Google Discover feed now accepts natural language customization requests

Google Discover has historically relied on implicit signals, search history, YouTube views, and site visits, to populate its feed. The new natural language interface lets users override the algorithm with explicit directives, such as requesting deep dives on smart home tech while excluding product announcements. This shift moves control from passive observation to active instruction, but the effectiveness depends on how well the system interprets free-form text.

The immediate update and memory of preferences reduce the friction of iterative refinement. However, the feature does not expose the underlying rules or confidence scores, so users may need to experiment with phrasing to achieve the desired filtering. There is no indication of a feedback loop to confirm whether the system understood the request, which could lead to silent misalignment between intent and output.

While the feature is positioned as a personalization tool, it also serves as a lightweight alternative to blocking publishers or marking content as uninteresting. Engineers who follow niche technical topics may find it useful for surfacing long-form analysis while suppressing press releases. However, the system may struggle with ambiguous or contradictory instructions, and there is no mechanism to prioritize one directive over another if multiple rules conflict.