Personalization skepticism on X
Ivan Lee questioned whether typical personalization approaches in news apps learn effectively from listening versus explicit preferences, suggesting many systems fail to adapt correctly to audio or passive signals. The post raises doubts about current personalization models' ability to reflect actual user tastes. (x.com)
A debate on X over news-app personalization zeroed in on a basic problem: apps may be reading clicks and settings, but missing what people actually choose to hear. (x.com) News personalization systems usually sort stories by signals such as past reading, similar users’ behavior, popularity, and any preferences a user states directly. A 2023 research review said those systems are now “ubiquitous” in digital media and shape both what publishers distribute and what audiences see. (academic.oup.com) The dispute in this case centers on explicit feedback versus implicit feedback. Explicit feedback is what a user says they want, like following a topic; implicit feedback is what a system infers from actions such as clicking, scrolling, or listening. (academic.oup.com) That distinction has become more important as publishers push beyond article ranking into format changes such as summaries, translations, and text-to-audio. Nieman Lab reported on June 24, 2025 that generative artificial intelligence is expanding personalization from story selection into the way the same story is delivered. (niemanlab.org) Publishers are investing in those features even as audience demand remains mixed. Reuters Institute said in its 2025 newsroom-leaders survey that 75% were exploring text-to-audio, 70% summaries, 65% translation, and 56% chatbots for audience-facing products. (reutersinstitute.politics.ox.ac.uk) Audience surveys point in a different direction. Reuters Institute’s Digital News Report 2025 found interest highest for summaries at 27% and translations at 24%, while customized homepages, recommendations, and alerts each drew 21%; text-to-audio ranked among the lowest-interest options. (reutersinstitute.politics.ox.ac.uk; niemanlab.org) Researchers are also questioning whether the old explicit-versus-implicit split is too simple. A 2025 study based on interviews with 34 active users of algorithm-driven social platforms said people use “intentional implicit feedback” too, taking actions they believe will steer recommendations without using formal preference controls. (arxiv.org) That finding lines up with the skepticism in the X post: if a user listens to audio versions, skips others, or samples stories passively, the system still has to decide whether that behavior signals taste, convenience, or just habit. The harder part is not collecting more signals, but interpreting which ones reflect a stable preference. (arxiv.org; academic.oup.com) Reuters Institute said its 2025 report was based on survey data across 48 countries, collected as publishers face declining engagement, low trust, and more news avoidance. In that setting, personalization is being sold both as a product feature and as a way to keep people from drifting away. (reutersinstitute.politics.ox.ac.uk; niemanlab.org) The argument on X does not settle whether personalization works. It does sharpen the question publishers keep running into: whether a news app can tell the difference between what a user consumes and what that user actually wants more of. (x.com; academic.oup.com)
