🆕 We've published a comprehensive article (we've been working on it for weeks) on @1492_Vision about SDF: Google Discover removes outdated articles before even ranking them. A Google paper for the CIKM 2026 conference describes this system, SDF, as "fully deployed in Google Discover." We compared it to our Discover data 👇 1492.vision/research/discove… What the paper describes: 🔁 Filter 1: An article is removed when a newer article on the same topic replaces it, like a pre-game preview once the result is known. ⏳ Filter 2: A model predicts the proportion of its lifetime traffic that an article has already received. At 90%, the article is removed. 📉 Only one filter is needed, and the article never makes it to ranking. In two years, "outdated" reports have decreased by 54.9%. What we see in our data: 🏀 Some still get through: pre-game content for the Michigan-UConn final was served more than 15 hours after the game ended. 📅 The articles displayed are rarely more than two days old, and their audience drops off after three days. Regarding scores: 🎯 A score called topic potential (from 0 to 1) seems to estimate the engagement an article can expect, across all readers. It depends more on the topic than the publisher and says nothing about quality. 🚪 It acts as an entry threshold: below approximately 0.10 to 0.12, an article was rarely amplified before July. Predicted engagement scores, specific to each reader, then determine the level of amplification. 📊 During the week of June 30th, several Discover scores changed, and the potential score dropped by about a quarter until October. This coincides with the beginning of the decline in reach noted in our logbook. 🧭 SDF was already running at the beginning of 2026: on its own, it doesn't fully explain this decline. It sheds light on its form: the articles retain their best day but lose their end-of-life. CC @gaganghotra_ @VorticonCmdr @nicola_agius @ClaraSoteras #googlediscover