# LinkedIn algorithm: how the feed is ranked and what changed in 2026

> What LinkedIn documents about how it ranks the feed, what changed on March 12, 2026, and what to adjust if you post for a brand or from your own profile.


**LinkedIn builds your feed from posts in your network plus recommendations from people you don't follow, and ranks them by what it estimates is relevant to you: your profile, your activity and what each post is about.** To show a post outside its author's network, it also looks for enough context and content that isn't overly promotional. On March 12, 2026, LinkedIn [announced](https://news.linkedin.com/2026/ImprovingTheFeed) two changes: ranking backed by language models that better understands what each post is about, and less room for engagement bait, recycled content and fake conversations.

LinkedIn doesn't publish a formula or how much each signal weighs. What follows is what the platform does document, kept apart from what circulates without a source.

## What LinkedIn looks at to rank the feed

According to its [help page on relevance](https://www.linkedin.com/help/linkedin/answer/a1339724?lang=en), your feed includes posts from your connections, from people and companies you follow and from your groups, plus relevant updates from outside your network. To rank them, it combines three types of signals:

| Signal type | What it includes, per LinkedIn | What it means if you post |
| --- | --- | --- |
| Identity | The reader's profile details, such as location, workplace and skills | Your post is competing for the attention of a specific professional profile |
| Content | How often the post is viewed or engaged with, what it's about and whether it provides knowledge or advice, how recent it is, whether it comes from a connection or someone the reader follows, its language, how constructive or professional the conversation is, and tagged people, companies or topics | Popularity counts, but so do the topic and the tone of the conversation it starts |
| Activity | What the reader has reacted to, commented on or shared, who they interact with frequently or recently, the content they spend the most time viewing, who they follow and the topics they've shown interest in | Two people see different feeds even if they follow the same accounts |

The same page says these systems reduce the visibility of low-quality or unsafe content, and that feed distribution isn't influenced by payments from third parties to LinkedIn, except for promoted content, which is clearly labeled.

Readers can also change the order: see the newest posts first for a single session, or make that the default in Settings. The options are called **Most recent posts** and **Most relevant posts**. LinkedIn adds that outside the European Economic Area, the per-session option may only be available on desktop.

## What a post needs to reach people who don't follow you

LinkedIn recommends posts from people you don't follow or aren't connected to when it considers them professionally relevant. Its [help page on recommended posts](https://www.linkedin.com/help/linkedin/answer/a1499047?lang=en) explains that when it recommends content outside the author's network, it tries to serve only what is valuable to those people, for example by making sure it provides sufficient context, isn't overly promotional, and has related elements such as text, images and video.

As example topics it lists professional growth, development and education, being inspired at work, professional products or services, and education on professional topics such as economics or technology. It also considers member disinterest signals and reports, and may review content to confirm it is high quality and professionally relevant.

For a brand, that comes down to two questions before you post. Would someone who has never heard of you understand it? Is it useful to them even if they never buy anything? A post that says "As we mentioned yesterday, it's live, link in the comments" fails both.

## What changed on March 12, 2026

That day LinkedIn published the [announcement](https://news.linkedin.com/2026/ImprovingTheFeed) and a [post from its engineering team](https://www.linkedin.com/blog/engineering/feed/engineering-the-next-generation-of-linkedins-feed) with the technical detail. They describe two separate changes.

### Ranking that understands what each post is about

LinkedIn started ranking the feed with larger sequence models, which it calls Generative Recommenders, augmented with large language models. According to the announcement, they better understand what a post is about and how your professional interests evolve, based on what you choose to share (industry, skills, experience and geography) and how you engage with content over time. What you're shown can come from your network or from professionals you haven't met yet.

The engineering post adds four details that are useful when you write:

- To represent each post, the system uses its text, its format, information about the author (name, headline, company and industry) and its engagement counts.
- It connects related topics even when they share no words. Its example: someone interested in electrical engineering who engages with posts about small modular reactors may be shown related content that keyword matching would miss.
- The ranking model processes more than a thousand of each person's past interactions as a sequence, to pick up patterns over time and long-term interests.
- Ranking relies on professional signals and engagement patterns and, according to LinkedIn, never on demographic attributes.

### Less engagement bait, recycling and fake conversations

The announcement names four practices. Read the verb for each one, because it doesn't say the same thing about all of them:

| Practice | What it says it does | Original wording |
| --- | --- | --- |
| Engagement pods | Work to make them ineffective | *We're working to make engagement pods ineffective* |
| Comment automation and third-party tools that create fake conversations | Curb them | *curb comment automation or third party tools that create fake conversations* |
| Engagement bait, such as "comment to agree" prompts or videos that don't match the text | Reduce it, along with repetitive, low-substance posts | *reduce repetitive, low-substance posts and engagement bait, such as “comment to agree” prompts or videos that don't match the text* |
| Recycled and click-driven posts | Reduce them | *We're reducing recycled and click-driven posts* |

In its section on authentic conversations, LinkedIn says it takes action to stop behaviors that promote inauthentic engagement, and adds that this isn't about limiting reach or discouraging participation: it wants to make sure that when you engage with a post, you're interacting with a real person and a real point of view. Its stated goal is a feed that is relevant to your interests, "not a popularity contest." The announcement said the changes were already rolling out and that it would keep testing.

**What it says, and doesn't say, about comment automation.**
LinkedIn names comment automation alongside engagement pods and "unauthorized third-party tools," always in the context of fake conversations and inauthentic engagement. The announcement doesn't define what counts as automation, doesn't mention AI in comments or replying to comments on your own posts, and doesn't say what happens to the reach of people who use it. The test it does make clear: a real person with a real point of view behind every interaction.

## What to do if you post for a brand or from your own profile

None of this guarantees reach. Each point rests on something LinkedIn documents; where it's our own reading, we say so.

1. **Write for someone who doesn't know you.** Make the post explain who is speaking, what happened and why it matters, without relying on an earlier post. Sufficient context is one of the conditions LinkedIn mentions for recommending outside your network.
2. **Make the topic clear in the text.** The system reads the text and connects related topics, and one of the content signals is whether the post provides knowledge or advice. A specific decision from your industry, well explained, says more about the topic than a row of hashtags (that last part is our reading).
3. **Keep the author's headline consistent with what they post.** The system uses the author's name, headline, company and industry to represent each post. Our reading: if your headline says "Head of operations at a food plant" and you post about the cold chain, the system has more context to connect the two.
4. **Ask for a reply only your audience can give.** LinkedIn is reducing "comment to agree" prompts and counts how constructive or professional the conversation is. A question people answer from experience starts a different conversation.
5. **If you come back to an idea, add something new.** Reposting the same text gets close to the recycled content LinkedIn says it reduces (our reading). The same goes for a video unrelated to the text that goes with it.
6. **Stay out of engagement pods and away from tools that comment for you to simulate conversation.** LinkedIn says it's working to make the first ineffective and to curb the second.
7. **Reply to comments as a real person or a real brand.** Read what each person said and answer that, in the account's voice. If you use AI to draft, review before you post: a reply that would fit under any comment drifts away from the test of a real person with a real point of view (our reading). In HeyMark, if you post from a LinkedIn Company Page, Mark drafts the public reply in your brand's voice and you approve it before it goes out; making sure it answers that specific comment is still your call.

*Fictional example: a made-up distribution company, with no real data.*

**Works**

**"This quarter we stopped promising 24-hour dispatch on orders over 30 boxes, because one in three was arriving late. We told customers about the new lead time before they asked. How do your teams handle it?"** Context, a specific decision and a question people answer from experience.

**Doesn't**

**"AI is changing logistics. Agree? Comment YES."** Nothing an outsider can use, and the kind of prompt LinkedIn says it reduces.

The official sources cited here don't document a difference in reach between a personal profile and a Company Page. To decide where to post from, see [personal profile vs. Company Page on LinkedIn](/en/guides/linkedin-profile-vs-company-page/). And if you're worried your posts sound generic, LinkedIn also lets members flag AI content they find generic or repetitive, as we explain in [how to avoid AI slop on LinkedIn](/en/guides/avoid-linkedin-ai-slop-brands/).

## What LinkedIn hasn't confirmed

**The "golden hour."** A common claim is that what happens in the first hour decides a post's reach. LinkedIn does list how recent the content is and how often it is viewed or engaged with among its signals, but none of the official pages cited here mentions a time window, whether an hour or any other length. There's no basis to confirm it or to rule it out.

**Exact weights.** The official sources cited here don't say how much a comment is worth compared with a reaction or a minute of reading. A percentage like that needs a source to back it up.

To find out whether a change affected you, compare your own posts by topic and format, over similar periods, with one change at a time. A single drop doesn't prove LinkedIn reduced your content. If you manage a Page, you can [export its analytics](/en/solutions/export-linkedin-page-analytics/) to compare months.

## Plan your next post in HeyMark.

Keep the idea, review the draft with your team, and see how it performed in the accounts you connected.

[Start free](https://app.heymark.ai) · [See how it works](https://heymark.ai/en/#product)

## Developer and agent resources

- [HeyMark MCP documentation](https://heymark.ai/en/mcp/)
- [llms.txt](https://heymark.ai/llms.txt)
- [Full site content for language models](https://heymark.ai/llms-full.txt)
- MCP protocol endpoint: `POST https://mcp.heymark.ai`
- OAuth protected-resource metadata: [/.well-known/oauth-protected-resource](https://mcp.heymark.ai/.well-known/oauth-protected-resource)
- MCP server card: [/server-card](https://mcp.heymark.ai/server-card)
