X has adjusted its recommendation system to give greater weight to mutual followers, the accounts users follow back and who follow them in return. According to product head Nikita Bier, the platform identified that this basic relationship data had not properly influenced how replies appear, resulting in familiar voices getting buried under responses from strangers. The change, described as a modest tweak, seeks to surface posts from these connections more readily and make their replies easier to spot in threaded conversations.
Social platforms have long grappled with the tension between broad engagement metrics and genuine interaction. Early networks built around explicit connections, where seeing updates from people you chose to link with felt intuitive. Over time, however, predictive algorithms prioritizing clicks, replies, and dwell time shifted priorities toward whatever generates measurable activity, often amplifying conflict or novelty at the expense of quieter, existing relationships. X’s latest move circles back toward those original signals, acknowledging that friendship graphs were already present in code released in May 2026 yet somehow underutilized in ranking decisions. The disconnect raises questions about implementation gaps that the company has not fully detailed.
At its core, the adjustment leverages data on reciprocal follows to influence visibility. When mutuals engage, their contributions could stand out more clearly amid the noise of unrelated replies. Bier suggested this might encourage the formation of interest-based groups by making conversations feel less random. Such an outcome would align with what many users recall from earlier social experiences, before feeds became dominated by algorithmic predictions of behavior rather than deliberate choices. Yet the platform’s Grok-powered models, trained to forecast replies and clicks, excel at spotting high-activity content without distinguishing welcome discussion from exhausting arguments. An angry stranger’s response registers as success in the same way a thoughtful reply from a mutual does, highlighting a deeper limitation in how these systems evaluate quality.
Critics might note that surface-level ranking changes address symptoms rather than root issues. Boosting mutuals will not automatically filter out coordinated harassment, outrage-driven bait, or the broader incentive structures that reward polarization. The effectiveness remains uncertain, as X has shared no specifics on the strength of the signal, its interaction with other factors, or metrics for evaluating success. Historical parallels, from Facebook’s repeated experiments with meaningful social interactions to Instagram’s attempts to prioritize close friends, show that algorithmic nudges often deliver incremental shifts at best. Users frequently adapt, or the platform tweaks priorities again in pursuit of growth metrics.
In the wider context of social media’s evolution, this development reflects ongoing pressure to restore some semblance of community amid widespread fatigue. Platforms face scrutiny over mental health impacts, addictive design, and the challenges of moderating scaled conversations. Enhancing mutual visibility represents a pragmatic step toward balancing scale with familiarity, though it underscores how much of the original social promise has been subordinated to engagement optimization. Whether the change noticeably improves daily experience on X will depend on execution details that have yet to emerge. For now, it stands as a reminder that even sophisticated AI systems sometimes require reminders about basic human connections they were ostensibly built to understand.
