X’s New Grok-Powered Feeds Finally Solve Your Timeline Chaos

X’s New Grok-Powered Feeds Finally Solve Your Timeline Chaos
Key Takeaways

  • X launched an update to its Custom Feeds feature this week, integrating the Grok model to automate list building and real-time content filtering through natural language.
  • The system replaces manual keyword tagging with semantic understanding, allowing feeds to curate content based on the intent of a conversation rather than just hashtags or specific phrases.
  • Unlike static lists, these feeds use real-time reasoning to prioritise high-signal accounts while suppressing low-effort engagement farming and repetitive bot activity.

X’s new Grok-powered Custom Feeds let Premium subscribers describe what they want to see in plain language and have the feed built for them automatically. For anyone who has spent years manually curating Lists only to watch them go stale, that is a meaningful shift. It puts curation logic in the user’s hands rather than inside a black-box algorithm.

Instant Curation via Natural Language Prompts

The core of the update is a creation sidebar where, instead of hunting for accounts or wrestling with muted-word filters, you simply describe what you want. Enter a prompt, the system suggests a title and icon, and the feed appears as a swipeable tab next to “For You” and “Following.” Feeds populate within seconds.

Because they are dynamic, new accounts posting relevant content get surfaced automatically without a manual follow. That active discovery model is a genuine departure from how social platforms have worked: users are no longer locked into their social graph or a recommendation engine they cannot steer. The manual process of building a quality List has always been labour-intensive: find reliable accounts, verify activity, prune constantly as topics drift. This automates most of that. In testing, a prompt for “Real-time updates on decentralised finance security audits” bypassed generic crypto-price noise and surfaced technical reports from security researchers directly.

Semantic Filtering vs. The Keyword Trap

The classic problem with keyword-based feeds is ambiguity. Search for “Apple” and you get fruit recipes alongside product reviews. X’s system uses semantic filtering, reading the meaning of a post rather than just matching characters. In testing, a feed set to “Professional photography gear” correctly ignored posts using a camera emoji to flag a party and focused instead on lens reviews and sensor specs.

Grok processes posts in real-time as they are indexed, assigning a relevance score based on the original prompt. A “Refine” button lets users tell the feed to show more of something or exclude a sub-topic, creating a feedback loop that improves accuracy over time. Allowing users to weight verified researchers and long-form posts over high-velocity short-form content would further improve the platform’s signal-to-noise ratio.

Navigating the Algorithmic Bias of Grok

Any AI-curated experience carries the bias of its underlying model because Grok is trained on a specific slice of internet data and X‘s own posts, the feeds it builds reflect those training weights. In testing with “Global trade policy,” the feed showed a tendency to surface accounts that post frequently, producing a loudest-voice effect rather than a most-expert-voice one.

X has faced criticism before for how its algorithms handle political and social content. These custom feeds shift some responsibility to the user, since the prompt sets the direction. But if you ask for “Unbiased news,” the model still has to make a subjective call about what unbiased looks like. In practice, prompts requiring high objectivity tended to produce feeds that were safe but repetitive, leaning on major outlets rather than independent analysts.

There is also a visibility question. Posts from accounts flagged by X’s safety systems are absent from these feeds even when relevant to the prompt. That keeps feeds cleaner, but it means the system still operates inside X’s moderation guardrails. Anyone expecting a truly unfiltered view of the platform will find these feeds are a curated experience by design. For teams building pipelines that need to monitor broad social signals without platform filtering, that constraint matters. For a deeper look at how guardrails shape what AI systems surface, see our piece on keeping LLMs in check.

Reclaiming Control Over Information Density

For power users, the practical win is controlling information density. The standard “For You” timeline is optimised for time-on-app, which means controversy and emotional charge. A well-prompted custom feed can function as a professional tool instead. Journalists, researchers and domain specialists can build a “Work Mode” tab that surfaces only field-relevant content, effectively muting the broader platform during working hours.

X is also letting users share their prompts publicly. Publishing a “Feed Recipe” lets others subscribe to a curation logic rather than a static account list. That could create a small ecosystem of curators who do not produce content themselves but are skilled at writing prompts that filter information well. If someone has precisely tuned a feed for “Quantum Computing Developments,” others can follow that logic and inherit the refinement work.

Grok-powered custom feeds are a solid step toward X’s everything-app ambitions. The semantic filtering is genuinely better than keyword matching, the interface is fast and the sharing mechanic has real potential. The engagement-farming and algorithmic bias problems are real limitations, not minor footnotes, and they will take more than a signal-to-noise toggle to fix. But for enterprise users who have long felt the main timeline was optimised for everyone except them, this is a tool worth building with. Stay up to date with the latest AI developments at Auton AI News.

Alex Chen
Alex Chen

Alex covers AI tools, apps, and consumer technology for Auton AI News. With a focus on making AI accessible, Alex helps everyday readers understand and use the latest AI developments.

📰 Journalists welcome — cite Auton AI News with attribution. Press & Media → | press@autonainews.com