Tracing Search Geographies with Google Trends: 6 lessons learnt

In a new article that will be presented at AGILE later this year, my colleagues Simon Scheider, Bas Spierings, and I explore the potential of Google Trends to understand how the search interest in geographic areas changes in space and time, looking at the Amsterdam metropolitan region as a case study [read the full paper in PDF].


This is what we learnt from this data science exercise:

  1. The geographic scale has a strong impact on the search behaviours for places. For example, patterns at the country and at the city level appear very different (e.g. city searches correlate to hotel bookings, country searches do not). Furthermore, for scales finer than a municipality, the data is not granular enough.
  2. The resolution of the Google Trends Index (GTI) is intrinsically limited. Our algorithm (see Section 4.1) can help increase the resolution of the index resolution for relatively low-interest areas. However, peripheral small neighbourhoods are likely to have a GTI = 0, which cannot be re-scaled. Many prominent points of interest can attract higher interest than large spatial units (e.g. Van Gogh Museum).
  3. Semantic ambiguity of search terms is a critical problem that cannot be solved completely. The manual inspection of search results for each query in the study is highly recommended. An approach worth pursuing might be to distinguish between tourism, reference, and housing searches through more specific terms (e.g. “Amsterdam hotels” as opposed to “Amsterdam rent”).
  4. Estimates from SEO companies such as SemRush can provide a quantification of search volume to complement the GTI, although the quality of such data is largely undefined.
  5. Searches for places tend to exhibit strong seasonality. Interest bursts are a serious issue and should be accounted for. Burst-causing events might be both unplanned (natural and man-made disasters) and planned (e.g. the World Cup).
  6. It is essential to possess local expert knowledge in order to interpret trends and spatial patterns both in interest origin and targets. Even so, in our case study, some patterns remained hard to interpret with obvious explanations (e.g. tourist flows). Place searches have their own peculiar, rather volatile geography, and some hard-to-explain variation must be expected.

Abstract. Search engines make information about places available to billions of users, who explore geographic information for a variety of purposes. The aggregated, large-scale search behavioural statistics provided by Google Trends can provide new knowledge about the spatial and temporal variation in interest in places. Such search data can provide useful knowledge for tourism management, especially in relation to the current crisis of tourist (over)crowding, capturing intense spatial concentrations of interest. Taking the Amsterdam metropolitan area as a case study and Google Trends as a data source, this article studies the spatial and temporal variation in interest in places at multiple scales, from 2007 to 2017. First, we analyze the global interest in the Netherlands and Amsterdam, comparing it with hotel visit data. Second, we compare interest in municipalities, and observe changes within the same municipalities. This interdisciplinary study shows how search data can trace new geographies between the interest origin (what place users search from) and the interest destination (what place users search for), with potential applications to tourism management and cognate disciplines.

Reference: Ballatore A., Scheider S., Spierings B. (2020) Tracing Tourism Geographies with Google Trends: A Dutch Case Study. In: Kyriakidis P., Hadjimitsis D., Skarlatos D., Mansourian A. (eds) Geospatial Technologies for Local and Regional Development. AGILE 2019. Lecture Notes in Geoinformation and Cartography. Springer, Cham [pdf] [web]