CLICK ON ANY FINDING TO EXPAND
"What to know" gets Featured 1.8× more often—but organic views trend lower (padj=0.10, not significant at α=0.05, n=16).
→ Use "What to know" when targeting Featured specifically. Don't apply it broadly.
Details ↓News brands and celebrity/entertainment content operate as two distinct notification ecosystems with 2.8× different baseline CTRs and non-overlapping formula signals. For news: "EXCLUSIVE:" (—) and attribution language (1.33×) drive CTR. For celebrity content: named person + possessive (1.17×).
→ Segment notification strategy by content type. Don't apply celebrity-content heuristics to hard news, or vice versa.
Details ↓Sports is #2 on Apple News and #8 (last) on SmartNews—the largest topic-rank gap by platform (6 rank positions).
→ Write platform-specific sports briefs. Don't reuse Apple News sports content on SmartNews.
Details ↓On Apple News, high-view and high-engagement articles are largely separate populations (r=-0.003)—a headline optimized for reach won't advance depth, and vice versa. These are two distinct editorial strategies, not one.
→ Define success separately for reach campaigns vs. depth campaigns. Don't optimize headline variants for views and expect active time to follow.
Details ↓Number leads climbed from 0.68× (Q1 2025) to 1.05× (Q1 2026)—the only formula to cross into above-baseline territory. Question format dropped from 0.66× to 0.47×.
→ Lean into number leads; deprioritize question-format headlines. Re-check quarterly as 2026 data accumulates.
Details ↓Featured articles reach 47% non-subscriber audiences. Non-featured articles reach only 3% non-subscribers—even among the top-quartile most-viewed non-featured stories (5%).
→ Treat featuring as audience acquisition, not just a traffic bump. Without it, Apple News delivers your content almost exclusively to existing subscribers.
Details ↓Weather articles get featured at 41%—28× the Sports rate (1.4%). Shopping and Opinion are never featured. Within Business, situation/event stories are featured 2.0× more than individual-person stories. Question format lifts featuring probability 2.0× across all sections.
→ Choose topic before formula when targeting Local News featuring. Weather and Business are the highest-leverage sections; Sports and Shopping essentially opt you out.
Details ↓Finding 1 · Featured on Apple News
Among the 973 Featured articles in our dataset, "What to know" headlines are dramatically overrepresented: 13 of 29 (45%) were Featured, versus 25.1% overall. This is a statistically robust formula signal (χ²=5.1, padj=0.03 *).
Question-format headlines are also Featured more often than expected (44%, 1.75× lift, padj<0.001 ***)—but they significantly underperform other Featured articles once selected. Apple's editors favor questions; the format itself doesn't follow through on views.
Quoted ledes present the inverse pattern: Featured at roughly the baseline rate (34%), but once Featured they deliver among the highest within-Featured percentiles. Questions get into the Featured tier and stall; quoted ledes get in and overperform.
Causal note: The association between "What to know" and Featured placement is observational. The causal direction is ambiguous: editors may independently choose the same stories that writers frame as "What to know," rather than the format itself driving featuring.
| Formula | n | Featured rate | Lift | padj (BH–FDR) | Within-Featured median %ile |
|---|---|---|---|---|---|
| ★ What to know | 29 | 45% | 1.79× | padj=0.03 * | 81% (1.18× Featured avg) |
| ★ Question | 178 | 44% | 1.75× | padj<0.001 *** | 44% (0.64× Featured avg) |
| ★ Here's / Here are | 28 | 39% | 1.57× | padj=0.15 | 54% (0.79× Featured avg) |
| Quoted lede | 178 | 34% | 1.37× | padj=0.008 ** | 86% (1.26× Featured avg) |
| Possessive named entity | 124 | 23% | 0.93× | padj=0.74 | 63% (0.93× Featured avg) |
| ↓ Number lead | 219 | 15% | 0.60× | padj=0.002 ** | 77% (1.13× Featured avg) |
Read this table as: "Featured lift" is how much more often Apple selects this formula for Featured. A high rate means Apple's algorithm rewards it—not that it organically outperforms.
Featured placement: reach vs. reading depth
Featured articles average 50 seconds of active reading time versus 56 seconds for non-Featured. The difference is statistically significant (Mann-Whitney p<0.0001). Apple's editorial promotion drives discovery; readers who find an article because the algorithm surfaced it are slightly less engaged than readers who sought it out.
All 3,884 Apple News articles (2025–2026). Chi-square test: each formula vs. all other articles combined. BH–FDR across all 7 formula tests. Causal direction of "What to know" → Featured is unconfirmed.
Finding 2 · Push Notifications
Across 1783 Apple News push notifications (Jan 2025–Feb 2026), the dataset contains two functionally different content types. Us Weekly (entertainment/celebrity) runs at 4.01% median CTR; the four news brand outlets (Miami Herald, KC Star, Charlotte Observer, Sacramento Bee) run at 1.41%. The formula signals that predict CTR are almost entirely non-overlapping between the two populations.
News brand signals (n=1036, median CTR 1.41%)
| Feature | n (present) | Median CTR (present) | Median CTR (absent) | Lift (95% CI) | Effect size r | padj (BH–FDR) |
|---|---|---|---|---|---|---|
| Attribution (says/told) | 97 | 1.83% | 1.38% | 1.33× [1.20×–1.46×] | -0.32 | padj<0.001 *** |
| Short (≤80 chars) | 758 | 1.45% | 1.36% | 1.06× [0.99×–1.18×] | -0.03 | padj=0.58 |
| Named person + possessive | 128 | 1.38% | 1.42% | 0.97× [0.85×–1.04×] | 0.07 | padj=0.29 |
| Contains number | 248 | 1.40% | 1.44% | 0.97× [0.87×–1.03×] | 0.02 | padj=0.68 |
| Full name present | 526 | 1.39% | 1.46% | 0.96× [0.88×–1.02×] | 0.05 | padj=0.29 |
| Question format | 87 | 1.25% | 1.43% | 0.87× [0.75×–0.99×] | 0.14 | padj=0.08 |
For hard news notifications: "EXCLUSIVE:" earns clicks when the story justifies it. Attribution language ("says"/"told"/"reports") signals source credibility and is associated with higher CTR. Question format consistently hurts. Notification length shows no significant effect when analyzed within this population.
The serial/escalating story as a content type
The top news brand notifications by CTR are dominated by a single story: Nancy Guthrie's disappearance and its connection to Savannah Guthrie. This defines a content type: the serial/escalating story with a celebrity anchor. The formula: possessive named entity + new development + escalating stakes, published in installments. The structural recipe: "[Celebrity]'s [family member/associate] [new disclosure/development]."
Us Weekly / celebrity signals (n=747, median CTR 4.01%)
| Feature | n (present) | Median CTR (present) | Median CTR (absent) | Lift (95% CI) | Effect size r | padj (BH–FDR) |
|---|---|---|---|---|---|---|
| Full name present | 599 | 4.27% | 2.92% | 1.46× [1.34×–1.56×] | -0.48 | padj<0.001 *** |
| Named person + possessive | 279 | 4.37% | 3.74% | 1.17× [1.11×–1.29×] | -0.23 | padj<0.001 *** |
| Short (≤80 chars) | 384 | 4.13% | 3.89% | 1.06× [0.99×–1.12×] | -0.06 | padj=0.23 |
| Attribution (says/told) | 38 | 4.17% | 3.99% | 1.05× [0.99×–1.12×] | -0.03 | padj=0.84 |
| 'Exclusive' tag | 64 | 4.00% | 4.01% | 1.00× [0.86×–1.11×] | 0.01 | padj=0.89 |
| Question format | 13 | 3.81% | 4.01% | 0.95× [0.53×–1.07×] | 0.31 | padj=0.10 |
| ↓ Contains number | 243 | 3.21% | 4.32% | 0.74× [0.68×–0.81×] | 0.38 | padj<0.001 *** |
For celebrity/entertainment notifications: named person + possessive ("Smith's…") is the dominant signal. Numbers in the headline hurt CTR—specific counts and statistics feel out of place in celebrity context. "EXCLUSIVE" shows no significant effect, suggesting the word has different valence in entertainment contexts. Notification length shows no significant effect within this population.
News brand CTR trend (post-activation, H2 2025–Q1 2026)
News brand Apple News notifications went live in June 2025. CTR has declined quarter-over-quarter since activation—from 1.77% in Q2 2025 to 1.26% by Q4 2025—with a partial stabilization at 1.37% in Q1 2026. This pattern is consistent with audience accommodation to a new push channel. Monitoring this trend as the channel matures is important for interpreting future CTR benchmarks.
Apple News push notifications Jan 2025–Feb 2026 (n=1783 with valid CTR; 2025 includes Us Weekly all year, news brands from June 2025 only). Analyses run separately within each brand-type population; BH–FDR correction applied within each set of tests. Mann-Whitney U; effect size = rank-biserial r; 95% CIs via 1,000-iteration bootstrap. Feature classifier unvalidated.
Finding 3 · Platform Topic Inversion
Sports ranks #2 on Apple News (percentile index 1.10× platform median) but #8—last—on SmartNews (index 0.78×). This is not a small difference: sports sits well above the Apple News median and well below the SmartNews median. The inversion is statistically significant (Mann-Whitney U, p<0.001 ***) across the full year of 2025 data.
Nature/wildlife shows the reverse: it underperforms the Apple News median (0.72×) but outperforms the SmartNews median (1.23×). Among the top 30 most frequent words in top-quartile headlines on each platform, only 8 words appear on both lists (cops, death, found, photos, say, shares, star, years)—generic reporting terms rather than shared topical vocabulary, suggesting the platforms reward very different content angles.
MSN shows a third distinct ranking (orange bars). Sports scores 0.00× the MSN platform median—below average, consistent with the SmartNews pattern. Politics indexes at 0.00× on MSN—near the platform median.
Sports subtopic performance by platform
Within the sports inversion: which sports specifically drive Apple News performance, and which are weakest on SmartNews? The table below breaks sports into subtopics (via two-level headline classifier).
| Sport | Apple News n | Apple News median %ile | SmartNews n | SmartNews median %ile |
|---|---|---|---|---|
| college | 4 | 67% | 37 | 45% |
| hockey | 8 | 64% | 17 | 30% |
| football | 160 | 60% | 540 | 49% |
| baseball | 5 | 54% | 34 | 39% |
| sports_other | 79 | 50% | 1404 | 35% |
| basketball | 20 | 32% | 77 | 35% |
| soccer | 0 | — | 6 | 29% |
Politics subtopic performance by platform
Within politics, which story type drives the most engagement on each platform?
| Subtopic | Apple News n | Apple News median %ile | SmartNews n | SmartNews median %ile |
|---|---|---|---|---|
| Federal | 0 | — | 0 | — |
| State | 0 | — | 0 | — |
| Local Govt | 0 | — | 0 | — |
| Election | 0 | — | 0 | — |
| Politics Other | 0 | — | 0 | — |
Top quartile politics headlines
Bottom quartile politics headlines
Topic tagged via unvalidated regex classifier applied to headline text. Coverage: 50% of Apple News articles match a named topic; 50% fall into "other/unclassified" and are excluded from this analysis. Results describe the classified minority—generalizing to all content requires caution. Percentile index = median percentile_within_cohort / platform overall median percentile. Apple News 2025–2026 (n=3,884); SmartNews 2025 (n=37,695); MSN 2025 (n=0). Subtopic classifier unvalidated. No significance testing—treat as descriptive. Subtopics with n<3 show "—".
Finding 4 · Views vs. Reading Depth
The Apple News dataset includes both Total Views and average active time per article. Pearson r = -0.003 (p = 0.86), Spearman ρ = -0.033 (p = 0.04). Both agree: views and reading time show near-zero correlation across 3,884 articles—high-reach and deep-read articles are largely distinct populations. The view count spans a 537× range across deciles; active time moves only 5 seconds.
| Metric | Correlation with Total Views | What it measures |
|---|---|---|
| Avg. active time | r = -0.003, ρ = -0.033 (not significant) | Depth of the current read |
| Saves | r = 0.81 (strong) | Intent to return / bookmark behavior |
| Likes | r = 0.78 (strong) | Affirmation / social signal |
| Article shares | r = 0.66 (strong) | Distribution / word of mouth |
Featured articles illustrate this split directly: 2.9× median view lift, but only 50s active time vs. 56s for non-Featured (p<0.0001). Apple's editorial promotion drives discovery; readers who arrive via Featured spend slightly less time than readers who actively sought the article. Separately, subscribers average 48s active time vs. 55s for non-subscribers—likely a usage behavior difference (subscribers browse more, read less per article) rather than a content quality gap.
Apple News 2025–2026 (n=3,884 articles with valid active time). 29 articles have active time <10s; 1 have >300s—not filtered, ~1% of records. Spearman ρ is the preferred test for independence given skewed views distribution.
Finding 5 · Trends Over Time
Chart shows the three highest-volume formulas (≥15 articles/quarter). Here's/Here are and What to know are in the table below—their n=4–9/quarter is too small to distinguish trend from noise.
Lift by formula across periods
Each cell shows the median lift for that formula in that period, relative to unclassified articles published in the same period. A dash means fewer than 5 articles qualified.
| Formula | Q1 2025 | Q2 2025 | Q3 2025 | Q4 2025 | Q1 2026 | |||||
|---|---|---|---|---|---|---|---|---|---|---|
| Number lead | 0.68× | (n=38) | 0.75× | (n=72) | 0.87× | (n=41) | 0.88× | (n=21) | 1.05× | (n=46) |
| Question | 0.66× | (n=45) | 0.64× | (n=20) | 0.96× | (n=31) | 0.66× | (n=28) | 0.47× | (n=53) |
| Possessive named entity | 1.32× | (n=20) | 1.25× | (n=16) | 1.10× | (n=29) | 1.23× | (n=22) | 0.92× | (n=36) |
| Here's / Here are | 1.23× | (n=9) | — | (n=—) | 1.41× | (n=7) | 0.87× | (n=8) | 0.12× | (n=4) |
| What to know | 1.56× | (n=9) | 0.42× | (n=4) | 1.52× | (n=4) | 0.98× | (n=4) | 0.42× | (n=8) |
Quarters: Q1=Jan–Mar, Q2=Apr–Jun, Q3=Jul–Sep, Q4=Oct–Dec. Q1 2026 = Jan–Feb 2026 only. Lift = formula median cohort percentile ÷ untagged baseline median within same quarter. Minimum 3 articles required per cell. Data through March 2026.
Finding 6 · Featuring Reaches Non-Subscribers
Among 10,929 news articles published in 2026 across 5 publications (Charlotte, KC, Miami, Raleigh, Sacramento), the subscriber composition of article audiences splits sharply by featuring status:
| Article type | Median subscriber share | Non-subscriber reach |
|---|---|---|
| Featured in Local News (821 articles) | 53% | 47% |
| Non-featured (all) | 98% | 3% |
| Non-featured (top-quartile views only) | 95% | 5% |
The top-quartile comparison is the most important row: even non-featured articles that are performing exceptionally well still reach only 5% non-subscribers. High organic traffic does not substitute for featuring when it comes to audience composition. The only path to non-subscriber audiences at scale is a Featured placement.
What this means operationally: Subscriber-audience articles still have value—subscribers are your core, highest-intent readers. But if the goal is discovery, growth, or reaching readers who haven't yet subscribed, featured placement is the lever. The editorial strategies that earn featuring (see Finding 7) are therefore also the strategies that expand your audience beyond the existing subscriber base.
Apple News Publisher data, Jan–Feb 2026. News publications only (Charlotte Observer, KC Star, Miami Herald, News & Observer, Sacramento Bee). Politics excluded per team policy. Minimum 10 views per article. Subscriber/non-subscriber split from Apple News Publisher "Unique Viewers, Subscribers" and "Unique Viewers, Non-subscribers" columns.
Finding 7 · Topic Predicts Featuring—Formula Doesn't
Featuring rate by section (2026 articles, news publications)
| Section | Articles | Featured rate | Median view percentile |
|---|---|---|---|
| Weather | 164 | 40.9% | 0.89 |
| Business | 196 | 21.9% | 0.79 |
| Crime | 94 | 7.4% | 0.21 |
| Sports | 2,556 | 1.4% | 0.37 |
| Shopping | 1,388 | 0.0% | 0.45 |
| Opinion | 562 | 0.0% | 0.44 |
Within-section signals
Business section—individual-person stories penalized: Business articles that do not mention a named person are featured at 30% vs. 15% for articles that do—a 2.0× difference. Stories about market conditions, economic trends, or institutional changes are favored over profiles and personnel stories. This holds across local city sections too (Charlotte, KC Metro, Sacramento, Miami) at 1.4–1.8× magnitude.
Question format lifts featuring 2.0× across all sections: 14.3% of featured articles use a question-format headline vs. 7.2% of non-featured. The pattern in featured question headlines is consistent: a hook statement paired with a community-concern follow-up ("How cold will it get?", "What's next?", "Will they fix it?"). These are service-journalism signals, not clickbait—Apple's editors appear to favor questions that answer a community need.
Weather is the highest-leverage section by a large margin. Nearly every major weather event that receives traffic also receives a Featured placement. This is partly because weather is inherently local, timely, and directly useful to readers—all qualities Apple's Local News curation rewards. The implication is that weather coverage quality has outsized editorial ROI relative to its production cost.
What headline formula can and can't do: After controlling for section, headline formula effects on featuring are small relative to topic choice. You cannot formula-optimize your way into a Featured placement from the Sports or Shopping sections. Choose the topic first, then optimize the headline within that topic.
Apple News Publisher data, Jan–Feb 2026. 10,929 articles across 5 news publications. Politics excluded. Featuring = "Featured in Local News" placements (Apple News Publisher). Named-person detection via regex (two consecutive capitalized words). Business section n=196 articles.