> **The idea, in one sentence.** This article isn't a stat list — it explains the **mechanism**. How the LinkedIn algorithm **decides who sees what** in 2026: the first-minutes window, the engagement signals ranked by weight, reading time (*dwell time*), and how reach concentrates on big creators. We look at the machinery on the post side **and** the comment side, backed by our first-party LinkHub data. Third-party figures are estimates from cited, dated studies — flagged as such. No invented numbers.

## Key takeaways

- **The algo tests, then amplifies.** A post is first shown to a small fraction of your network; engagement in the **first 60–90 minutes** decides whether it gets wider distribution or dies. That's the famous "golden hour." *(see [when to comment](/en/blog/quand-commenter-sur-linkedin))*
- **Not all signals weigh the same.** A **comment weighs more than a like** — estimated ~2x ([AuthoredUp](https://authoredup.com/blog/linkedin-algorithm)) to ~5–15x depending on the source *(third-party, disputed)*. On the LinkHub side, a 10+ word comment clearly weighs more than a like. *(see [comment vs like weight](/en/blog/poids-commentaire-vs-like-algorithme))*
- **Dwell time (reading time) has become a core signal**: LinkedIn measures how long you linger on a piece of content, not just whether you tap "like." *(2026 industry data, cited)*
- **Reach concentrates on top creators.** Per [van der Blom](https://www.linkedin.com/posts/richardvanderblom_chapter-1-algorithm-insights-report-2025-activity-7322514599126130688-Q895) (1.8M posts), "Top Creator" visibility rose from **15% (2022) to 31% (2025)** — which is why **commenting where reach is already concentrated** pays off.
- **For anyone who comments, this changes everything**: commenting **early** (0–30 min) earns **~3.8x** more impressions than after 24h, and **under the right creator** up to **~10x** the median. *(LinkHub — see [when to comment](/en/blog/quand-commenter-sur-linkedin))*
- **The mechanism doesn't judge who wrote it, but what the content triggers.** AI isn't penalized as such: the algo evaluates **relevance and engagement**, not authorship. *(see [are AI comments detectable](/en/blog/commentaires-ia-detectables-linkedin))*

## 1. How the algorithm distributes: test, then amplify

LinkedIn doesn't show content to everyone at once. The mechanism runs in successive stages:

1. **Quality filtering.** On publish, the content is shown to a **small fraction** of your network (on the order of a few percent). It's a test, not a broadcast.
2. **Engagement test (the golden hour).** The algo watches how that first sample reacts during the **first 60–90 minutes**. Likes, comments, shares — but above all **time spent** on the content — determine what happens next ([SocialBee, 2026](https://socialbee.com/blog/linkedin-algorithm/)).
3. **Extended distribution.** If the signals are good, the content is pushed to 2nd-degree connections and profiles with similar interests. That amplification can keep going for **48–72h** as long as engagement holds ([SocialBee, 2026](https://socialbee.com/blog/linkedin-algorithm/)).

The practical consequence is huge: **a post's first hour determines its trajectory**. It's true on the post side, and just as true on the comment side — because a comment posted early gets exposed to the host post's **audience peak**. That's exactly what our timing data shows: commenting in the **first 30 minutes** earns **391 average impressions**, vs **104 after 24h**, i.e. **~3.8x** (LinkHub, n = 261,137 — see [when to comment](/en/blog/quand-commenter-sur-linkedin)). The golden window isn't a metaphor: it's the test/amplify machinery that creates it.

To never miss that window, the practical blocker — monitoring the native feed nonstop — is solved by [personalized feeds](/en/features/feeds-personnalises): you group your target creators and see their posts **the moment they go live**.

## 2. Signals ranked by weight (comment > reply > like)

Not all engagement signals are equal. The algo ranks them, because they don't cost the same effort and don't signal the same intent:

- **The comment is the strongest signal.** It takes thinking and writing; it often draws a reply from the author → a **thread** that extends distribution. The size of the multiplier is debated — from **~2x** ([AuthoredUp, 2025](https://authoredup.com/blog/linkedin-algorithm)) to **~5–15x** per industry estimates — but the direction is constant: a relevant comment weighs clearly more than a like. *(third-party, disputed)*
- **The reply / conversation pickup** counts too: when a thread forms, LinkedIn distributes **beyond** the original author's audience.
- **The like / reaction** is the weakest signal — one click, no content, no strong intent.
- **Dwell time (reading time)** has become a core signal in 2026: the algo measures **how long** you linger on content, not just whether you react. Content that holds attention is judged higher quality and distributed more ([meet-LEA, 2026](https://meet-lea.com/en/blog/linkedin-algorithm-explained)). That's why a **substantive comment** (one that makes people read) is worth more than a hollow phrase.

On the LinkHub side, this shows up in **length**: a **10+ word** comment clearly weighs more than a like, and the optimal zone sits between **15 and 40 words** (see [comment length](/en/blog/longueur-commentaire-linkedin-impressions)). Too short and it reads like a reaction; too long and it dilutes. The full numbers duel is in [comment vs like weight](/en/blog/poids-commentaire-vs-like-algorithme) and [comments vs likes](/en/blog/commentaires-vs-likes-linkedin).

To quickly write a comment that triggers the right signals (long enough, relevant, reply-inviting), LinkHub offers [personalized AI comments](/en/features/ia-commentaires-personnalises) — always validated by you — written in **~29s median** (n = 44,523).

## 3. Why reach concentrates on top creators

The test/amplify mechanism has a side effect: it **favors accounts that already engage**. A creator whose posts quickly trigger comments clears stage 2 more often, so gets amplified more often. Reach concentrates.

Industry numbers confirm it. Per [van der Blom (1.8M posts)](https://www.linkedin.com/posts/richardvanderblom_chapter-1-algorithm-insights-report-2025-activity-7322514599126130688-Q895), the visibility share of **"Top Creators"** rose from **15% in 2022 to 31% in 2025**, while "other creators" fell from **57% to 28%**. In other words: reach is shifting toward the top of the pyramid.

For anyone who **posts**, that's bad news (harder to break through without already being big). But for anyone who **comments**, it's an **opportunity**: commenting under a big creator means **borrowing** an already-concentrated audience — without having to build it ([Botdog, 2025](https://www.botdog.co/blog-posts/linkedin-algorithm-2025)). Our data puts numbers on it: across **4,861 creators** (≥5 comments), the median earns **36 impressions/comment**, a **top 10% 124 (~3.4x)**, a **top 1% 357 (~10x)**, and the observed maximum **1,522 (~42x)**. Same comment, same quality — placed under the right host, it's seen **3 to 40x more** (see [who to comment on](/en/blog/sur-qui-commenter-linkedin)).

Hence the stakes: picking **the right hosts**, those whose audience is your [ICP](/en/blog/icp-commentaire-linkedin). LinkHub automates that choice with its [AI profile recommendation](/en/features/ia-recommandation-profils): it infers your ideal customer, then searches **100,000+ profiles** for the creators with the best ROI for your niche — not just the biggest accounts.

## 4. Does the algorithm penalize AI-generated comments?

It's the honest question to ask, given how the mechanism works. The short answer: **no, not as such.**

The algorithm doesn't "know" whether you wrote it yourself or with AI. What it measures are **signals**: does your comment draw replies? Does it hold attention (dwell time)? Is it relevant to the post? It evaluates **what the content triggers**, not its **authorship**. A generic AI comment that sparks nothing will rank poorly — exactly like a hollow human comment. Conversely, a relevant, engaging comment performs, assisted or not.

So the real risk isn't algorithmic but **human and reputational**: an AI comment that sounds fake, off-topic or robotic hurts your credibility — and that's what costs you relationships, not an "AI penalty." On detection, it's nuanced: the detail is in [are AI comments detectable?](/en/blog/commentaires-ia-detectables-linkedin), and best practices to stay natural (and avoid any account trouble) in [commenting with AI without getting banned](/en/blog/commenter-linkedin-ia-sans-ban).

The healthy rule: use AI to go **faster and better**, not to spam. A [personalized AI comment](/en/features/ia-commentaires-personnalises) properly calibrated to the post and your tone ([authentic tone](/en/blog/commentaires-ia-ton-authentique-linkedin)) ticks every box the algo rewards — relevance, length, ability to trigger a reply.

## 5. What this changes in practice (commenting AND posting)

Once you understand the mechanism, the levers become obvious. Here's what to take from both sides:

**When you comment:**

- **Be in the first half hour.** That's the host post's audience peak — **~3.8x** more impressions than after 24h (see [best hours to comment](/en/blog/meilleures-heures-pour-commenter-linkedin)).
- **Pick the host well.** Reach is concentrated: a top 10% in your niche is worth **~3.4x** the median. It's lever #1 (see [who to comment on](/en/blog/sur-qui-commenter-linkedin)).
- **Write a comment that triggers a strong signal**: 15–40 words, an angle, a question that invites a reply — not a "great post 👏" (see [write a good comment](/en/blog/ecrire-bon-commentaire-linkedin)).

**When you post:**

- **Play the first hour.** Alert your network, reply fast to early comments: you feed stage 2 (the test) and trigger amplification. The post-side detail is in [LinkPost's algorithm playbook](https://www.linkpost.gg/en/playbooks/linkedin-algorithm-playbook-2026).
- **Optimize for dwell time**: a format that holds attention (strong hook, easy read, document/carousel) beats a post skimmed in 2 seconds — see [what makes a post viral](/en/blog/post-viral-linkedin-statistiques).
- **Provoke comments, not likes**: ask questions, take a stance. One comment is worth several likes (see [comment vs post](/en/blog/commentaire-vs-post-linkedin) and [post format](/en/blog/format-post-linkedin-engagement)).

The takeaway: the algorithm rewards **early and substantive engagement**, and concentrates reach on those who generate it. For anyone starting small, **commenting is the shortcut** — you borrow an already-concentrated audience instead of waiting to build your own (see [grow without posting](/en/blog/grossir-linkedin-sans-poster)).

## FAQ

**How does the LinkedIn algorithm work in 2026?**
It tests then amplifies: your content is first shown to a small fraction of the network, engagement in the first 60–90 minutes (likes, comments, and above all reading time) decides whether it gets wider distribution, then amplification can last 48–72h. The first hour is decisive.

**What weighs more: a comment or a like?**
A comment, clearly — estimated ~2x to ~5–15x depending on the source *(third-party, disputed)*. It takes more effort, holds attention and often triggers a reply thread that widens distribution. Detail in [comment vs like weight](/en/blog/poids-commentaire-vs-like-algorithme).

**What is dwell time and why does it matter?**
It's the time a user spends reading/watching your content. In 2026 the algo uses it as a quality indicator: content that holds attention is judged better and distributed more than content skimmed in 2 seconds.

**Does the algorithm penalize comments written with AI?**
Not as such: it evaluates relevance and engagement, not authorship. A generic AI comment that triggers nothing will rank poorly, like a hollow human comment. The real risk is reputational — see [AI comments detectable](/en/blog/commentaires-ia-detectables-linkedin) and [commenting with AI without getting banned](/en/blog/commenter-linkedin-ia-sans-ban).

**Why do big creators capture so much reach?**
Because the test/amplify mechanism favors accounts that already engage. The "Top Creators" share rose from 15% to 31% (van der Blom, 2025). For anyone who comments, it's an opportunity: commenting under them means borrowing an already-concentrated audience.

## Sources & methodology

- **First-party LinkHub data** — impressions per comment (`creator_stats_cache`, 4,861 creators: p50 = 36, p90 = 124, p99 = 357, max = 1,522), timing (261,137 comments: 391 avg impressions at 0–30 min vs 104 at 24h+), optimal length (15–40 words), measured writing time (~29s median, n = 44,523). See [when to comment](/en/blog/quand-commenter-sur-linkedin), [who to comment on](/en/blog/sur-qui-commenter-linkedin), [comment vs like weight](/en/blog/poids-commentaire-vs-like-algorithme).
- [van der Blom — Algorithm InSights Report 2025](https://www.linkedin.com/posts/richardvanderblom_chapter-1-algorithm-insights-report-2025-activity-7322514599126130688-Q895) · [AuthoredUp (2025)](https://authoredup.com/blog/linkedin-algorithm) · [LinkPost — Algorithm 2026 Playbook](https://www.linkpost.gg/en/playbooks/linkedin-algorithm-playbook-2026) · [SocialBee (2026)](https://socialbee.com/blog/linkedin-algorithm/) · [meet-LEA (2026)](https://meet-lea.com/en/blog/linkedin-algorithm-explained) · [Botdog (2025)](https://www.botdog.co/blog-posts/linkedin-algorithm-2025)
- Third-party figures on signal weight (comment vs like) remain industry estimates — flagged as disputed. For a full numbers overview, see the [2026 LinkedIn statistics roundup](/en/blog/statistiques-linkedin-2026).
- Browse all our studies on the [LinkHub blog](/en/blog).