Google Trends is a free Google service that reports how often a term is searched relative to every other search made in the same place and period. It publishes no volume figures. What it publishes is an index running from 0 to 100, where 100 marks the highest point the selected term reached inside the selected window and geography, and every other reading is a fraction of that peak. The service exists because Google holds the largest continuous record of what the public states it wants, will not release the underlying counts, and has chosen instead to expose a normalised, sampled and heavily filtered view of them.

How the index is built

Normalisation runs in two steps, both described in Google's own data FAQ. Each data point is divided by the total searches for the geography and time range it represents, which stops dense markets from ranking highest on every query by default. The resulting proportions are then scaled from 0 to 100. A term searched two hundred times a week and a term searched two million times a week can both return a reading of 100.

The input is a sample rather than the full log. Google's stated reason is processing speed: a complete dataset would be too large to query within minutes of an event. Several categories never reach the sample at all. Terms with little volume return 0. Repeated searches from the same person within a short period are stripped out. Queries containing apostrophes and other special characters are filtered. Internal searches issued by Google products, including AI Mode and AI Overviews, are excluded, which means the tool measures typed demand and not the machine traffic now generated on top of it.

Two further caveats sit in the same document. Google adds statistical noise for privacy reasons, most visible on low-interest queries, and warns that apparent one-off spikes on such terms may be artefacts rather than searches. And automated or manipulative queries are sometimes deliberately left in the data, because filtering them would signal to their authors that they had been detected. Graphs covering 30 days or more use Coordinated Universal Time; ranges of seven days or less follow the local time zone of the browser.

Three surfaces, one dataset

The homepage leads to Explore, Trending now and Year in Search. Explore is the analytical surface: up to five terms or topics compared at once, geography from worldwide down to city level, category filters, and a property selector covering web search, images, news, Shopping and YouTube search. A tutorial series published in August 2024 documented the option of leaving the query field empty and driving the filters alone, which surfaces the dominant topics for a region or period without a hypothesis attached. A second episode covered the operators, including quotation marks for phrase matching, year-on-year comparison mode, dynamic sharing URLs, embeddable charts and per-card CSV export.

Trending now was rebuilt in August 2024 around a forecasting engine that Google said detects ten times more emerging trends, refreshes every ten minutes, and covers more than 100 countries with regional breakdowns in 40 of them. The table shipped the following month carries fields the Explore page does not: an approximate count of actual searches, a growth rate measured against predicted volume, a trend start time, an active or ended status marker and a 24-hour spark line. Search Advocate Daniel Waisberg tied the time filters to editorial use, noting that a story about the opening week of the Olympics calls for a seven-day window rather than the past hour.

From 2006 to an API

Trends launched on 11 May 2006 with history reaching back to 2004. Maintenance was erratic at first, with bloggers noting long gaps in updates through 2007. Google released Insights for Search, a more granular sibling, on 5 August 2008, then folded it back into Trends on 27 September 2012. Two spin-offs did not survive. Google Flu Trends, launched in 2008, was retired in 2015. Google Correlate, which ran the logic in reverse by matching an uploaded series against queries with similar shapes, closed on 15 December 2019 for lack of use.

Programmatic access arrived late. Google opened alpha applications for a Trends API on 24 July 2025, offering a rolling window of 1,800 days, daily, weekly, monthly and yearly aggregations, and geographic restriction by region and sub-region under ISO 3166 codes. Its important departure is consistent scaling: because values are not rebased to 0 to 100 on every call, results from separate requests can be joined and compared, and the five-term ceiling of the website does not apply. Access remains gated. According to the scraping vendor Scrapfly, the documentation still described the interface as alpha and open to applications in August 2026, with no published pricing or general availability date. The community filled the gap unofficially for years through pytrends, a Python library archived on 17 April 2025.

What the data is used for

Google has been explicit about the commercial applications. In an October 2024 session on search data, Waisberg and Trends engineering manager Omri Weisman set out four: reading vertical demand to time stock and content, benchmarking against named competitors, monitoring the terms attaching to a brand over 30 or 90 days, and expanding keyword sets. Their worked example took a retailer fixed on cheese platter and found charcuterie carrying far more interest. A companion session for journalists covered cross-country comparison and minutes-old trend detection, and Google published a dedicated documentation page for content creators in the same month.

The stakes rose as search volume itself became contested. Research from Fractl covering 1,010,848 high-volume keywords, published on 2 July 2026, measured a 29% year-on-year fall in overall search demand, with fintech down 38%. A free public index of relative interest is one of the few instruments available to test claims of that kind outside a paid dataset, which is also why its construction matters.

Where it fails

The recurring failure is treating the index as a volume series. A June 2026 analysis of NBA Finals search activity drew the distinction plainly: Trending now reports approximate absolute searches, while Explore reports a normalised score, and spend decisions taken on the second as though it were the first carry a hidden error. A vendor claim that Casamigos doubled brand searches on $19.5m of World Cup television spend rested on a relative index with no control brand and no model, only coincident timing. An August 2026 claim that interest in the phrase King of SEO grew 250% compressed a twelve-month window that said nothing about the trailing weeks. Google is not exempt: a consumer post on home decor searches used the word breakout three times without citing the threshold Trends applies to it.

Academic work targets the sample rather than the interpretation. Because Google does not disclose its sampling method, the error cannot be quantified from outside, and repeated identical queries return different series. Carriere-Swallow and Labbe reported average standard deviations above 15% for the term Chevrolet across 50 extractions. Cebrian and Domenech found correlations between 0.79 and 0.94 across six extractions of the same queries and argued the spread was wide enough to threaten reproducibility. Eichenauer and colleagues recommend averaging at least twelve pulls.

The cautionary case remains Google Flu Trends, which overestimated influenza-like illness in 100 of 108 weeks from August 2011 and ran at roughly twice the CDC figure in the 2012-13 season. Lazer, Kennedy, King and Vespignani published the autopsy in Science in March 2014, attributing the failure to a static model exposed to a changing search product. Google's own FAQ makes the modest claim explicitly: Trends is not a poll, and a spike shows only that people searched.

Not the same as

Keyword Planner is an advertising tool reporting monthly volume ranges and cost forecasts, and it gained city, region and device-level breakdowns in July 2025; Trends reports relative interest across the whole search corpus and has no commercial dimension. Search Console reports queries that reached one property, not the market. Autocomplete is a prediction feature subject to removal policies and filtering, which makes it a poor proxy for popularity. Within Trends itself, Explore and Trending now run on different metrics, the first normalised and the second approximate absolute, and mixing readings between them produces nonsense.

Recent developments

Google added Gemini to the Explore page on 14 January 2026, with a side panel generating up to eight related terms from a stated area of interest, doubled rising-query counts, per-term colours and a raised comparison ceiling. Lead product manager Nir Kalush framed it as a research shortcut for journalists and creators. The same capability reached mobile web in April 2026, promoted by the Google News Initiative as field kit for reporters; a demonstration comparing nine jam variants over twelve months in the United States returned a peak of 71 for the leading query against 0 to 21 for the rest. The underlying 0 to 100 normalisation did not change. Separately, Google shipped an open-source Gemini tool for YouTube trend detection in early 2026, built on Vertex AI and the YouTube Data API rather than Trends data.

Timeline

  • 2004: Earliest period covered by Google Trends historical data
  • 11 May 2006: Google Trends launches
  • 5 August 2008: Google releases Insights for Search
  • November 2008: Google Flu Trends launches
  • 2011: Google Correlate launches
  • 27 September 2012: Insights for Search merged into Google Trends
  • 2015: Google Flu Trends discontinued
  • 15 December 2019: Google Correlate shut down for low usage
  • 6 August 2024: Trending now page redesigned with a new forecasting engine and filters
  • 28 August 2024: Tutorial covers search operators, comparison mode and export
  • 14 September 2024: Trending now table adds search volume, growth rate, start time and status
  • 23 October 2024: Google publishes guidance for journalists using Trends
  • 24 October 2024: Search Trends documentation page published for content creators
  • 17 April 2025: The pytrends library is archived
  • 24 July 2025: Trends API alpha announced with 1,800 days of consistently scaled data
  • 14 January 2026: Gemini term suggestions added to the Explore page
  • April 2026: Gemini-powered Explore page reaches mobile web
  • August 2026: Trends API documentation still describes the interface as alpha

Summary

Who: Google builds and operates the service, with Search Advocate Daniel Waisberg, engineers Hadas Jacobi and Omri Weisman and product lead Nir Kalush as its public voices. Marketers, SEO practitioners, journalists, academics and public health researchers are the main users.

What: A free index of relative search interest, scaled 0 to 100 against the peak of a chosen term, period and geography, drawn from a sample of Google Search and YouTube queries and stripped of low-volume terms, duplicates, special characters and internal product searches.

When: Launched on 11 May 2006 with data from 2004, absorbed Insights for Search in 2012, rebuilt its Trending now surface in 2024, opened an alpha API in July 2025 and added Gemini suggestions in January 2026.

Where: Available at trends.google.com worldwide, with geographic granularity from global down to city level and regional breakdowns for Trending now in 40 countries.

Why: It is the only free public read on aggregate search demand, which makes it the default evidence for seasonality, brand interest and category shifts. The same accessibility makes it the most frequently misread dataset in marketing, since a relative index carries no volume, no confidence interval and no guarantee that a repeated query returns the same series.