Tony Wang17 min readGoogle's Trending Now Board Barely Repeats Itself: A 26-Day Churn Study (2026)
Google Trends real-time board, 26 daily snapshots: ~96% of everything that ever cracks the top 50 shows up on exactly one sampled day, then never returns.
Google's Trending Now board is the opposite of a stable ranking: it's a live feed of whatever's spiking right now, US-wide, updated continuously. We wanted to know something a single glance at the board can't tell you: if you check it once a day, every day, for a month, how much of it is actually the same list? We pulled the top 100 items once a day for 26 sampled days and tracked every query's presence, rank, and recurrence across the window. The short answer: almost none of it repeats, and the little that does repeat tells you something real about which stories have staying power.
The board resets almost completely, every single day
Take any two consecutive sampled days and compare their top 25. On average, 99.4% of the second day's list is brand-new: queries that were nowhere in the previous day's top 25. The same is true for the top 50. This holds steady across the whole window, not just on a couple of unusually volatile days:
View the data: day-over-day % new (top 25)
| Sampled day | % new vs. prior sampled day |
|---|---|
| 7/23 | 100% |
| 7/25 | 100% |
| 7/28 | 100% |
| 7/30 | 96% |
| 8/1 | 100% |
| 8/3 | 100% |
| 8/5 | 100% |
| 8/7 | 100% |
| 8/9 | 100% |
| 8/11 | 100% |
| 8/13 | 100% |
| 8/16 | 100% |
| 8/18 | 100% |
Zoom out from day-over-day comparisons to the whole 26-day window and the picture holds: of every unique query that ever cracked a given board size, the overwhelming majority appeared on exactly one sampled day and then vanished from the dataset entirely.
| Board size | Unique queries seen | Appeared on exactly 1 of 26 days | Appeared on 2+ of 26 days |
|---|---|---|---|
| Top 25 | 625 | 608 (97.3%) | 17 (2.7%) |
| Top 50 | 1,247 | 1,205 (96.6%) | 42 (3.4%) |
| Top 100 (full board) | 2,431 | 2,294 (94.4%) | 137 (5.6%) |
And when a query does come back, it's almost never the next day. Looking at consecutive-day runs specifically (how many sampled days in a row a query holds a spot before dropping out), the longest streak in the entire dataset is 2 sampled days for the top 25 and top 50, and 3 for the full top 100:
| Board size | 1-day runs | 2-day runs | 3-day runs | Longest run found |
|---|---|---|---|---|
| Top 25 | 642 (99.4%) | 4 (0.6%) | 0 | 2 sampled days |
| Top 50 | 1,286 (99.5%) | 7 (0.5%) | 0 | 2 sampled days |
| Top 100 | 2,548 (99.1%) | 20 (0.8%) | 4 (0.2%) | 3 sampled days |
When something does come back, it's a legible short list
The 42 queries that cracked the top 50 more than once are not random noise. They cluster around exactly the kinds of stories you'd expect to have a longer shelf life: a team on a playoff run, a recurring scheduled event, or a story that keeps generating new developments.
| Query | Sampled days present (of 26) | First seen | Last seen |
|---|---|---|---|
| Dodgers | 5 | 2026-07-30 | 2026-08-12 |
| Powerball numbers | 4 | 2026-07-27 | 2026-08-03 |
| Alexandra Eala | 4 | 2026-07-29 | 2026-08-10 |
| Dow Jones industrial average | 3 | 2026-07-25 | 2026-08-07 |
| Padres | 3 | 2026-07-29 | 2026-08-13 |
| Astros | 3 | 2026-07-30 | 2026-08-17 |
| Ben Shelton | 3 | 2026-08-01 | 2026-08-13 |
| Lindsay Clancy trial | 2 | 2026-07-31 | 2026-08-10 |
| Supreme Court | 2 | 2026-07-29 | 2026-08-17 |
| Jamal Adams | 2 | 2026-07-28 | 2026-08-16 |
That spread ("Dodgers" showing up on 5 separate days across a 2-week window rather than 5 days in a row) is itself informative: MLB playoff races and legal cases don't trend continuously, they trend in bursts tied to games, hearings, and news cycles, and our once-a-day sample happens to catch some of those bursts and miss others.
The higher a story peaks, the more likely it comes back
We also tested whether a story's best-ever rank predicts whether it resurfaces later in the window. It does, though the effect is modest, not dramatic:
View the data: recurrence rate by peak rank
| Best rank ever reached | Reappeared on a later sampled day | Sample size |
|---|---|---|
| #1-10 | 8.8% | 250 queries |
| #11-25 | 6.1% | 375 queries |
| #26-50 | 6.6% | 622 queries |
| #51-100 | 4.3% | 1,184 queries |
Read plainly: stories big enough to crack the top 10 are about twice as likely to show up again later as stories that never made it past the bottom half of the board, consistent with genuinely major stories (playoff runs, market-moving days, unfolding legal cases) generating fresh spikes over multiple days, while most of the board is one-off news that spikes once and is gone.
What's actually on the board: mostly names, not themes
We clustered every sampled item into five broad categories using rough keyword matching against the query and title text, the same "coarse, disclosed, and imperfect" approach as any keyword heuristic. No single theme comes close to a majority:
View the data: category mix across all sampled rows
| Category | Share of sampled rows | Row count |
|---|---|---|
| Sports | 17.0% | 443 rows |
| Politics / Government | 3.6% | 93 rows |
| Entertainment | 2.8% | 73 rows |
| Weather / Disaster | 2.6% | 67 rows |
| Finance / Lottery | 2.3% | 60 rows |
| Everything else (names, one-off news) | 71.7% | 1,864 rows |
That "everything else" bucket isn't a black box we're waving away. We looked inside it. A large share of it is short, bare queries that our theme keywords simply can't catch because they aren't about a theme at all. They're a specific person's name: travis kelce, tony romo, selena gomez, kelsey plum, gerard butler, edwin diaz. Running a simple shape check (a 2-3-word query, no digits, no filler words like "of," "the," or "vs") over every "Other" row finds that about 42% of every sampled item on the entire board (not just the "Other" bucket) matches that bare-name pattern. Reliably telling a real person's name from a two-word news phrase needs proper entity recognition, which this heuristic doesn't do, so we're not claiming an exact person-count. But directionally it's clear: Google's real-time trending board is dominated by who, not what: individual athletes, actors, musicians, and public figures generate more trending queries than every news theme we could name, combined.
Weekday pattern: sports peaks on Saturday (small sample, treat as suggestive)
With only 26 sampled days spread unevenly across the week (3 to 5 sampled days per weekday; see the counts in each bar's note below), any single weekday's share is a small-sample estimate, not a settled finding. With that caveat up front, the pattern that shows up is the one you'd expect:
View the data: sports share by weekday
| Weekday | Sports share of that weekday's rows | Sample size |
|---|---|---|
| Saturday | 22.3% | 400 rows, 4 sampled Saturdays |
| Thursday | 19.0% | 400 rows, 4 sampled Thursdays |
| Sunday | 17.0% | 300 rows, 3 sampled Sundays |
| Wednesday | 17.0% | 300 rows, 3 sampled Wednesdays |
| Monday | 16.3% | 400 rows, 4 sampled Mondays |
| Tuesday | 15.4% | 500 rows, 5 sampled Tuesdays |
| Friday | 11.3% | 300 rows, 3 sampled Fridays |
Saturday (college football, MLB doubleheaders, international soccer) tops the list and Friday is the trough. That's directionally plausible, but with 3-5 sampled instances per weekday we'd want a much longer run before calling this a confirmed cycle rather than a suggestive one.
Why the number is this extreme: read the sampling cadence first
Before drawing a broader conclusion, it's worth being explicit about what "96% one-and-done" actually measures. This dataset is captured once a day, at a fixed time (09:15 UTC, roughly 5:15am ET / 2:15am PT), not a continuous feed. Google's real Trending Now board updates constantly through the day; our snapshot catches whatever is on it at that single moment each morning. A topic that spikes hard at noon Eastern and fades out by midnight never appears in this dataset at all, even if it dominated the real board for twelve hours. Conversely, a topic still active at 9:15am UTC on two different mornings genuinely persisted through the night in a way our sample can detect.
So the finding here is precisely: at a once-daily, fixed-time sampling cadence, the Trending Now board is almost entirely different from one snapshot to the next. That's a real and useful fact about how volatile the board is at daily resolution: useful for anyone deciding how often to poll it, or whether "trending" content strategies need same-day turnaround. It is not a claim about how many hours a topic trends for in real time; measuring that would require intraday sampling, which this panel doesn't do (yet).
How we measured this (and the caveats)
We pull Google's real-time Trending Now board via Crawlora's Google Trends API: GET /google/trends/trending?geo=US&window=24h&sort_by=search_volume&limit=100, which returns the identical ranked list you'd see live at trends.google.com/trending. A daily cron job (scripts/trending-snapshot-cron.mjs) runs this once a day at 09:15 UTC and appends that day's top 100 items to a committed, append-only dataset (src/data/trending-now-panel.ts); the pipeline refuses to write if the endpoint returns zero or malformed items, so a transient upstream error can't silently blank out a day.
- One sample a day, not a continuous feed. As covered above, this measures churn at daily-snapshot resolution. Anything that trends and fully fades between two 9:15am UTC samples is invisible to this dataset.
- 26 sampled days, not 26 calendar days. The window runs 2026-07-21 to 2026-08-18 (29 calendar days), but the cron missed 3 days (2026-07-22, 2026-07-26, 2026-08-14) to job/infra hiccups. Each gap is a single missed day, not a multi-day outage, and the panel has been running without interruption since. We analyze the 26 days we actually have and don't interpolate the gaps.
- US only, top 100 only.
geo=US,window=24h, capped at the endpoint'slimit=100. Findings here describe the US board specifically and say nothing about other countries or about items ranked below #100 on days the board was deeper than that. - Category clustering is a rough keyword heuristic, disclosed as such. We match query and title text against hand-picked keyword lists per category (Sports, Politics/Government, Entertainment, Weather/Disaster, Finance/Lottery); anything that doesn't match falls to "Other." It will misclassify edge cases (an athlete's name with no team/sport keyword lands in "Other," not "Sports") and undercounts every theme as a result: the 17.0% Sports figure is a floor, not a precise total. The 42% "name-shaped query" estimate is a structural pattern match (2-3 words, no digits, no filler words), not named-entity recognition, so it will include some non-person phrases and miss some person names that don't fit the 2-3-word shape.
started_unixandrelated_termsare mostly empty in this endpoint's response (only populated for some items), so day-over-day rank and presence in the top-N (the signal this study is built on) is the reliable field, not those two.- Rank changes for recurring queries aren't analyzed here. We track whether a query reappears, not whether its rank climbed or fell between appearances, a natural follow-up once the panel has more history.
This is the same fetch-and-normalize discipline behind our Search vs. Store study (a different Google Trends endpoint, tracking a fixed list of AI apps over time rather than whatever the board says is spiking) and our anti-bot adoption index: pull data that's normally locked behind a rendered, rate-limited page as clean JSON, then let the numbers say something honest, including "this sample is too small to conclude X," when that's the truth.
Pull Google's real-time Trending Now board as clean JSON
Crawlora turns Google Trends (the real-time trending board, interest-over-time, related queries) into normalized JSON, with proxies, rendering, and anti-bot handled for you. Pay-on-success billing, 2,000 free credits a month, no card.
Frequently asked questions
How much does Google's Trending Now board change day to day?
Almost completely. Sampling the board once a day for 26 sampled days (2026-07-21 to 2026-08-18), on average 99.4% of the top 25 (and top 50) is brand-new compared to the previous sampled day. Across the whole window, about 96-97% of every query that ever cracks the top 25/50 appears on exactly one of the 26 sampled days and never shows up again. This measures churn at a once-a-day, fixed-time sampling cadence — not the true real-world lifespan of a trending topic, which our single daily snapshot can't fully capture.
Do the same topics ever come back on Google's Trending Now board?
Rarely, and almost never on consecutive sampled days. Of 1,247 unique queries that cracked the top 50 across the 26-day window, only 42 (3.4%) appeared more than once, and the longest unbroken streak of consecutive sampled days found in the entire dataset was 2 days. The queries that do recur are legible: a playoff-hunt baseball team (Dodgers, 5 of 26 days), a twice-weekly lottery draw (Powerball numbers, 4 of 26), a tennis player's tournament run (Alexandra Eala, 4 of 26), and an ongoing legal case.
What kind of topics dominate Google's Trending Now board?
Not broad themes — specific names. Rough keyword clustering across 2,600 sampled item-rows finds Sports as the largest identifiable category at 17.0%, followed by Politics/Government (3.6%), Entertainment (2.8%), Weather/Disaster (2.6%), and Finance/Lottery (2.3%) — combined, under 30% of the board. About 42% of every sampled item is a bare 2-3-word query with no theme keyword at all, a shape that looks far more like an individual person's name (an athlete, actor, or public figure) than a category.
Does a story's peak rank predict whether it comes back later?
Weakly, yes. Bucketing every query by the best rank it ever reached: items that cracked the top 10 reappeared on a later sampled day 8.8% of the time, versus 4.3% for items that never broke the bottom half of the board (rank 51-100). The two middle buckets (11-25 and 26-50) sit close together at 6.1% and 6.6%, so the relationship is a real gradient, not a clean straight line.
How was this measured?
Via Crawlora's Google Trends API (GET /google/trends/trending, geo=US, window=24h, sort_by=search_volume, limit=100) — the same data behind trends.google.com/trending. A daily cron appends one snapshot of the top 100 items per day to a committed dataset; this study covers 26 sampled days between 2026-07-21 and 2026-08-18 (3 days were missed to job hiccups). It's a once-a-day, fixed-time (09:15 UTC) sample, not a continuous feed — a topic that spikes and fully fades between two snapshots never appears in the dataset. Category clustering is a disclosed, rough keyword heuristic, not exact.