a mutual follow makes your next original post more visible to that person. Replies are also scored by a Grok model (0-3) ✅, so an empty “Great post!” probably scores low (our inference).

5. Post when you have two hours

1. Schedule your best post at a time when you can respond for two hours afterwards.
2. Answer every comment with a real follow-up question or addition.
3. LinkedIn: 2-5 posts per week (LinkedIn itself ⚠).

Make something people forward outside X

“Share via copy link” weighs 20, the highest positive weight in the X code. ✅ Templates, checklists, data and strong one-liners that people paste into Slack or WhatsApp are therefore rewarded directly. (That last part is our inference, not a statement by X.)

6. From content to client

1. List of 100 companies or people, by hand.
2. One to two weeks of real comments on their posts.
3. Connection request without a pitch, with one reason.
4. First DM only after interaction: give something (observation, mini-audit), ask one question, no calendar link.
5. One in four or five posts ends with a concrete offer: “This month I have room for 2 audits of your customer service flow.”
6. Build a newsletter or email list: Van der Blom estimates that’s where most converting happens. ⚠

Grey: possible, with risk

Tactic	Risk	Safer
Automation tools for requests and DMs	High. User Agreement §8.2 prohibits “bots or other unauthorized automated methods”. ✅ Restriction or ban, including of your main account.	By hand, 10-20 personal requests per day. Sales Navigator is official.
Scraping leads	High. Prohibited in §8.2 ✅, and in the EU you need a GDPR legal basis.	LinkedIn’s search filters, public company data, opt-in via lead magnet.
“Comment GUIDE and I’ll send you the PDF”	Medium-high. Engagement bait; LinkedIn: “Don’t do things to artificially increase engagement”. ✅	Link to a landing page, or simply put the core in a carousel.
AI ghostwriting at scale	Medium-high. Slop classifier and report button. ⚠	AI for structure and editing; the experience, number and opinion are yours.

Risk: rage hooks

Polarising gets replies, but mute (−58.8) and block (−31.2) weigh heavily negative in the X code ✅, and you attract the wrong audience. Go for “respectfully contrarian”: a well-argued opinion against the consensus, with data.

Black: how it works and why you don’t do it

Tactic	How it’s detected and penalised
Engagement pods	LinkedIn policy: “don’t agree with others ahead of time to like or re-share”. ✅ LinkedIn removes pod groups and limits the reach of artificially inflated posts. ⚠ On X, engagement via a group chat doesn’t count: “has no ranking impact”. ✅
Bought followers, likes or views	Prohibited in the X developer policy and the LinkedIn User Agreement. ✅ Fake followers don’t respond, so your engagement rate and predicted probabilities drop.
Bot networks and fake accounts	X prohibits multiple accounts responding to the same posts and publishes part of its detection systems (not all rules). ✅ LinkedIn only allows a profile for “a real person”. ✅
AI comments at scale	X scores replies with Grok and runs reply-spam classifiers. ✅ LinkedIn removes them from “Most Relevant” and restricts the account on repeat. ⚠
Automated replies on X	“Sending automated replies to posts based on keyword searches alone is not permitted”; can lead to permanent suspension. ✅ (via search result from help.x.com)

Security

• Fake recruiters spread malware, also via hijacked LinkedIn accounts. Real recruiters don’t send programs or archives: open nothing outside the browser.
• An app permission bypasses 2FA. Never click “Authorize” via a link from a DM, and regularly review your connected apps on X.
• Never give a growth tool your password or session: besides a ban risk, it’s a takeover risk. Turn on 2FA on LinkedIn and X; X also supports passkeys. More in Security.

Myths

• “A reply is 27× a like.” Outdated (2023) and the weights aren’t count values. ✅ debunked
• “Replying as the author is 150× a like.” That weight (75) is no longer in the 2026 code. Replying back is still smart, for other reasons. ✅ debunked
• “Golden hour on LinkedIn.” No official source. On X there is a 2-hour cold-start chance for small accounts. ✅
• “Premium gives a 4× boost.” Not traceable in the current code. ✅
• “Links on X are dead.” Contradictory: then X head of product Nikita Bier said in April 2026 “Links are not deboosted”; Buffer measured a median 0% engagement for link posts without Premium (2024-2025 data). ⚠ Test it yourself.
• “Hashtags help on LinkedIn.” Van der Blom saw posts without hashtags do 5-10% better. ⚠

Measuring

• Not just impressions. Measure profile visits, new followers per post, newsletter sign-ups, inbound DMs, conversations, quotes and clients, in one sheet of your own.
• Compare with the median for your follower group (Taplio, March 2026): 136 views with fewer than 1,000 followers, 318 with 1,000-5,000. Below that? Fix your hook and topic first, not your frequency.
• “Am I shadowbanned?” First check X’s Under the Hood (x.com/i/jf/under_the_hood, still in pilot, not available to everyone): there you see labels that limit your visibility. ✅

Your first 30 days

For a freelancer selling AI services (chatbots, automations):

1. Pick your platform. Selling to small businesses? LinkedIn. To makers and developers? X.
2. Day 1-2: rewrite headline, About and Featured around one topic and one client type.
3. Day 3: list of 20-30 accounts your clients read; start with 5 comments per day.
4. Week 1: two posts: one mini-case with a number, one explainer with an image or carousel.
5. Week 2: start a list of 100 potential clients and comment on their posts.
6. Week 3: make one lead magnet (flow template, checklist, mini-audit) with a landing page with one field.
7. Week 3-4: connection requests without a pitch; DM only after interaction.
8. Day 30: count conversations and quotes, not views. Keep what works, cut the rest.

Checklist

Checklist

0/4

• My headline names one topic in my client's words
• I have a list of 20-30 accounts to comment on substantively every day
• I only post when I can respond for two hours afterwards
• I don't use automation tools, pods or "comment GUIDE" posts

Your progress is only stored in this browser.

Quiz

Test yourself

1. 1Which action weighs the most of the positive actions in the October 2026 X code?

   A likeA profile clickShare via copy link

2. 2Why has your LinkedIn headline been so important since 2026?

   It's in the prompt of the language model that decides who sees your postOnly people with the same headline see your postsIt determines how many posts you may publish per day

3. 3You have 3,000 followers on X. When do you post your best post?

   In the middle of the night, when there's less competitionAt a time when you can respond for two hours afterwardsThe time doesn't matter