Do External Links Always Reduce Reach?

Do External Links Always Reduce Reach?

user avatar

by me1afon

14 minutes ago


AI tools helped gather research and write the first draft of this article. I edited it and checked every figure and quote against the page it comes from, listed at the end. The cover illustration is AI-generated. Details at the end.

On 28 July 2026, X's head of product, Nikita Bier, replied to Mark Zuckerberg: "you do not need to put the links in replies anymore". The next day Paul Graham asked Elon Musk whether X no longer penalised posts with links. Musk answered: "We haven't for over a year." Both exchanges are reported by PPC Land [1].

Yet a year earlier, in August 2025, a Buffer analysis of 18.8 million posts put the engagement rate of link posts from regular X accounts at 0% [5]. In April 2026, Nieman Lab found that the news publishers who link the most on X get the least engagement [6].

So which is it? The statements and the data answer different questions.

No platform examined here documents a general penalty for an ordinary external link. The lower numbers that link posts often show are real, but they are observations, not proof of a rule, and they shrink a lot when posts are compared fairly. This article sets out what the platforms say officially, what the evidence for and against shows, and how to tell a measurement from an anecdote.

This is part 4 of the Platforms & Metrics series. Part 1 looked at the LinkedIn version of this question in detail; here the focus is on X, with LinkedIn, YouTube and Telegram for comparison.

Three different things called a "link penalty"

  • A rule. The ranking system lowers a post because it contains a link. This is what "penalty" means, and what platforms deny.
  • A behaviour. Readers act differently on link posts. Someone who taps a link leaves the post and may never come back to like or reply to it. The post then collects fewer of the signals that ranking systems use, with no rule against links at all.
  • A restriction. The platform limits where links can go or whether they can be tapped, openly, in its documentation.

Each leads to different advice. A rule would justify hiding the link. A behaviour calls for a post that earns a reaction before the reader leaves. A restriction just has to be followed.

What the platforms say officially

X. The clearest denials come from X's leaders. The original posts do not open without signing in, so each quote here comes from a page that reports it.

  • In October 2025, Bier posted "Links are not deboosted", according to roboin.io [2]. The same report paraphrases his explanation: when users view a link, they forget to like or reply to the original post, so the system gets no clear signal of whether the post is good [2].
  • On 13 April 2026 he answered a claim that links are penalised with "Fake news" and advised: "Just post the link in the body of the post and provide sufficient context about what the user will see before they click it" [3].
  • The July 2026 exchange with Zuckerberg and Musk is quoted above [1].

The roboin.io report from October 2025 also quotes a senior staff machine learning engineer at X, Jay Baxter, saying that "organic like rates on posts with links are naturally lower even without any algorithmic penalty" [2]. In other words, X's own staff describe the behaviour effect while denying the rule.

There is a complication. The same October report says the belief in a link penalty came partly from an earlier claim by Musk himself that link posts are deprioritised and that links are better placed in a reply [2]. His July 2026 answer, "We haven't for over a year", says nothing about earlier periods. An author who remembers being told to put links in replies is not imagining it.

LinkedIn. LinkedIn has published no statement on links in its engineering blog. Ordinal, a social media tool vendor, notes that LinkedIn "has never confirmed a deliberate downranking mechanism for outbound links" [8]. The advice from a LinkedIn product director, quoted in part 1 of this series, is that a post should stand alone without the link.

YouTube. YouTube's help page on links is a restriction, not a penalty: "To reduce spam and scam attempts, URLs placed in YouTube Shorts comments and Shorts descriptions are non-clickable." Clickable links in long-form video descriptions require the channel's advanced features [9]. The page does not say that links affect recommendations.

Telegram. Telegram describes channels as a way "to reach people directly, sending a notification to their phones with each post" [10]. There is no ranked feed of channel posts, so there is nothing for a link to be penalised in. Part 3 of this series covers what that means for the tap.

What X's open code shows

X is the only one of these platforms that publishes ranking code. The repository xai-org/x-algorithm describes itself as "the core code that determines which posts a viewer sees in the For You feed on X" [4]. Its README lists the actions the model predicts, and among the clicks it names "link" alongside post, profile and photo clicks [4].

In the current parameter file, read for this article on 16 November 2026 (it carries a sync date of 13 November 2026), the weight for opening a link is positive: OpenLinkWeight is 0.2. For comparison, a like is 0.708 and a reply is 5.0 [4]. No parameter named after a URL appears in the file [4]. Dapp.Expert's own reading of an earlier release found no "has a link" input to the ranking model, and explicit URL rules only for malicious, unsafe or spam links shown to people who do not follow the author [12].

Two limits matter. The repository contains the scoring code, not the trained model. If the model has learned that link posts tend to get fewer likes and replies, that effect would not be visible as a rule. And the code changes: PPC Land's review of the January 2026 release also found "no explicit rule targeting URLs", but wrote that the model "conspicuously omits external link clicks as a tracked engagement type" [1]. The current file includes such a weight. Different versions, different readings.

What the code supports is narrow: there is no visible rule against ordinary links, and opening a link counts for a little, much less than a like or a reply.

The evidence that links get less

  • Buffer, 18.8 million posts from 71,000 X accounts [5]. As of August 2025, link posts from regular accounts had an engagement rate of 0%, against about 0.20% for images, 0.25% for video and 0.40% for text. For Premium accounts, links reached about 0.28% against about 0.90% for text. Buffer writes that link engagement fell "to absolute zero after March 2025" for regular accounts [5].
  • Nieman Lab, the 200 most recent posts from each of 18 news and news-style accounts, April 2026 [6]. The New York Times linked in 88% of its posts, CNN in 90%, the Wall Street Journal in 98%. Fox News linked in only 9%, relied on video and graphics, and had the third-highest median engagement in the sample. The author's conclusion: "Links do seem to hurt news publishers on X/Twitter" [6].
  • PPC Land reports that posts with links received 94% fewer views than comparable posts without them [1]. The page does not show how that comparison was made.
  • On LinkedIn, Ordinal puts the gap at a median of 26.5% and says it grew from 5% in 2023 to 42% in 2025 [8].

Buffer measures engagement, not reach, and does not describe how it separated links from other factors; it calls platform incentives "the most likely explanation", an inference rather than a demonstrated cause [5]. Nieman Lab's own article says "there are lots of factors" [6], and Bier's reply blamed paywalls and caption style rather than links [6]. None of these studies compares the same author's link and non-link posts while holding format constant.

The evidence that the effect is smaller, or elsewhere

When a study does make that comparison, the gap shrinks.

MagicPost, a LinkedIn tool, compared 124,893 link posts from 9,931 authors with the same authors' posts without links in the same quarter, out of 865,980 posts between July 2024 and July 2026 [7]. Without adjustment, link posts got 17% fewer impressions. Comparing posts of the same type and similar "salesiness" brought it to 11% [7]. By format, the gap was 35% when the link preview card was the only visual, and 2% for an address in an ordinary text post, which the authors call "no measurable difference" [7]. MagicPost says plainly: "We don't see how LinkedIn ranks posts, so we don't claim a cause" [7].

Ordinal, whose average figure is larger, also finds the gap concentrated in one place: "company pages absorbing the reach penalty from external links while personal profiles show almost none" [8].

So the largest effects cluster in specific cases. A post that is only a preview card. Accounts that link almost every time and rarely post anything else. Company pages on LinkedIn. In ordinary posts with real content in the text, the measured difference is small or absent.

Why both sides can be right

A lower number has more than one possible cause, and the behaviour explanation fits most of the data without any rule.

A link asks the reader to leave. Some readers tap and go, and the like or reply they might have left never happens. Baxter's remark at X and Bier's October 2025 explanation both describe this [2]. The same logic appears outside social feeds. A Pew Research Center analysis of 900 U.S. adults' browsing in March 2025 found that people who saw an AI summary on Google clicked a traditional result in 8% of visits, against 15% for those who did not [11]. When the answer is already in front of readers, fewer of them click. A social post works the same way in reverse: if it promises everything elsewhere, readers have little reason to react where they are.

Then there is selection. Accounts that post mostly links tend to be publishers with short headline-style captions, often in front of paywalls, and may get less engagement for reasons unrelated to the link. Nieman Lab's sample shows the pattern; it cannot separate the causes [6].

Neither proves that no ranking penalty exists. They show that the data quoted as proof of one does not require it.

How to tell a measurement from an anecdote

Most of what circulates about links is a screenshot or a percentage in a headline. Part 1 of this series traced one such figure, LinkedIn's "60% penalty", and found no dataset behind it. A few questions sort the rest:

  1. Who measured, and how many posts? One account's experience is a hypothesis, not a finding.
  2. Compared with what? The useful comparison is the same author, the same format, the same period. Comparing link posts with all other posts mixes in everything else.
  3. Which number? Impressions, engagement rate and clicks are different things. A post can reach fewer people and still send more readers to the article.
  4. Official or paraphrased? A platform's own page, a vendor's study and a quote from a post nobody opened are three different levels of evidence.
  5. Which period? X's code and its leaders' statements have changed over two years. Advice from 2024 may describe a system that no longer exists.

What this means in practice

Moving every link into a reply is not supported by the evidence. On X, the code by default filters replies from accounts the reader does not follow out of the For You feed [4], so a link in a reply reaches new people mainly through the parent post. On LinkedIn, MagicPost found that a link in the first comment did no better than a link in the post [7].

What the evidence does support is the same thing parts 1 and 3 arrived at from different directions: give the post enough content that it is worth reacting to before the reader leaves. Put the link in the body with a plain line explaining what is behind it, as Bier advised [3]. Do not let the preview card be the only thing in the post.

The most reliable answer for your own account is your own data: ten to twenty pairs of comparable posts, one with the link in the body and one with it in a reply or left out, compared on impressions and, above all, on visits to the article from tagged links. A small sample will not settle the question for everyone, only for your audience, which is the question that matters to you.

A short check before you decide

  • Is the claim about a rule, a behaviour or a restriction?
  • Does the study compare the same author's posts in the same format?
  • Is it measuring reach, engagement or clicks to your site?
  • Was the platform's statement read on its own page, or in a paraphrase?
  • Are you measuring visits to the article, not only impressions?

What this article cannot tell you

No platform publishes the trained model that ranks its feed, so a learned effect against links cannot be ruled out from code. X's statements were read only through secondary reports, because the original posts do not open without signing in; the dates and wording come from PPC Land and roboin.io. Musk's earlier claim about deprioritised links is known here only from roboin.io's description, without the original wording or date. PPC Land's 94% figure is reported without a visible method. A dataset of 53,568 X posts by the vendor Xboost, cited in the research notes, could not be checked because its page shows no figures, so it is not used. None of the studies measures how many readers a post sends to an article. Nothing here reports results from Dapp.Expert's own posts; those have not been measured in a way that allows a claim.

How this article was made

Task What the AI did What a person did How it was checked
Research Collected platform statements, code notes and link studies into research notes Chose the question and the sources to rely on Figures and quotes checked against the pages listed below [1]–[11]; secondary quotes are marked as such; [12] is unpublished research
Code Read the parameter file of X's public repository Chose which weights to report Values read from the file on 16 November 2026
Draft Wrote the first draft from the notes Edited the text and structure A separate review compared figures and claims with the sources
Images Generated the cover photograph-style illustration Chose the illustration; designed the explanatory diagrams Labelled as AI-generated in the disclosure line and alt text
Not done by AI — The decision to publish and responsibility for the text —

Conflicts of interest: none. The author has no commercial relationship with X, LinkedIn, YouTube, Telegram or the companies whose studies are cited.

Responsible for this article: Dmitriy Lastovka, author. Corrections: contact form.

Sources

Labels: official (the platform's own page or code), study (a published analysis, read on its own page), secondary (a quote or figure reported by another page).

  1. Luis Rijo, PPC Land, on X dropping its link penalty (30 July 2026), quoting posts by Nikita Bier (28 July 2026) and Elon Musk (29 July 2026), and reviewing X's January 2026 code release. https://ppc.land/x-drops-year-old-link-penalty-musk-tells-zuckerberg-on-platform/ Secondary for the quotes; study for the code review.
  2. roboin.io, "Is 'posts with links are penalized' on X a myth? Head of Product denies it" (20 October 2025), quoting Nikita Bier and Jay Baxter. https://roboin.io/article/en/2025/10/20/x-product-chief-denies-rumor-that-link-posts-are-deboosted/ Secondary.
  3. roboin.io, on X denying link deboosting (13 April 2026), quoting Nikita Bier's post https://x.com/nikitabier/status/2043428612108644848 (not opened). https://roboin.io/article/en/2026/04/13/x-denies-link-deboost-recommends-in-post-links/ Secondary.
  4. xai-org, "x-algorithm", README and home-mixer/params/param.rs (file sync date 13 November 2026, read 16 November 2026). https://github.com/xai-org/x-algorithm Official.
  5. Tamilore Oladipo, Buffer, "Links on X" (14 October 2025). https://buffer.com/resources/links-on-x/ Study.
  6. Laura Hazard Owen, Nieman Lab, "Do links hurt news publishers on Twitter? Our analysis suggests yes" (8 April 2026). https://www.niemanlab.org/2026/04/do-links-hurt-news-publishers-on-twitter-our-analysis-suggests-yes/ Study.
  7. Naïlé Titah, MagicPost, "Do Links Really Hurt LinkedIn Reach? What 865,980 Posts Say" (29 October 2026). https://magicpost.in/blog/linkedin-external-links-reach Study.
  8. Ordinal, "LinkedIn Link Penalties: Does Posting with Links Hurt Reach?" (undated; read 16 November 2026). https://www.tryordinal.com/blog/linkedin-link-penalty-study Study.
  9. YouTube Help, sharing links in videos and Shorts. https://support.google.com/youtube/answer/13748639 Official.
  10. Telegram, "Telegram Channels" (feature tour). https://telegram.org/tour/channels Official.
  11. Athena Chapekis and Anna Lieb, Pew Research Center, "Google users are less likely to click on links when an AI summary appears in the results" (22 July 2025). https://www.pewresearch.org/short-reads/2025/07/22/google-users-are-less-likely-to-click-on-links-when-an-ai-summary-appears-in-the-results/ Study.
  12. Dapp.Expert research notes on X's ranking code and on platform adapters (5 November 2026), reading the repository release of 13 August 2026. Author's research; not a published source.

Tier I

Sector: #18291

Sealed Hiding Place Room

Resource Cache

Resource Cache

Tier I

Requires 25% Tier Progress to Claim
Meme Cache

Meme Cache

Tier I

Requires 50% Tier Progress to Claim
Equipment Cache

Equipment Cache

Tier I

Requires 75% Tier Progress to Claim

After collecting, hiding places will be stored in your inventory and can be opened with Keys.

Other articles

Giving Telegram Readers a Reason to Open the Full Article

A Telegram channel post reaches subscribers without a ranked feed, so the hard part is the tap, not the reach. What Telegram documents, and how to earn the tap.

user avatarme1afon

One Article, Two X Accounts, Two Different Jobs

A personal X account and a publication's X account can carry the same article, but they should not say the same thing. Here is how to give each its own job.

user avatarme1afon

Turning an Article Into a LinkedIn Post

A LinkedIn post that only points to an article gives readers nothing to stay for. What LinkedIn measures, what link data shows, and a post that stands alone.

user avatarme1afon

Comparing Creators Without Rewarding Only Incumbents

A ranking by totals mostly measures who started first. Why early leads persist, and five corrections to compare creators fairly without punishing leaders.

user avatarme1afon

Making an Expert Ranking Explainable

A ranking of experts is a claim about real people. Four parts make it checkable: a published formula, stated weights, a per-entry breakdown and an appeal path.

user avatarme1afon

Audience Size Is Not Topic Authority

A large following records popularity, not knowledge of a subject. Why reach and authority need separate evidence, and five questions to ask about any expert.

user avatarme1afon

Important disclaimer: The information presented on the Dapp.Expert portal is intended solely for informational purposes and does not constitute an investment recommendation or a guide to action in the field of cryptocurrencies. The Dapp.Expert team is not responsible for any potential losses or missed profits associated with the use of materials published on the site. Before making investment decisions in cryptocurrencies, we recommend consulting a qualified financial advisor.