The Update Nobody Checked
LinkedIn never built a "Depth Score." Everyone confidently repeating it as fact just proved the exact problem the myth claims to solve. For months, marketing blogs have circulated a specific, detailed story: LinkedIn quietly launched an AI model called "360Brew" that scores every post for authenticity, and anything falling short on a real-sounding "Depth Score" gets buried in the feed. The story has a name, a mechanism, and a villain. It is also not what LinkedIn's own engineers actually said.
What we found checking the primary source
We checked the primary source directly: LinkedIn's engineering blog, published March 12, 2026. The actual disclosure describes a real rebuild of the platform's ranking system, and a larger one than the "360Brew Depth Score" story ever claimed. A unified semantic retrieval model replaced what had previously been separate ranking pipelines running independently. A new system, described as a Generative Recommender, now ranks posts against a member's ongoing engagement history rather than scoring each individual post in isolation.
Nowhere in that disclosure does the phrase "360Brew" appear. Nowhere does "Depth Score" appear. There is no authenticity-scoring filter described under either name, or any name resembling them, anywhere in LinkedIn's own account of what it built. The specific, quotable version of the story that spread across the industry for months does not match the primary source it was supposedly summarizing.
Why the real update is more consequential, and spread slower
The genuinely striking part of this is not that a fabricated detail spread. Fabricated details spread constantly in fast-moving industries. It is that the real update, the one LinkedIn's engineers actually described, is a more significant change to how the platform works than the invented version, and it spread far more slowly because it did not come with a catchy name attached.
A unified semantic retrieval model replacing separate ranking pipelines is a structural change to how the entire platform decides what to surface. A recommender system that evaluates posts against a member's ongoing engagement history, rather than scoring individual posts on their own, changes the unit of analysis the algorithm is even working with. Both of these are more consequential to understand, for anyone building a content strategy around LinkedIn's ranking behavior, than a fictional authenticity filter ever was. But "Generative Recommender using semantic retrieval" does not spread through marketing Slack channels the way "360Brew Depth Score" does. The real story lost the distribution race to the more memorable fake one.
How a detail like this gets invented
Stories like this rarely start as a fabrication. They typically start as a directionally correct observation, that LinkedIn was rewarding depth and penalizing generic content, mixed with speculation about the mechanism, which someone eventually wrote down as if it were confirmed fact. Once one source states it with specific-sounding detail, a name and a mechanism, later sources cite that source instead of checking LinkedIn's own disclosure, and the fabricated specificity compounds with each retelling. By the time it has circulated for months, the "360Brew Depth Score" story sounds more authoritative than the actual engineering blog post it was supposedly based on, purely because it has been repeated more often, by more sources, each one lending it borrowed credibility rather than independently verified credibility.
This is exactly the failure mode that makes secondhand reporting dangerous in a fast-moving industry. The general direction of a story is rarely the problem. Directional claims are usually at least partially right, which is part of why they spread. The specific, quotable detail, a name, a mechanism, a precise number, is where secondhand reporting quietly invents, because specificity is what makes a story feel authoritative and shareable, and nobody along the chain wants to be the source that only offered the vague, accurate version.
The discipline that catches this before it ships
This is the exact discipline a real signal-sensing process exists to enforce: check the primary source before repeating the specific version of a claim, no matter how many other credible-sounding outlets have already repeated it. The general direction of a story circulating in an industry is worth noting. The specific mechanism, name, or number attached to it needs to trace back to something the original source actually said, not to how many secondary sources have since repeated it with growing confidence.
That discipline is slower than simply repeating what is already circulating. It is also the only thing standing between a content strategy built on real information and one built on a fabricated detail that happened to spread further than the truth. If a content strategy was built on a feature that does not exist, the strategy itself was never wrong. The source it was built on was, and the only way to catch that before it costs anything is to have checked the primary source before repeating the specific claim in the first place.
The cost of skipping that check rarely shows up immediately, which is part of why it keeps happening. A content plan built around "the 360Brew Depth Score" would not have failed on day one. It would have quietly optimized for a mechanism that never existed, producing content shaped around avoiding a penalty LinkedIn never actually built, while missing whatever the real ranking change, the unified semantic retrieval model and the engagement-history-based recommender, was actually rewarding instead. The failure would have surfaced eventually, as unexplained underperformance with no obvious cause, which is a much harder problem to diagnose than simply checking the primary source would have been in the first place.