Elon Musk told Y Combinator co-founder Paul Graham on X that the platform has not penalized posts containing external links “for over a year,” a confirmation that surfaced inside a reply thread rather than in any formal policy update. The admission touches a workaround, moving a link out of the main post and into the first reply, that has shaped how brands and publishers post on X since Musk’s 2022 acquisition of the company. Whether the underlying mechanism that produced the old suppression has actually changed is harder to verify than the exchange makes it sound.
The thread started when Meta’s Mark Zuckerberg posted a link to his own piece on artificial intelligence. X product chief Nikita Bier replied first: “Hello Mark, you do not need to put the links in replies anymore.” Graham then asked the direct question: “You no longer penalize tweets with links in them?” Musk answered: “We haven’t for over a year.” Developer Wes Bos read the admission as confirmation of the very practice X had long denied: “Anymore?! So you were nerfing links.”
The more substantive part is the evidence PPC Land supplies from its analysis of X’s open-sourced recommendation algorithm, published to GitHub on January 20, 2026. Put a link in a post and the view count came in 94 percent below an otherwise comparable post without one, and nothing in the code targeted URLs at all. The suppression traced instead to Phoenix, the transformer model that ranks posts by predicting engagement across replies, likes, and reposts. Phoenix’s prediction framework does not track external link clicks as an engagement type, so a user who clicks a link and leaves the platform generates no follow-on likes or replies for the model to learn from, and the training data ends up associating linked posts with weaker engagement.
That distinction, emergent statistical suppression instead of a written rule, is the part worth sitting with. Nobody at X coded a line that says “demote links.” A ranking system learned, across billions of interactions, that posts pointing off-platform correlate with lower engagement, and the effect showed up in the output anyway. Most modern ranking penalties work this way now, on Google and elsewhere, which is why debating whether a penalty formally “exists” is frequently the wrong question. The better question is whether the correlation the model learned is still in the training data.
Link-in-reply has been standard practice in content distribution playbooks for years, treated as settled advice by social teams who never had a way to test it against X’s actual ranking logic. If the suppression genuinely eased a year ago, as Musk claims, that workaround has spent twelve months trading real reach for a precaution that stopped paying off.
AI safety researcher Steven Adler’s reply captured the ambiguity better than either side of the exchange: “I believe it’s been said that they were never directly penalized. But there were various indirect negative pressures on them? Which might still exist.” PPC Land’s January analysis measured a code snapshot. Musk’s statement describes an outcome, not a specific change, and neither the exchange nor X’s promised monthly developer notes has pointed to the moment the mechanism shifted.
Treat this as a claim made in a social media exchange, not an official X policy statement. No independent measurement has confirmed that the 94 percent gap has closed, narrowed, or disappeared since January. The only way to know how your account performs under the current system is to run the comparison yourself: publish matched posts with the link in the body against posts with the link in the first reply, hold the rest of the format constant, and compare views over a large enough sample to separate signal from noise. Do that before retiring a habit your team has followed for three years on the strength of one reply-thread admission.
PPC Land, reporting July 30, 2026, on an X exchange dated July 28-29, 2026, and its own January 2026 technical analysis of X’s Phoenix recommendation model.
