Google Discover Feb 2026 Core Update Scorecard: Data Shows What Actually Changed

In early February 2026, Google began rolling out a Discover-focused core update for English-language users in the United States. In its messaging, Google highlighted three intended outcomes:

1. More locally relevant content, with increased prominence for sites based in the user’s country
2. Less sensational and clickbait-style content in Discover
3. More in-depth, original, timely content from sites with demonstrated topic expertise, evaluated topic-by-topic

This is not a “Discover redesign”, it’s an underlying scoring change that affects what gets distributed, how strongly, and to whom. Because the rollout overlaps with an active news cycle, these early findings should be treated as directional, not final.

Methodology

We compared pre-update (Jan 25–31, 2026) vs post-update (Feb 8–14, 2026) using NewzDash DiscoverPulse (Google Discover Real-Time Tracking) panel data across the following datasets:

• US Top 1000 domains
• US Top 1000 articles (pre vs post exports)
• California Top 1000 articles (pre vs post exports)
• New York Top 1000 articles (pre vs post exports)

Important note on comparability

Google Discover Visibility Score is a relative representation of clicks within each selected time window:

• The top item in each list receives 100 points
• All other items are scaled relative to that top item
• For US Top 1000 domains, we use normalized scores so pre and post are comparable on the same scale

Panel-data caveat

Our dataset reflects Discover content surfaced to a large, diverse panel of millions of real US users. It is a strong signal for detecting distribution shifts over time, but it is still a sample. The right use is to identify changes in patterns and composition, not to claim absolute totals for the full Discover ecosystem.

TL;DR: The Scorecard

Key takeaway: These goals do not behave like three separate switches. The data reads more like a unified scoring change where regional relevance, topic authority, and content utility and timeliness reinforce one another.

Goal 1: More locally relevant content from websites based in the user’s country

Google’s stated goal here is country-level relevance, specifically increasing visibility for sites based in the user’s country.

To evaluate this claim, we looked at it from two angles:

• Country-level “domesticity” (directly aligned to Google’s goal): Are US-based publishers taking a larger share of Discover visibility versus non-US publishers?
• Regional personalization (an additional test we chose to run): Google did not explicitly mention state-level personalization in its announcement, but we evaluated it anyway to see whether Discover feeds meaningfully differ between California, New York, and the US national view in ways that reflect local relevance.

We evaluate both using (1) the US Top 1000 domains normalized dataset and (2) the Top 1000 articles exports for the US, California, and New York.

1) Domain-level “domesticity” (US Top 1000 domains, normalized)

To test Google’s “based in the user’s country” intent, we bucket domains into four groups using NewzDash classification rules:

• Platforms: YouTube, X, and similar distribution platforms
• International: known non-US publishers plus strong non-US signals
• Global/unclear: ambiguous country signals
• US: conservative default bucket

Result (score-weighted share, normalized): the international share of total normalized score declines from 8.52% → 7.04%, while the US share rises from 88.86% → 89.94%.

Examples of international-leaning domains that decline in Discover Visibility during the post-update window include:

• theguardian.com (−11%)
• reuters.com (−20%)
• independent.co.uk (−57%)
• the-sun.com (normalized score 150 → 50, −67%)
• and additional long-tail international publishers (for example: ndtv.com, elfinanciero.com.mx)

At the same time, several large US publishers increase visibility during the same window (for example: cbsnews.com, nbcnews.com, axios.com, apnews.com, cnbc.com, usatoday.com, forbes.com).

Important nuance: there are exceptions that require careful interpretation. For example, bbc.com rises sharply in the post window. Rather than over-interpret this as “Google treating BBC as US,” the safer read is simply: some global publishers can still win visibility when their content matches high-utility, high-interest topics for US users (or when their US edition signals are strong).

Another nuance: platform share in the normalized domains view increases slightly (2.03% → 2.63%). So, while the “more domestic websites” signal is strong in the domains mix, it does not imply that platforms are being broadly suppressed at the domain level.

2) What users saw in feeds (US, California, New York article exports)

At the article level, the feeds remain platform-heavy, but platform visibility declines modestly across all three Top 1000 article lists. Here, “platform share” is defined as YouTube + X score-weighted share in each Top 1000 export:

• US Top 1000 articles: 63.16% → 62.68%
• California Top 1000 articles: 58.43% → 56.53%
• New York Top 1000 articles: 59.34% → 56.34%

In other words: the algorithm remains highly social/platform-influenced in the Top 1000 articles, but there is a consistent directional step down across US, CA, and NY in the post window.

Note: The top 1000 articles lists also suggest that within “websites only” (excluding platform domains), international visibility trends down across the three geographies. This conclusion depends on the same domain bucketing used in the domains analysis, so it should be interpreted as directional.

3) Regional personalization (state-level local relevance)

Google’s stated goal is country-level relevance, but the most visible behavior change in the article exports is sub-national personalization. When we compare California, New York, and the US national Top 1000 articles lists, the three feeds share a large common core, but each state shows a meaningful “local layer” of content that differs from the national feed.

California: local content increases in the post window

Post-update, the California feed contains California-specific stories that do not appear in the US national Top 100 during the same window, including examples from publishers such as SFGate, LA Times, Sacramento Bee, ABC7, EdSource, Farm Progress, and SF Chronicle. This is the clearest “local relevance” pattern in the regional data.

New York: local layer persists, but with a different pattern

The New York feed’s “local layer” is visible through a different set of local and metro-area items (for example: NJ.com, Newsday, North Jersey, plus local weather and retail updates appearing via YouTube). The specific local items differ substantially from California, which is exactly what you would expect if geographic personalization is functioning.

The strongest proof: cross-regional distribution differences

When we compare local-domain presence across feeds, the difference is stark:

NY-local domains appear roughly 5x more in the NY feed than in the CA feed, and CA-local domains appear roughly 5x more in the CA feed than in the NY feed. That is unambiguous evidence of a meaningful state-level personalization layer in Discover.

At the same time, the feeds still share most of their Top 100 items, meaning this is not “three totally different feeds.” It is better described as: a common national core plus a meaningful regional layer.

Goal 1 verdict

Strong directional support. The normalized US domains mix supports a shift toward domestic visibility share during the post window. The state-level article exports show a clear personalization layer, most pronounced in California, and clearly differentiated between California and New York.

Goal 2: Reducing sensational content and clickbait

This is the hardest goal to validate using spreadsheets alone. “Clickbait” is often driven by thumbnails, promise vs delivery, and on-page experience, not just a few headline words.

So we used two spreadsheet-friendly proxies:

• Conservative headline markers: obvious terms and patterns (for example: “shocking”, “secret”, “revealed”, “you won’t believe”), excessive punctuation, and similar signals.
• Title structure signals: average title length, question-based titles, and the mix of editorial articles vs social/platform items.

1) Conservative headline-marker check (directional, not definitive)

Headline markers remain rare across all feeds, and the count rate stays broadly flat. However, the score-weighted share of flagged items rises in every region, suggesting that a small number of higher-visibility items can heavily influence the score distribution.

• US: flagged titles 0.4% → 0.3%, flagged score share 0.19% → 0.40%
• California: flagged titles 0.5% → 0.4%, flagged score share 0.29% → 0.38%
• New York: flagged titles 0.4% → 0.5%, flagged score share 0.23% → 0.43%

What this means: the marker method does not prove a broad “clickbait reduction.” At best, it suggests that blunt, obvious clickbait terms are not increasing, but a few high-performing items can still carry engagement-style framing.

2) Title structure and content mix (US Top 1000 articles)

We also measured how titles and content types shift in the US Top 1000 articles. This is not a clickbait detector, but it does indicate whether Discover is leaning toward more descriptive headlines and more editorial content.

Interpretation: titles become longer and editorial volume rises modestly. These are consistent with a feed that is less dependent on short, curiosity-gap framing, even though engagement-style packaging still exists.

3) What is actually being suppressed: templated curiosity-gap patterns

While the headline-marker method is limited, the data does show clear demotion of specific templated patterns, especially when those patterns come from sites that rely heavily on them.

Yahoo: fewer top-tier placements

In our US Top 1000 export, Yahoo’s presence shrinks from 11 → 6 articles. More notably, Yahoo goes from having multiple items in the Top 100 pre-update to zero items in the Top 100 post-update. That is consistent with curiosity-gap distribution losing ground at the very top of the feed.

Autoevolution: template-style clickbait disappears

Autoevolution had 5 articles in the pre-update US Top 1000, all following a near-identical “dramatic reveal” formula. Post-update, Autoevolution has 0 articles in the US Top 1000. This is one of the cleanest examples of Discover potentially identifying and suppressing repetitive sensational templates.

Geediting: listicle-style framing loses rank

A pre-update Geediting listicle that ranked very high in the feed drops materially in the post-update window (in our exports, it falls from roughly #14 to roughly #153). This supports a broader pattern: listicle-driven “psychology says…” style content appears less favored in the post window.

Important nuance: engagement-optimized titles do not disappear entirely. Some brands with clear topical authority still rank well with playful or curiosity-driven framing. The change appears less like “no engagement allowed” and more like “no shallow templates at scale.”

Goal 2 verdict

Directional, not proven. The spreadsheet proxies do not allow us to definitively prove that clickbait decreased across Discover. What we can say with confidence is:

• Obvious clickbait markers remain rare and do not surge in count, but flagged items can still command a higher score share.
• Titles become more descriptive on average, and editorial volume rises modestly.
• Several high-reliance, templated curiosity-gap patterns are visibly demoted in the post window.

Goal 3: More in-depth, original, timely content from topic-expert websites, evaluated topic-by-topic

This goal bundles three things that are difficult to measure cleanly from spreadsheets alone:

• Timeliness: does the feed tilt toward high-utility, current coverage?
• Topic expertise: are specialist publishers gaining visibility within their niches?
• Depth and originality: are the promoted items actually original or in-depth?

From the provided exports, we can measure timeliness and topic mix shifts strongly, we can measure publisher concentration vs breadth clearly, and we can point to directional signals of topic expertise. We cannot directly confirm “original” or “in-depth” without content-level review.

1) The clearest visible shift: more “news utility”, less entertainment gravity

Across the US, California, and New York, the score-weighted category mix shifts toward News and Sports and away from Arts and Entertainment. The Sports uplift is influenced by a major live-sports cycle (Super Bowl, Winter Olympics, and World Cup coverage).

US (score-weighted share)

• News: 15.94% → 19.16% (up)
• Sports: 5.06% → 8.52% (up)
• Arts & Entertainment: 24.40% → 17.90% (down)

California (score-weighted share)

• News: 16.00% → 18.71% (up)
• Sports: 5.51% → 8.49% (up)
• Arts & Entertainment: 21.51% → 15.71% (down)

New York (score-weighted share)

• News: 17.39% → 19.52% (up)
• Sports: 5.57% → 8.10% (up)
• Arts & Entertainment: 21.62% → 15.29% (down)

In the US Top 100 articles, the same story shows up as a visible “top-of-feed” shift: entertainment-heavy categories lose prime placement, while news-forward categories occupy more of the top positions.

What this supports: The data strongly supports the “timely” and “utility” portion of Goal 3. Discover is leaning more toward news-forward consumption during the post window, and less toward entertainment gravity at the very top of the feed.

2) Topic breadth increases, while publisher concentration also increases in some markets

Rather than treating “equal opportunity” as mixed, the exports tell a clear pattern:

• Discover shows more distinct topics (more unique categories),
• but it often does so using fewer publishers (fewer unique domains) in the US and California.

Diversity and concentration signals (Top 1000 articles)

• US Top 1000 articles: unique domains 172 → 158 (more consolidated), Top 10 score share 75.31% → 76.65% (more concentrated)
• California Top 1000 articles: unique domains 187 → 177 (more consolidated), Top 50 score share 86.95% → 88.00% (more concentrated)
• New York Top 1000 articles: unique domains 192 → 194 (slightly broader), Top 10 score share 70.57% → 69.71% (less concentrated), Top 50 score share 87.82% → 86.30% (less concentrated)

Category breadth (unique categories)

• US: 163 → 173 (up)
• California: 162 → 170 (up)
• New York: 168 → 183 (up)

Interpretation: Google appears to be expanding topical reach while applying stricter publisher selection in at least two of the three views (US and CA). New York behaves somewhat differently, showing slightly more publisher diversity post-update, but the category-breadth expansion is consistent everywhere.

3) The platform timeliness layer: X.com grows as a high-velocity news surface

We covered X.com's Discover rise back in November 2025, and this update appears to have accelerated that trend significantly. Most of the X articles ranking in the top 100 come from institutional accounts — NYT, news outlets, political figures. Google may be treating tweets from authoritative accounts as "timely, original content" under Goal 3.

At the article level, X increases its presence near the top of the feed in the post window:

• US Top 100 articles: X.com items rise from 3 → 13
• NY Top 100 articles: X.com items rise from 2 → 14

In the normalized US domains view, X’s Discover Visibility rises sharply (+38%). At the same time, YouTube remains the dominant platform in absolute Discover volume, generating roughly 5x the videos and clicks of X.com in our panel feeds. 

X grows materially in top placements and becomes a more prominent “timely updates” surface inside Discover during the post window.

Many of the very top-performing X items are from established media brands, but across all X items in the Top 1000, it’s a mixed set.

Measuring Content Shelf life from the US Top 1000 articles dataset, X.com posts and editorial articles show a similar shelf life in the post-update window. Post-update, X items average ~48.5 hours in-feed (median 49 hours) versus ~48.7 hours for editorial articles (median 44 hours). 

Does posting on X diminish the chance your article appears in Discover?

From the current top-1000 articles data alone, we cannot prove or disprove cannibalization. That said, we can do a directional sanity check in the US exports:

• X.com items increase in top placements in the post window (for example, US Top 100: 3 → 13).
• For at least one major newsroom brand, nytimes.com appears less often as a site domain in the US Top 1000 post window versus pre (US Top 1000: 8 → 5 items; score sum 116 → 43), while an NYT X post is the #1 item in the post window.

Interpretation: this pattern is consistent with the possibility of “surface substitution” (some distribution shifting from site URLs to X posts), but it could also be driven by the news cycle, editorial mix, or other scoring changes. It is not proof of cannibalization.

4) Topic expertise

We do see directional signs consistent with “topic-by-topic” selection: more category breadth, fewer generalist placements at the top in some markets, and specialist publishers appearing more frequently within relevant categories.

Discover is expanding into more niches, and the publisher set that benefits most appears to be the one that combines timeliness, utility, and clear topical focus, rather than broad, entertainment-driven general coverage.

Goal 3 verdict

Strong support for timeliness and topic-mix shift. The category composition and top-of-feed behavior move toward news utility and live-event coverage. Topic breadth increases (more unique categories), while publisher concentration increases in the US and California, with New York showing a modest counter-pattern. Originality and depth are not testable from these exports alone and require content-level review.

Open questions and anomalies (kept tight and evidence-based)

• Paywalled publishers: Some paywalled brands decline in the post window.
• Rollout timing: Google’s public dashboard messaging indicates the rollout may take up to about two weeks. This analysis reflects an early window, so patterns may shift as rollout completes.
• News cycle confound: The post window includes major ongoing stories and live sports, which can independently increase News and Sports visibility. 

What This Means for Publishers: 5 Takeaways

1) Build topic authority, not “general coverage” breadth

The post-update exports point to a clear direction: Discover is expanding into more categories while becoming more selective about which publishers get the most distribution in many markets. The strongest winners are not always the biggest brands, they are often publishers that combine clear topical focus, utility, and fresh coverage in the categories they own.

This aligns with Google’s “topic-by-topic” framing: a site can be broad overall, but still be evaluated as an expert within specific topic areas.

2) Engagement is still allowed, templated “curiosity gap at scale” is not

Our spreadsheet proxies do not let us “prove clickbait is gone.” What we can observe is that several templated curiosity-gap patterns lose visibility in the post window. For example:

• Autoevolution goes from having multiple templated items in the US Top 1000 to zero.
• Yahoo shrinks from 11 → 6 items in the US Top 1000 and goes from having items in the Top 100 to zero in the post window.
• Geediting shows a sharp drop for a high-ranking listicle item in the post window.

The practical takeaway: you can still write compelling headlines, but relying on repeatable “mystery click” templates is increasingly risky unless the content is backed by clear authority and value.

3) Local publishers should treat this as a real distribution opportunity

Goal 1 shows the strongest, most provable behavior change: state-level feeds (California and New York) include a meaningful “local layer” that differs from the national feed. If you are a local or regional publisher, your advantage is structural: you can produce the content that is most relevant to a specific geography and community, and Discover appears to be reinforcing that relevance layer.

The best strategy is not “be local” only, it is: be locally relevant and topically useful in the areas where you have real coverage depth.

4) Your X presence may now be part of your Discover surface area

X is not replacing YouTube, and platforms still dominate a large share of the Top 1000 article visibility. However, X increases its presence near the top of the feed in the post window (for example, X items rise from 3 → 13 in the US Top 100, and 2 → 14 in the NY Top 100).

Publishers should monitor whether their X posts are appearing in Discover and treat X as a complementary “timely updates” surface: short, fast, high-velocity distribution that can coexist with long-form editorial coverage.

One caution for publishers: if a Discover card sends the user to Discover → X → your site, that creates an extra step versus Discover → your site. That added friction will almost certainly reduce click-through to your owned pages for a share of users, even if the story still “wins” distribution.

If X continues to grow as a Discover surface, publishers should treat it as both a visibility channel and a potential leak in referral efficiency. Two practical next steps:

• Measure conversion drop: compare sessions and downstream conversions for Discover traffic that lands directly on your site versus Discover traffic that lands on X first (needs referral chain or link tracking).
• Test monetization on X: if more discovery happens on X first, explore monetization formats that work natively on X, especially video and short, high-velocity updates that can be monetized without relying on the second click.

5) Treat the update as one system, not three separate checklists

In practice, “local relevance,” “reduced templated clickbait,” and “topic-by-topic expertise” behave like a single scoring model. The publishers best positioned to benefit are the ones that combine:

• Topical authority (clear expertise and coverage depth)
• Timeliness and utility (fresh, useful, informative content)
• Geographic relevance where applicable (state and metro relevance)

Final Take

After analyzing the US Top 1000 domains, we get a clearer directional read on the “based in their country” claim: the international share declines on a normalized, apples-to-apples basis (based on our domain bucketing method).

At the article level, we also see consistent shifts across US, California, and New York:

• Platform share (YouTube + X, score-weighted) declines modestly across all three Top 1000 article exports.
• Category mix shifts toward News and Sports and away from Arts and Entertainment in the post window.
• Regional personalization is clearly present: California and New York feeds share a national core, but each contains a meaningful local layer.

Two important items remain unresolved or require richer data to validate:

• Clickbait reduction: headline markers alone cannot prove “clickbait is reduced,” even though templated curiosity-gap patterns appear to lose distribution.
• Publisher opportunity vs consolidation: category breadth increases in all regions, but publisher diversity declines in the US and California while New York shows a modest counter-pattern. The net effect looks like: more topics from a more curated set of publishers in many markets.