Hello - We use the geonames gazetteer in our application CLIFF which does extraction and geolocation of places from news articles -- cliff.mediameter.org
We are currently having a problem because in the geonames.org - "São Paolo" is the main way that geonames.org spells the capital city of Brazil so it is being ignored by our entity extraction engine. The correct English spelling is São Paulo.
If this poses a problem to you then I strongly suggest to use the premium data extraction instead of the free one. The main advantage of the premium extract is exactly that the data goes through a releases cycle and changes like this are undone before the release.