LinkedIn algorithm penalties: actions that kill post impressions
A LinkedIn algorithm penalty is the label people put on a post whose impression count collapsed. LinkedIn does not publish a penalty score, a shadowban flag, or a percent deducted from reach. It publishes a short list of actions that change how many screens a post is allowed to occupy. Some of those actions limit distribution past the people who already follow you. Some remove the post. Some restrict the account, so the next post never becomes a display at all.
An impression is a display. LinkedIn's combined post analytics defines it as the number of times a post was shown on screen, and it says the number is an estimate. Members reached is the count of distinct members and Pages. The same screen splits displays into in network (people who follow you or are connected to you) and out of network (everyone else). How retrieval and ranking assign those displays is the LinkedIn algorithm guide. This page is the action list in front of that system: what you did, which public note covers it, and what the impression row can still tell you the next morning.
Boykli drafts from a note, a link, a file, or a Voice Agent transcript, and it publishes an approved draft to the personal profile you connect. It does not detect a penalty, score impressions, join a pod, leave a comment, send an invitation, or message anyone. The Algorithm page is a reading of LinkedIn's public notes. Your impression number still lives on LinkedIn.
Logged-in visitors can browse Radar and Algorithm with no plan. Writing, carousels, Voice Agent, publishing, and scheduling need a paid seat. BYOK is $29 a month. Managed is $88 a month. Prices are the public list on boykli.com as of 27 October 2026.
Three ways an impression count dies
Treat these as separate incidents. A fix for one does nothing for the other.
| What you see the next day | The name people use | What the public record supports | What the next post can still do |
|---|---|---|---|
| The post is on your profile. Impressions are almost all in network. Members reached is a small slice of followers | "The algorithm penalized me" or "shadowban" | Distribution was tapered. 4 June 2026: content that looks generated and lacks a clear perspective is less likely to travel beyond the immediate network. Relevance help: algorithms filter or taper low-quality and unsafe content. 15 September 2026: views of content classified as AI slop were down 40% | It can still display. A later post with a specific perspective is a new candidate. There is no published trust score you have to wait out |
| The post is gone from the profile | "They suppressed it" | Removal. Professional Community Policies cover spam, untargeted or gratuitously repetitive promotion, and artificial engagement. Violating content can be removed | A new post can go up. The removed one does not return because you changed the hook |
| LinkedIn shows a restriction, and a feature comes back at a time written on the notice | "Permanent shadowban" | The account was restricted for automated activity on the website. Automated activity: disable the software or extension. The account is re-enabled at the time on the notice | Until that time, the restricted feature does not run. New posts do not display because they do not publish. After the time, content rules still apply |
| A comment you "left" with a tool never appears under Most Relevant, or never leaves the commenter's network | "My comments have no reach, so my posts are dead" | Automated comments are handled as inauthentic. The 13 March 2026 note says detected ones may stay out of Most Relevant and sometimes out of view beyond the commenter's network, and that LinkedIn may restrict the member | That row is about the comment, not the impression count on your own post. A restriction, if it arrives, is the third row |
The impression split is the audit for the first row. Boykli does not import it. A full account of the restriction notice versus an OAuth queue is automated LinkedIn posting.
Actions LinkedIn has tied to fewer displays
Each row is an action a founder, operator, or creator can actually take this week. The effect column uses only figures and mechanisms LinkedIn has printed. Vendor charts that assign "47% less reach" to a phrase are unpublished.
| Action | What you did | What happens to displays | Where LinkedIn said it |
|---|---|---|---|
| Engagement-bait closer | "Comment YES if you agree." "Like this if you are a founder." "Comment the keyword and I will send the doc." The useful object is off the post, or the reaction is the product | The feed is being tuned to show less of this. On 12 March 2026, Tim Jurka named "Comment 'Yes' if you agree" as engagement bait the feed would show less of, set against an actionable breakdown or a specific perspective | Jurka, 12 March 2026. No public penalty percent, and no list of banned phrases with a score |
| Engagement pod | A group agrees in advance to like, comment, and re-share inside a window so the early counts look busy | Pods are a named target. LinkedIn said it had removed groups that showed pod behavior and had warned members whose actions showed pod participation, including possible removal from programs such as Top Voices. Jurka said the company is working to make pods ineffective. Pre-agreeing to like or re-share is artificial engagement under the community policies | 13 March 2026 newsroom, Jurka, Professional Community Policies |
| Automated comments and likes | A browser extension, a script, or a third-party tool posts the comment or the like without you clicking | Those comments may never reach Most Relevant, and may never leave the commenter's network. The September note says hundreds of thousands of automated comment attempts are blocked a day, and that billions of automated attempts (posting at scale, and slop) were blocked over a couple of months. The member can be restricted. Prohibited software names bots that comment, like, share, or otherwise drive inauthentic engagement | 13 March 2026, 15 September 2026, prohibited-software help |
| Generic, recycled, or attention-gaming text | A polished paragraph any account could sign. The same lesson recycled. A list built to game attention. No incident, no number you measured, no decision you will live with | Less likely to be distributed beyond the immediate network. Early tests identified generic content correctly 94% of the time. Views of content the classifier calls AI slop were down 40%. More than one million members had used "Seems like AI slop" since August. If reports cross LinkedIn's threshold, the author may see a tip in post analytics. LinkedIn does not publish that threshold | 4 June 2026, 15 September 2026 |
| Mismatched video | A video that has nothing to do with the text, attached because someone said video gets more impressions | Jurka named this as inauthentic distribution the feed would show less of | Jurka, 12 March 2026. Format-relative dwell is a separate subject in video vs carousel vs text. A native video that matches the text is a different object |
| Repetitive promotion | The same offer, untargeted, every weekday. Invitations used as the ad | Community policies tell members not to post untargeted, irrelevant, unwanted, or gratuitously repetitive promotional content, and not to use invitations as a promotion channel. Jurka also called out recycled thought leadership that does not say much | Professional Community Policies, Jurka. There is no published daily posting cap in these notes |
| Website automation | Software or an extension that scrapes LinkedIn, holds the session, or clicks Post, Like, Comment, Connect, or Message on the website | This is the restriction path. While the feature is off, nothing new displays, because nothing new publishes. The help page tells you to disable that software. The account is re-enabled at the time on the notice | Automated activity, prohibited software |
Two boundaries keep this table from being abused.
Using a model to turn a real incident into a draft is a different act from the generic row. The June note says using AI to help write is acceptable when the post is still your perspective. September's 40% is views of content their classifier labeled, and it is LinkedIn's figure. A line-by-line test of a bait closer against a post that already contains the point is engagement bait versus deep value. A closer that says “tag a founder who needs this,” then a row of names who were not in the work, is the same bait class. A mention of the person the sentence is about, picked from LinkedIn’s list, is a different act. That split is how to tag people on LinkedIn.
Publishing through an application you approved on LinkedIn's OAuth screen is a different act from the website-automation row. Boykli is the OAuth case. Sign-in happens on LinkedIn. The scopes are openid, profile, email, and w_member_social. The queue sends one approved item to the personal profile at the slot you saved. It does not browse the feed, and it does not upgrade distribution. A generic post sent through that grant is still a generic post.
Write the post you can defend, then queue it
Create posts and Voice Agent sit on a paid seat. BYOK is $29/month with your keys. Managed is $88/month with platform models and monthly caps. The queue does not comment, connect, or message, and it does not score impressions.
boykli.com →Penalties that circulate, and the record behind them
Buyers search this topic after a roundup told them a single habit "kills reach." These are the habits those roundups name. The right column is what the sources above actually support.
| Claim that circulates | What you can rely on |
|---|---|
| An external link kills impressions by a fixed percent | None of the 2026 notes above publish a link tax. A post can be thin with a link or without one. Judge the discovery split, and the object in the text |
| Hashtags, or repeating the keyword, are a ranking penalty or a ranking boost | Retrieval in the March 2026 engineering account matches embeddings, including topics that do not share words. The post text is in the prompt. A stuffed hashtag block is a writing choice. It is not a published impression switch. The mechanics are in the algorithm guide |
| Missing the first hour applies a penalty for the life of the post | LinkedIn has not published a first-hour cohort size. A new post is embedded in near real time, and popularity percentiles refresh within minutes. Early reading can move the bucket. Minute 60 is not a published switch |
| Hold the reader for 30 seconds, or 61 seconds, or the post is penalized | The 2024 dwell note, still named by the 2026 ranker, says a universal long-dwell cutoff failed. Their illustration of that failure is 30 seconds, because 30 seconds is a different event on an image than on a video. Long dwell is a percentile inside similar posts. Zhang, Kothari, and Tiwana |
| Editing a post after it is live zeros the impressions | That rule is not in these notes. If you edit, you still only know the outcome from the discovery block the next day |
| An approved scheduler is itself an impression penalty | Content rules apply to the text however it leaves. The restriction help page is about software that automates activity on the website. LinkedIn's own composer clock is first-party scheduling. An OAuth grant is a permission LinkedIn issued |
| One weak post permanently tanks later posts | A restriction is a notice with a time. A distribution taper is described per the class of content. These sources do not publish a decaying account score you can look up |
| Leaving a pod "resets" you, so impressions must fall as punishment | The early likes were the pod. When the subsidy stops, the count drops to what strangers actually did. That drop is the measurement becoming honest. The month after you leave is the subject of growing without pods |
If a figure is not in the right column, do not operate on it. Changing the hour, the hashtag count, and the closer in the same week makes the next impression delta unreadable.
A teardown of the last eight posts
Run this on a personal profile. Boykli publishes only to the personal profile you connect with OAuth. A company Page uses Page analytics, and this product does not queue to it. Budget about an hour. Do it once, then hold the variables still for three posts.
- Write the discovery row (about 20 minutes). For the last eight posts that are still on the profile, copy date, format (text, image, document, or video), impressions, members reached, in-network percent, and out-of-network percent. Add two marks the analytics screen will not compute for you: is there a restriction notice anywhere on the account, and did any of these posts disappear. LinkedIn describes the combined analytics graph as complete through yesterday, so ignore a post from this morning.
- Tag the action, not the mood. Give each post one tag from the action table, or "specific." Specific means a stranger can repeat a rule, a number you measured, a sequence, or a constraint without reacting. "The algorithm hates this topic" is not a tag. If the closer is a YES, a keyword, or a like-for-option, tag bait even when the paragraph above it is sharp.
- If a restriction notice is up, stop the teardown. Disable the extension, the cloud browser, or the tool that asked for the LinkedIn password or the session cookie. Wait for the time on the notice. A second tool that also acts in the LinkedIn tab is the same class of software. Do not publish a "comeback post" through it.
- If posts are missing, read the removal rules before you rewrite. Untargeted repetitive promotion and artificial engagement are named. A new hook on a new post is a new object. It does not restore the removed one, and it does not require a pod to "recover authority."
- If the posts are live and almost entirely in network, read the text against the June test. Could this paragraph sit on another author's profile about another week? If yes, it is the class the note says is less likely to travel past people who already follow you. Replace it with one decision you can sign. The four-line receipt — incident, refusal, mechanism, transfer — is the draft shape in the thought leadership framework. The reason the sentence has to come from material you brought is the constraint in building Boykli.
- Stop one action for the next three posts. Pick the highest row that you actually did: website bot, then pod or automated comments, then bait closers, then recycled text. Keep the format the same. Keep one clock: the weekdays you toggle, one time of day, one timezone. The hour itself is chosen in best time to post on LinkedIn. Changing the hour and the topic in the same week makes the impression delta unreadable.
- Read the split again at least a day later. Look at whether out-of-network displays moved, and whether members reached moved with impressions. A post that only recirculates among people who already follow you can still be the right post for buyers in that room. It is a different outcome from one that retrieval is handing to strangers. A flat out-of-network share after you removed the bait closer is information. It is not proof of a hidden flag.
- Leave the pod before you trust step 7. Pod likes arrive inside the window whether the post is specific or empty, so they hide the stranger test. The follow spike on pod day is the same subsidy. Judge the three clean posts after the agreement is off.
Boykli can hold steps 5 and 6. Create posts drafts from the note you supply. Voice Agent may use only the transcript of an interview you sat. The preview shows the line breaks before the post leaves. The queue sends the approved item. None of those steps read the discovery block, apply a penalty label, or add a first comment. Managed caps are 210 posts, 150 carousels, 150 images, 25 Voice Agent sessions, and 130 post refinements a month. BYOK uses your OpenRouter key for chat and images and your Google AI Studio key for Voice Agent, with no Boykli generation cap. You pay those providers for usage.
What to stop this week, in order
Stop the action that can remove the post or restrict the account before you tune a hook.
| Priority | Stop | Leave in place while you test |
|---|---|---|
| 1 | Any tool that acts on the LinkedIn website: comments, likes, connects, messages, or a cookie-driven poster | LinkedIn's own composer, and an app you approved on LinkedIn's screen that only publishes what you already signed |
| 2 | The pod, and any extension that comments for you | Five comments you type yourself, each with a fact, on posts you actually read. A script that does fifty before lunch is the March prohibition |
| 3 | YES closers, keyword gates, and "comment for the checklist" | A question that asks for a fact, after the post already contains your answer |
| 4 | The recycled promo and the topic-only prompt ("write five viral posts about leadership") | One incident from this week, drafted with a model if you want, then edited until you would say it on a call |
| 5 | Only after 1–4 are off: arguments about the posting minute, the hashtag count, and whether a link "killed" the post | One clock, one format, three posts, then the discovery split |
The fifth row is where most penalty advice starts. It is the row with the least published mechanism. Running it first teaches you nothing, because the bait, the pod, or the generic paragraph is still in the prompt the ranker sees.
FAQ
Does LinkedIn apply an algorithm penalty with a hidden score?
LinkedIn has not published a penalty score, a shadowban status, or a percent taken off impressions. It has published distribution limits for generic and low-quality content, removal under the community policies, restrictions for automated activity on the website, and separate handling for automated comments. The number you can read is the impression estimate plus the in-network and out-of-network split on the post.
Do external links kill LinkedIn impressions?
The March, June, and September 2026 notes cited here do not publish a link tax. A link on a post that already contains the point is a different object from a post whose only content is "link in comments" or a keyword gate. Read the discovery split on the post you shipped. Do not adopt a percent from a roundup LinkedIn did not print. Where 2026 samples put a typed URL, a preview card, and a first-comment line against the same author's posts is links in the first comment versus the post body.
Will an AI-drafted post get fewer impressions?
The June 2026 note says using AI to help write is acceptable when the post is still your perspective, and that generic, perspective-free content is less likely to travel beyond the immediate network. The September note reports a 40% drop in views of what its classifier calls AI slop: low-effort, polished, and missing a point of view, including text that is generic, repetitive, recycled, or built to game attention. A transcript of a decision you made is the first case. "Write a viral post about this topic" is the second. Boykli is built for the first. It does not promise a distribution outcome.
How long does a LinkedIn penalty last?
It depends which incident you have. A restriction lasts until the time written on the notice, and the help page tells you to disable the software first. A removed post stays removed. A live post that stayed in network has no published cooldown. The next post is a new candidate. These sources do not publish a 14-day or 30-day recovery ritual.
Do engagement pods still increase impressions in 2026?
They can still move early likes inside the group. LinkedIn's March 2026 notes say the company is making that coordination ineffective, has removed groups that showed pod behavior, and has warned members, including possible removal from programs such as Top Voices. The community policies forbid artificially increasing engagement. The impression count you want, strangers who do the work you do, is the count the pod hides. When you leave, the drop is the subsidy ending.
Can scheduling a post cause an impression penalty?
Scheduling is a clock. LinkedIn's own composer can hold a date. Boykli's queue holds the weekdays you toggle, one time of day, and one timezone, and it publishes one due item at a time through the permission you granted. The text still faces the content rules. A tool that schedules by clicking the website, or that also comments and connects from your session, is the automated-activity case. That distinction is the automated posting guide.
Can Boykli tell me a post was penalized?
No. The Algorithm page explains the public notes. It does not read your impression row, apply a penalty label, or recommend a recovery sequence from your archive. Library stores what you wrote in Boykli. LinkedIn's discovery block is the meter. If enough "Seems like AI slop" reports cross LinkedIn's unpublished threshold, the September note says the tip appears in LinkedIn's post analytics, not in Boykli.
What should I check before I blame the algorithm?
Confirm the post is still on the profile, confirm there is no restriction notice, then read in-network versus out-of-network. Tag the closer and the source of the paragraph. If the closer asks for a token, or the paragraph could move to another account unchanged, you already have an action the 2026 notes describe. The posting hour is the test you run after that action is gone. A drop across many posts, measured as a median against an earlier window of the same format, is the separate diagnosis in how to diagnose and fix declining LinkedIn reach.
Ship three posts with the penalty action removed
Boykli drafts from a note you supply and queues the approved post to your personal profile. BYOK is $29 a month. Managed is $88 a month. It will not detect a penalty or run a pod.
boykli.com →