Google Discover: Tailor Your Feed for niches

With "Tailor Your Feed" — now known as "Add topics to your feed" — Google opens an explicit control layer for the Discover feed for the first time. Users can type in natural language what content they want to see. For publishers, this means a potential third visibility path alongside implicit interest affinity and the Follow button. Especially notable: a minority of served cards comes from a query-intent fan-out that, according to analysis, favors niche websites and smaller creators — often content that had barely circulated in Discover before.

The feature launched through Search Labs in the United States and has been systematically observed since December 2025. Analysts tracked UI states, server responses, attribution tags, and feed behavior after each "Refresh / Update your feed." In parallel, each card was mapped to its underlying pipeline. This makes it possible to trace how historicalnaturallanguagetuningcontent.f differs from the rest of the feed — based on tracking data from 1492.vision. The analysis focuses on distribution outcomes, not absolute reach numbers.

Explicit control instead of pure behavioral inference

For years, Discover personalized implicitly: clicks, dwell time, and follows shaped the experience. "Tailor Your Feed" reverses that principle. At the top of the feed appears the question "What do you want to see?" with the input field "Add topics to your feed." A chat panel opens; users pick suggestions or write freely — for example "Show me more women's basketball" or "Keep me updated on country music."

On the server side, inputs are translated into actions: SEE_MORE, KEEP_UPDATED, CREATOR_MORE, and SEE_LESS. Only after confirmation with "Refresh / Update your feed" do the instructions take effect. Behind the interface, a large language model with a persistent chat thread apparently applies prompts to the feed in real time and over longer periods.

The historicalnaturallanguagetuningcontent.f pipeline

Technically, the historical variant historicalnaturallanguagetuningcontent.f delivers most cards from this control layer — as a counterpart to naturallanguagetuningcontent.f. In spring 2026, the interface was renamed and the chat entry point was placed more prominently. For SEO and publisher teams, the label matters less than pipeline behavior: it selects content differently from classic Discover channels that re-serve already popular articles.

Two mechanisms: entity expansion and fan-out

Content is selected in two ways. The majority follows entity or interest expansion: known topics and entities are broadened. The minority mechanism is query-intent fan-out — described in the Discover context as a GEO-like retrieval path. Here the system interprets the user request like a search intent and casts a wider net of sources.

This fan-out explains observations such as vegan recipe blogs, regional media like Mississippi Today, LinkedIn posts, or niche blogs on the Japanese property market in the feed. VentureBeat also appeared on a "niche sites" prompt — as evidence of retrieval behavior, not as a small publisher itself.

Transparency through visible attribution

Google makes the origin recognizable: the "You asked to see" label, the "resulting from natural language tuning" tag, and a prompt history in My Activity connect explicit user intent with individual cards. For publishers, this is a rare glimpse into otherwise opaque personalization — and a signal that user wishes flow more directly into distribution.

Popularity bypass and opportunities for small sites

The central finding: this pipeline mostly carries content that had barely or never circulated in Discover before. Classic pipelines favor already successful articles. The fan-out acts like a popularity bypass — visibility arises through explicit demand rather than historical reach. Selection power shifts more toward the user and opens its own channel for niche offerings.

Methodology and data interpretation

The analysis combines field research in the Google app with pipeline mapping of each feed card. "No prior Discover distribution" means no earlier serving trace appears in the tracking dataset — not necessarily missing overall reach. Examples are shown anonymized as prompt-to-result. Internal mechanisms are interpretations of observed data and public research.

Current rollout and limitations

As of the observations: the feature remains English-only, runs through US Search Labs, and is still young in adoption. In France, availability was close to zero percent. Publishers outside the United States cannot address the feature directly yet but should understand the logic: explicit user intent, fan-out retrieval, and moving away from pure popularity filters will change which content reaches Discover over time.

• Explicit prompts complement implicit personalization in Discover.
• The fan-out path can surface niche content without prior Discover history.
• Attribution labels make natural language control visible in the feed.
• US Search Labs limit the immediate international effect.
• Publishers with clear topic niches benefit through the third visibility path.

Anyone using Discover as a traffic channel should prepare content to match precise user requests: clear topics, strong images, recognizable entities, and editorial context instead of pure broad reach. "Tailor Your Feed" is still young — but the fan-out shows Google is testing an additional path for smaller and specialized publishers.