What percentage of songs contain at least one mention of a real place?

How often do English songs mention a real place? I took a random sample of 250,000 tracks from 3,233,212 English music pages on Genius, looked for place names in the lyrics, and only counted ones that check out as a real city, US state, country, or continent.

9.4% of songs mention at least one real place
Confidence range 9.2% – 9.5%
23,383 of 250,000 in this sample · 95% interval.

How often places show up

Share of songs that mention each kind of place. A song can land in more than one category. Click a card to filter.

Of the 23,383 songs with at least one real place, what kinds appear? Songs can count in more than one.

62.0%
City
24.2%
US state
30.3%
Country
3.5%
Continent

By genre

What share of songs in each genre name at least one real place? Click a genre to filter the whole page.

Genre% with a placeSongs
Rap
17.5%
13,072 of 74,593
Country
13.2%
895 of 6,759
Pop
6.0%
6,466 of 108,003
R&B
5.7%
681 of 12,011
Rock
4.7%
2,269 of 48,634

Where the places are

974 places on the map · 54497 mentions plotted.

Most-named places

Place names that show up most in the lyrics.

Los Angeles
4,002
New York
3,588
California
2,818
America
2,456
London
1,724
Texas
1,472
Paris
1,409
Chicago
884
Mexico
809
Tennessee
796
Japan
789
Miami
771

Over time

How often songs mention a real place, by decade (1960–2020). Click a point to filter.

0.0%6.0%12.0% 1960s · 7.7% (n=3135) 1970s · 9.7% (n=4944) 1980s · 8.1% (n=7364) 1990s · 8.7% (n=17035) 2000s · 8.9% (n=34795) 2010s · 9.9% (n=125576) 2020s · 8.8% (n=55943) 1960 1970 1980 1990 2000 2010 2020

How I did this

English Genius songs only, music genres. I left out the misc tag (poems, speeches, and other non-music pages).

A song counts if it names a real place: a city with at least 50,000 people, a US state, a country, or a continent. Some names are usually people in lyrics (Jordan, Austin), so I drop those.

I sampled 250,000 English music songs at random from 3,233,212 in the catalog.

Genius lyrics are often all caps, which trips up spaCy. I truecase with Moses first, then run en_core_web_sm and keep GPE and LOC mentions. A mention only counts if it matches a real place: cities via GeoNames (geonamescache), countries and US states via pycountry, continents from a short fixed list. Anything that doesn't match stays out of the rate.

The band around the rate is a Wilson 95% interval for this sample size.

  1. Truecase the lyrics with Moses (trained on this corpus).
  2. Run en_core_web_sm and keep GPE/LOC mentions.
  3. Check each mention against GeoNames (cities with population ≥ 50,000), pycountry, or the continent list. Apply the person/place denylist.
  4. If at least one mention checks out, mark the song. Write per-song flags (and hit rows when I want them) to parquet.
  5. Roll up the rates and attach a Wilson interval.