When clients and publishers ask whether links are “punished” on X, I do not give them a blanket yes or no. External links can still be a reach disadvantage, but the 2026 evidence does not support the simple claim that every outbound URL receives the same fixed algorithmic penalty.
When I audit link performance on X, I separate what the platform actually documents from what marketers repeat as folklore. X’s public For You system changed substantially in 2026. The current open-source recommendation system uses a Grok-based transformer to predict actions such as likes, replies, reposts, clicks, shares and dwell time, then combines those predictions into a relevance score. The public documentation does not identify “contains an external link” as a standalone negative scoring rule.
That does not mean link posts perform equally well. A large 2025 Buffer analysis found that link posts underperformed text, image and video posts across its dataset. That is useful evidence of a performance gap, but it does not prove that a hidden link penalty caused the gap.
My practical advice to publishers, creators and businesses is straightforward: use a link when the click is the goal, but judge that decision against your own account data. Do not assume links are harmless, and do not remove them automatically because of an old platform myth.
X does not publicly document a universal external-link penalty in its 2026 recommendation system, yet link posts can still earn less native engagement. I would keep the link when referral traffic is the objective and measure the business result rather than treating impressions as the only score that matters.
What changed in X’s recommendation system?
X published a major update to its open-source For You algorithm in May 2026. The system now describes a pipeline that retrieves posts from accounts a user follows and from out-of-network candidates, then ranks them with a Grok-based transformer.
Original editorial diagram based on X/xAI’s May 2026 open-source For You recommendation documentation.
The model predicts the probability of multiple actions, including likes, replies, reposts, clicks, profile clicks, shares, dwell time and negative actions such as blocks, mutes and reports. A weighted scorer combines those predictions into the final relevance score.
This matters because the current public architecture is different from the older idea of a feed controlled mainly by a collection of hand-written rules. X says it has removed every hand-engineered feature and most heuristics from the system.
What the public code does not establish is just as important: its published scoring overview does not list a universal “external URL” penalty.
So has X stopped demoting links?
This is where I would push back on both extremes. There is no strong basis for saying X has officially ended link demotion across the platform, but there is also no strong basis for claiming that the current public ranking system applies a fixed penalty to every post containing an external URL.
Those are different claims.
The public algorithm documentation tells us how the current recommendation architecture works at a high level, but X also notes that its released Phoenix model is representative rather than identical to production in every respect. Production models are larger and continuously trained.
That means the open-source code is valuable evidence, not a complete audit of every production behavior.
Meanwhile, X’s own safety documentation confirms that some links can receive reduced visibility for a different reason. URLs classified as potentially malicious or harmful may be blocked, receive warning notices, or have limited visibility. That is a safety and integrity mechanism, not evidence that ordinary outbound links are universally penalized.
Why link posts can still perform worse
In practice, I do not need a hard-coded “link penalty” to explain why a link post may lose reach.
If ranking is driven by predicted engagement, I expect different formats to produce different signals. A native video may generate viewing time, a strong text post may earn replies, and an image may prompt expansion or sharing. A link post asks the reader to leave the feed and visit another destination, so it is competing under a different behavioural pattern from the start.
The result can be a performance gap even without a simple URL penalty.
Buffer’s October 2025 analysis covered 18.8 million X posts from 71,000 accounts. It reported that link posts produced lower engagement than text, image and video posts in its dataset, including a particularly weak result among regular accounts.
Original chart based on Buffer’s October 2025 analysis of 18.8 million X posts from 71,000 accounts; observational evidence, not causal proof.
That is substantial observational evidence, but it should not be turned into a causal claim. Accounts choose which content gets links, audiences differ, Premium status matters, and platform behavior changes. The study shows that link posts underperformed in that sample; it does not isolate the algorithm as the sole cause.
Observational studies can show that link posts underperform without proving why. Treat platform-wide benchmarks as a starting point, then test your own audience and account type.
Reach and referral traffic are different goals
One of the most important points I make to clients is that a post can succeed without maximising impressions.
For a publisher, 10,000 impressions that produce qualified readers may be more valuable than 100,000 impressions with almost no site visits. For a software company, a product-page click can matter more than likes. For an ecommerce brand, revenue from a smaller audience may beat a high-reach post that keeps everyone on X.
This is why “Do links hurt reach?” is only half the question.
The more useful question is: does a link post produce enough valuable off-platform action to justify any reduction in native engagement or distribution?
When the answer is yes, removing the link to chase impressions can make the post worse for the business.
Should you put the link in the post or a reply?
I do not use a universal rule here.
Putting the link in the main post gives readers the shortest path to the destination and makes the call to action clear. That is often the better choice when traffic, registrations, downloads or purchases are the objective.
Putting the link in a reply keeps the main post focused on native engagement, but it adds friction. Readers must notice the reply, open it and then click. It can also make measurement and user experience less straightforward.
Treat the reply-link tactic as a format worth testing, not as an algorithm loophole.
A better way to test link performance on your account
My testing rule is simple: never compare one viral text post with one weak link post. Build a small repeated test instead.
- Choose one content theme and one audience objective.
- Publish several link posts and several comparable non-link posts over multiple weeks.
- Avoid changing too many variables at once. Keep topic, posting window and creative quality reasonably similar.
- Measure impressions and engagement inside X.
- Measure outbound sessions, conversions or revenue on your site.
- Compare medians rather than letting one unusually strong post decide the result.
I am not trying to reverse-engineer every internal ranking decision. I am trying to discover which format produces the outcome the account actually needs.
In a link-post audit, I pair X metrics with site analytics. Impressions and engagement show what happened on the platform; outbound sessions, conversions and revenue show whether the post actually did its job. Optimising for reach alone can make a commercially useful post look like a failure.
What to measure
For awareness-focused posts, I track impressions, engagement rate, follows and meaningful replies.
For traffic-focused posts, track outbound clicks and landing-page sessions alongside impressions. For commercial posts, continue through to sign-ups, leads, purchases or another business outcome.
A lower-reach link post can still win if it sends better traffic.
Also separate results by account type and format where possible. The Buffer data showed meaningful differences between regular and Premium accounts, which is another reason broad platform rules can mislead individual publishers.
When external links deserve extra caution
X can restrict links it identifies as unsafe, deceptive, spammy or associated with platform manipulation. Its help documentation says warning-notice links can also receive limited visibility.
I use the direct destination whenever possible and avoid deceptive redirects, misleading shorteners or any link practice that obscures where the reader will land.
If X flags a legitimate domain, its help documentation provides a route to report a potentially miscategorized link.
Do not treat reply links, URL shorteners, or other posting tricks as guaranteed reach workarounds. They can add friction, complicate measurement, or trigger safety systems if the destination looks deceptive.
What publishers and brands should do in 2026
I keep external links when the destination is integral to the post’s purpose. A news article, research report, product page, registration form or download often needs a direct route.
Make the post useful before the click. Give readers enough context to understand why the destination matters rather than publishing a bare URL.
Test native and link-led formats separately. A native post can maximize conversation while a link-led post can optimize referral traffic. They do not need to perform the same job.
Evaluate business outcomes, not impressions alone. Reach is an input. The value of the visitor after the click may be the outcome that matters.
Finally, I recheck the assumption rather than turning one test into permanent doctrine. X’s recommendation system is actively evolving, and the May 2026 open-source update demonstrates how materially the architecture can change.
My recommendation on links
External links on X can still underperform native formats, but I would not tell a client that “X automatically demotes every link.” That conclusion is stronger than the evidence available in 2026.
The public recommendation architecture centres on predicted user actions rather than exposing a simple universal link penalty. Independent performance data still shows a meaningful gap for link posts, while X separately limits some URLs for safety and integrity reasons. Those mechanisms should not be confused.
For publishers and businesses, the best strategy is not to hide every link. Use links when clicks matter, make the post valuable on its own, measure both on-platform reach and off-platform outcomes, and keep testing as the platform changes. If you're also deciding between X and Threads, see our Threads vs X (Twitter) in 2026 comparison.
