Which cities attract the most attention on Wikipedia? Population cannot answer this question. Neither can tourist arrivals, airline traffic or the number of Google searches. A different perspective comes from Wikipedia pageviews: how often people actually open an encyclopedia article because they want to know something about a place.

Using open data from Wikipedia and Wikidata, it is possible to measure this information demand on a very large scale. For this ranking, pageviews of city-related articles on the English Wikipedia were analyzed over a 12-month period from July 1, 2025 through June 30, 2026. The result is not a ranking of the world’s largest cities, the richest cities or the best places to visit. It is a ranking of information attention: which city-related topics English Wikipedia readers opened most frequently during this period.

Producing such a ranking, however, is considerably more complicated than simply downloading a list of Wikipedia pages and adding up their views.

Watch the full video:

Wikipedia as a sensor of information demand

Wikipedia is not only an encyclopedia. From a data-analysis perspective, it can also be regarded as a large-scale sensor of public demand for information.

When readers open an article, Wikimedia’s infrastructure records aggregated pageview statistics. These statistics are available through the Wikimedia Analytics API and through public pageview data dumps.

The data made available through the Analytics API is published under the CC0 1.0 public-domain dedication, which makes it particularly attractive for research, reproducible analysis and data journalism.

The scale of Wikipedia traffic is enormous. In a study published for Wikipedia’s 25th anniversary, the Pew Research Center estimated that Wikipedia articles accumulated around 1.9 trillion pageviews between 2015 and 2025, with English Wikipedia accounting for a very large proportion of this activity.

This makes Wikipedia useful for studying what might be called information demand. For instance, when a film becomes popular, people read about its actors. When elections take place, interest in politicians and political institutions increases. When a major scientific discovery is announced, related scientific articles receive additional attention. Sporting events, wars, anniversaries, celebrity deaths and viral stories can all leave visible traces in Wikipedia traffic.

Cities behave in a similar way. For example, a person might open the article about Paris while planning a holiday, read about Jerusalem after seeing it in the news, research Constantinople while studying history, or open New York City after watching a film. Wikipedia does not tell us exactly why an individual reader opened an article, but in aggregate, millions of such actions create a valuable indicator of information interest.

Wikipedia data has several important advantages:

  • it covers an extremely large global audience;
  • pageview statistics are available on a daily basis;
  • the methodology is relatively consistent over time;
  • the data is openly available;
  • Wikipedia articles can be connected to structured entities in Wikidata;
  • different language editions of Wikipedia can be studied separately or compared;
  • long time series make it possible to identify spikes, trends and seasonal patterns.

The current Wikimedia pageview analytics series begins on July 1, 2015, making more than a decade of comparable analysis possible. More information is available in the official Wikimedia Page View Analytics documentation.

There is an important limitation – Wikipedia pageviews measure attention to information. They do not directly measure tourism, economic importance, quality of life, popularity among residents or positive attitudes toward a place. So, five million pageviews are also not necessarily five million different people. One person can open an article several times. For this reason, the ranking should be interpreted as a ranking of Wikipedia information demand, not as a universal ranking of the world’s cities.

The first problem: what exactly is a “city” on Wikipedia?

Before counting pageviews, we first need to identify the Wikipedia articles that are about cities. That sounds simple – but it’s not 😉

One possible approach would be to use Wikipedia categories such as “Cities in the United States”, “Cities in Japan”, “Capitals in Europe” and thousands of similar categories. However, Wikipedia’s category system has developed organically over more than two decades. It contains an enormous number of interconnected and nested categories. Categories may overlap, and different Wikipedia editors and national communities may organize geographic topics differently. A city might appear primarily as a municipality, capital, metropolitan center, historical settlement, port, administrative division, archaeological site or several of these at the same time.

Following Wikipedia category trees recursively therefore creates two opposite risks:

  • false negatives — real cities that the algorithm fails to find;
  • false positives — articles that enter the candidate set even though they are not really cities.

The problem becomes even more complicated when historical cities, city-states, boroughs and special administrative entities are considered.

A more systematic solution is to connect Wikipedia articles with their corresponding Wikidata items.

Using Wikidata to identify cities

Wikidata is a structured knowledge base in which entities are represented by identifiers and connected through machine-readable statements.

For example, the general Wikidata entity for a city is city (Q515).

Two Wikidata properties are particularly important for our task:

The obvious first idea would be to search for all Wikidata entities with: instance of (P31) = city (Q515). Unfortunately, that would miss many cities. Real cities are frequently classified using much more specific values. Examples include:

There are many more such classes. Consequently, an algorithm that checks only whether P31 = Q515 would produce an incomplete list.

Following subclass relationships in Wikidata

One solution is to combine Wikidata’s instance of (P31) statements with the subclass of (P279) hierarchy. A useful first step is to search for Wikidata items that are used as values of instance of (P31) and are connected, directly or indirectly through subclass of (P279), to city (Q515).

For example, the following illustrative SPARQL query can be used in the Wikidata Query Service:

SELECT DISTINCT ?class ?classLabel ?directSubclass ?parent
WHERE {
  ?class wdt:P279* wd:Q515.

  OPTIONAL {
    ?directSubclass wdt:P279 ?class.
  }

  OPTIONAL {
    ?class wdt:P279 ?parent.
  }

  SERVICE wikibase:label {
    bd:serviceParam wikibase:language "en".
  }
}

The expression wdt:P279* means that SPARQL can follow zero or more “subclass of” relationships. This is useful because it allows us to discover a family of increasingly specific concepts that ultimately belong to the broader concept of a city.

Wikidata’s official SPARQL tutorial describes the same technique for retrieving instances of a class and its subclasses. A compact form for retrieving potential city entities is: ?item wdt:P31/wdt:P279* wd:Q515.. In plain English, this means: Find an item whose “instance of” class is either city itself or some class that eventually becomes a subclass of city. This approach is much more flexible than looking only for items whose P31 statement contains Q515 directly.

Why even Wikidata classification is not completely simple

Structured data helps enormously, but it does not eliminate every ambiguity. Real-world definitions of “city” vary between countries and historical periods. Wikidata also contains multiple overlapping classification systems created by thousands of contributors. A geographic entity may simultaneously be classified as a municipality, administrative territorial entity, city, capital, historical settlement or archaeological site.

Wikidata’s own WikiProject Cities and Towns discusses some of these classification issues. This is especially important when interpreting the final ranking. It contains conventional modern cities such as London and Chicago, but also city-states, boroughs, historical cities and other strongly city-related geographic entities. Examples include Brooklyn, Manhattan, Monaco, Vatican City, Babylon, Troy, Constantinople and Machu Picchu.

For that reason, the most precise methodological description of the dataset is a ranking of city-related entities rather than a claim that every entry has exactly the same present-day legal or administrative status.

Connecting Wikidata items to English Wikipedia

Once appropriate Wikidata entities have been identified, the next step is to find their corresponding articles in English Wikipedia. This is another major advantage of Wikidata. A Wikidata item can contain sitelinks connecting the same entity to articles in Wikipedia’s different language editions. Therefore, after identifying a geographic entity in Wikidata, it is possible to obtain the title of its English Wikipedia article.

Collecting 12 months of Wikipedia pageviews

The measurement period for this ranking is: July 1, 2025 – June 30, 2026.. Wikipedia pageview information can be obtained either through the Wikimedia Analytics API or from the much larger pageview dump files.

The API is convenient when collecting statistics for a more limited number of known articles. Bulk dumps can be more efficient when processing very large numbers of articles.

For this ranking, the pageview statistics correspond to the following conceptual API settings:

  • project = en.wikipedia.org
  • access = all-access
  • agent = user
  • granularity = daily
  • start = 20250701
  • end = 20260630

Parameter access = all-access means that the analysis is not restricted to desktop traffic alone. Traffic from the supported access methods is combined. Parameter agent=user is particularly important because Wikimedia also distinguishes traffic classified as spiders and other automated traffic. The official Wikimedia pageview documentation explains the different traffic categories and the rules used to identify pageviews. Filtering automated traffic makes the resulting values much more useful as an approximation of actual reader attention.

Why redirects must be included

There is another technical complication that can substantially change the final result: redirects and article-title changes. Suppose a Wikipedia article receives hundreds of thousands of views under one title and is later renamed. Wikipedia will normally leave the previous title as a redirect to the new title. Readers following an old link still reach the same subject. From the perspective of measuring interest in the subject, views of these different titles should often be considered together. However, historical pageview statistics are associated with the title that was actually requested. Wikimedia’s documentation specifically notes that views of redirects are not automatically counted as views of the final target article. Therefore, collecting pageviews only for the article’s current canonical title can underestimate its historical popularity.

Imagine, for example, that an article used the title City A during the first half of the study period and was renamed City B in January. If the analysis collects only historical pageviews for City B, traffic recorded under City A can be missed. For this reason, a robust analysis should identify relevant redirects and title history and aggregate the corresponding pageviews where appropriate.

The MediaWiki Redirects API and the redirect table can help identify pages that currently redirect to an article. Another useful interactive tool is Pageviews Analysis, which allows users to inspect pageview time series and optionally include redirects.

A pageview is not always natural information demand

After collecting and aggregating the pageviews, it might seem that the ranking is finished. Unfortunately, there is one more problem. Not every apparently human Wikipedia pageview necessarily represents natural reader interest.

The best-known example is the extraordinary Wikipedia traffic received by the article about Cleopatra. For a long period, Wikipedia editors noticed that Cleopatra was receiving remarkably high and persistent traffic without an obvious news event that could explain it. The mystery was investigated by Wikipedia contributors and journalists. One important explanation involved Google Assistant, which used a suggestion similar to “Show Cleopatra on Wikipedia” as an example voice command on some Android devices. The phenomenon is described in the article Why is Cleopatra constantly trending on Wikipedia? published by Inverse. The issue was also discussed in The Conversation.

This example demonstrates an important methodological problem. The requests may originate from real users and real devices, but an application, interface, recommendation system or default link can create large amounts of traffic that do not correspond to millions of independent decisions to research the subject. For popularity research, such traffic can distort a ranking.

Detecting anomalous Wikipedia traffic

For this reason, the raw totals used in this analysis were subjected to additional checks for anomalous traffic. The general approach was inspired by techniques previously discussed by WikiRank, where unusually inflated or “overclocked” popularity statistics have been investigated using additional contextual indicators. Potential signals can include:

  • the shape of the daily pageview time series;
  • unusually persistent traffic without an identifiable event;
  • extreme differences between desktop and mobile traffic;
  • changes in the number of edits;
  • changes in the number of active editors;
  • article protection events;
  • changes in links and citations;
  • external search interest;
  • major news events involving the subject;
  • traffic patterns of comparable Wikipedia articles.

A useful example of this approach can be found in WikiRank’s discussion of abnormal popularity measurements: WikiRank upgrade: fair popularity and new features. Another useful methodological example comes from the Pew Research Center’s 2026 study of Wikipedia.

Pew counted human-user traffic and excluded automated activity. Researchers also applied an additional heuristic based on the distribution of views by device: pages for which mobile traffic represented less than 5% or more than 95% of views were excluded from some popularity calculations. The researchers also excluded several pages with known or suspected artificial traffic distortions, including Cleopatra.

This illustrates an important principle: using agent=user is necessary, but for some kinds of research it may not be sufficient. A sudden traffic spike is not automatically suspicious. A city can legitimately receive millions of additional views because of an earthquake, war, sporting event, film, political event or other major development. The goal is therefore not simply to remove spikes. It is to distinguish explainable public interest from traffic patterns that appear to have been generated by technical or interface-related mechanisms. The final ranking therefore represent totals after redirect aggregation and additional anomaly checks and corrections rather than a blind sum of every raw page request.

The 50 most-viewed city-related articles on English Wikipedia

The resulting ranking for the period July 1, 2025 – June 30, 2026 is shown below. Pageview values represent the totals used in this study after the described processing, redirect handling and anomaly checks.

  1. New York City (5,448,443)
  2. Singapore (4,431,672)
  3. Hong Kong (3,249,417)
  4. Brooklyn (3,174,187)
  5. London (3,092,746)
  6. Washington, D.C. (2,974,686)
  7. Los Angeles (2,917,643)
  8. Monaco (2,708,837)
  9. Chicago (2,351,214)
  10. Dubai (2,138,711)
  11. Tokyo (1,927,396)
  12. Jerusalem (1,907,306)
  13. Paris (1,905,260)
  14. Mumbai (1,768,391)
  15. San Francisco (1,741,185)
  16. Toronto (1,735,944)
  17. Berlin (1,730,307)
  18. Philadelphia (1,670,222)
  19. Boston (1,640,581)
  20. Seattle (1,620,036)
  21. Rome (1,563,788)
  22. Vatican City (1,545,922)
  23. Macau (1,543,003)
  24. Istanbul (1,528,192)
  25. Mexico City (1,516,313)
  26. Saint Petersburg (1,453,139)
  27. Atlanta (1,433,038)
  28. Machu Picchu (1,373,177)
  29. Chongqing (1,343,670)
  30. New Orleans (1,321,826)
  31. Prague (1,318,991)
  32. Shanghai (1,313,562)
  33. Babylon (1,312,031)
  34. Vancouver (1,311,284)
  35. Montreal (1,301,798)
  36. Las Vegas (1,301,113)
  37. Vienna (1,262,813)
  38. Minneapolis (1,261,474)
  39. Constantinople (1,252,602)
  40. Manhattan (1,245,518)
  41. Detroit (1,238,274)
  42. Gibraltar (1,227,408)
  43. Sydney (1,220,463)
  44. Moscow (1,219,667)
  45. Beijing (1,217,741)
  46. Amsterdam (1,216,232)
  47. Miami (1,216,004)
  48. Houston (1,214,556)
  49. Barcelona (1,208,087)
  50. Troy (1,205,970)

An extended version of the ranking is presented in this video.

What does the ranking tell us?

The most obvious result is the strong lead of New York City, whose article accumulated almost 5.45 million views in the processed dataset during the analyzed year. Singapore follows with more than 4.43 million views, while Hong Kong, Brooklyn and London complete the top five.

However, the diversity of the ranking is more interesting than the identity of the winner. Modern global cities such as New York, London, Tokyo, Paris, Dubai and Singapore appear alongside places whose Wikipedia popularity is connected strongly with history, politics or special administrative status. Babylon, Troy and Constantinople demonstrate the importance of historical interest. Monaco and Vatican City combine the concepts of city and state. Brooklyn and Manhattan illustrate how Wikipedia readers often search for important urban entities independently from the city to which they belong.

The appearance of Machu Picchu is also a reminder that automated geographic classifications need interpretation rather than blind acceptance. These examples reveal something fundamental about Wikipedia usage: readers do not divide their curiosity neatly into modern geography, administrative law, history, tourism and popular culture. All these motivations become part of the same pageview stream.

Wikipedia rankings are also rankings of concepts

There is another important interpretation of the results. A Wikipedia ranking is partly a ranking of how knowledge has been divided into articles. For example, New York City has its own article, but Brooklyn and Manhattan also have extremely prominent independent articles. Readers searching for information about New York may therefore distribute their attention across several pages. Other cities may have a much larger proportion of relevant information concentrated in a single article.

This means that pageview totals cannot be interpreted as perfectly standardized measurements of cities themselves. They measure attention to the Wikipedia concepts and article structures representing those places. That is not necessarily a disadvantage. In fact, it gives us additional information about what specific aspects of an urban area attract independent reader interest.

Creating a ranking is easier than defining what it really means

Adding numbers together is the simplest part of creating a Wikipedia popularity ranking. The difficult work happens before and after the summation:

  • defining what counts as a city;
  • navigating Wikidata’s item hierarchy;
  • finding city-related subclasses rather than checking only Q515;
  • connecting Wikidata entities to English Wikipedia articles;
  • collecting daily pageview statistics;
  • reconstructing redirects and article-title changes;
  • filtering known automated traffic;
  • checking whether unusual pageviews represent genuine information demand;
  • interpreting heterogeneous entities consistently.

These methodological decisions determine what the final ranking actually means. The strength of the Wikimedia ecosystem is that much of the required infrastructure is openly accessible. Wikipedia provides the encyclopedia articles, Wikidata provides a machine-readable semantic layer, the MediaWiki API provides information about page relationships, and Wikimedia Analytics provides historical measurements of reader attention.

Together, these resources allow us to look at cities from an unusual perspective. Instead of asking only “How many people live there?”, “How many tourists visit?” or “How large is its economy?”, we can ask another question: How often do people decide that they want to know more about this place?

Seen this way, Wikipedia is not merely an encyclopedia. It is also a large-scale, continuously changing observatory of human curiosity.