Google Trends Scraper

Returns the searches that are rising right now in any country — with how many people searched, how fast it grew, what category it belongs to, and the queries people search alongside it.

One request per country returns the whole list. Not a sample, not a top ten:

Country	Window	Trends returned
United States	24 hours	526
Argentina	24 hours	154
United Kingdom	4 hours	97
Spain	7 days	1,762

What you actually get

A spike dies in hours. The cluster of searches around it does not — that is what you write, bid or rank against:

{
  "term": "jed york",
  "volume": 200000,
  "growthPercent": 1000,
  "startedAt": "2026-08-24T14:40:00.000Z",
  "categories": ["Law and Government"],
  "relatedQueries": [
    "49ers owner", "jed york arrested", "jed york wife",
    "danielle belluomini", "san francisco 49ers", "49ers news",
    "east palestine ohio", "nfl news"
  ],
  "window": "24h",
  "rank": 1
}

Sixteen related queries for one trend, and up to 57 on some. That is keyword research that is hours old, not a monthly export.

Input

{ "geos": ["US", "GB", "AR"], "window": "24h" }

Only the sports stories that actually moved, with their query clusters:

{
  "geos": ["ES", "AR", "MX"],
  "window": "48h",
  "categories": ["Sports"],
  "minVolume": 20000,
  "onlyWithRelatedQueries": true
}

Parameter	Type	Default	What it does
geos	array	["US"]	Two-letter country codes. One request each.
window	string	"24h"	4h, 24h, 48h or 7d. A wider window returns far more terms — roughly 100 at 4 hours against 1,700 at 7 days for the same country.
categories	array	—	By name: Sports, Politics, Entertainment, Business and Finance, Technology, and the rest. A name Google does not publish stops the run and prints the valid list, instead of silently returning nothing.
minVolume	integer	—	Drops the long tail. 20000 leaves what actually moved.
onlyWithRelatedQueries	boolean	false	Keeps only the trends that carry a query cluster.
keywords	array	—	Keep trends whose term or related queries match — how you watch a brand across countries.
maxItems	integer	500	Hard cap on rows saved and charged. 0 means no cap.
includeSeen	boolean	true	Set to false to save only trends never returned before. Built for scheduled runs.

Output

Field	Notes
term	The search as people type it.
volume	Estimated searches. Google publishes these rounded.
growthPercent	Growth against the previous period. Google caps it at 1000, which means "over 1000%", not exactly 1000.
startedAt	When it began rising.
relatedQueries	What people search around the term, strongest first. Google publishes these for roughly one trend in five — the rest come back as an empty list rather than a guess.
categories	Names, not numbers. Read from the page's own filter list, so a rename by Google shows up as the new name instead of breaking.
window	The window the row was requested with. Volume is not comparable across windows, so it travels with the row.
rank	Position within that country and window.

Tracking trends over time

Set includeSeen to false and schedule the Actor. Each run then saves only terms it has never returned, so the dataset becomes a clean timeline of what started trending rather than the same list repeated every hour.

Cost

No proxy. Google Trends serves this to an ordinary client, verified across four countries and three windows. That makes this the cheapest Actor we publish — one request covers a whole country, and residential bandwidth, which is the dominant cost everywhere else, is zero here.

Empty and duplicate rows are never charged.

Notes

The list is read from the state Google ships to fill its own trends table, not from the page markup — that is why obfuscated class names being regenerated on every deploy does not affect this Actor.

Two endpoints that every public guide still recommends, /trends/api/dailytrends and /trends/api/explore, no longer respond. Both were tried and both failed; this page is what remains.

Google accepts ?category= in the URL but ignores it — the response is identical with and without it. Category filtering here happens on the extracted rows, which is why it works.