OpenStreetMap Amenity Tags Are Not Businesses: What the Count Really Contains

Why we recounted the amenity tag

When a buyer compares POI data, the open-data options come up first because they are free and the headline numbers are large. OpenStreetMap is the best known of them, and its amenity key sounds like a list of places to go. We wanted to know what that key actually holds, so we pulled the counts straight from Taginfo, OpenStreetMap's public tag statistics service, rather than quoting anyone's summary. The result is below, with the exact requests so you can repeat it.

This is a statement about one tag, not about OpenStreetMap as a whole. The project is an extraordinary public resource, and some businesses may be tagged under other keys, so the amenity count is not a ceiling or a floor for businesses. The point is narrower. If a number built on the amenity key is offered to you as a count of businesses, it is not one.

What the amenity key contains

Taginfo reported 35,070,905 objects with the amenity key, spread across 9,539 distinct values. The data was current to October 15, 2026. By object type that is 17,990,857 nodes, 16,881,179 ways and 198,869 relations. The table lists every value we counted, in the order of the two groups used in the shares that follow.

The ten-value list totals 20,218,411 objects, or 57.7% of the key. The fifteen-value list totals 21,549,202 objects, or 61.4%. The five business values total 3,802,305 objects, or 10.8%. Those three percentages are only true for those exact lists, which is why we name every value. A share quoted without its list cannot be checked.

What this means for a POI buyer

• A tag count is not a place count. The key holds both parking (6,906,148 objects) and parking_space (5,071,238 objects) as separate values, and each object is counted separately.
• The business share is small. Benches and bins sit beside restaurants under the same key. That is fine for mapping and a hazard for anyone who sums the key and calls it a business list.
• Counts move. Taginfo is a live statistic, so the number you read today is not the number from last month. Always carry the date with the figure.
• Filter before you compare. If you plan to benchmark an open source against a licensed POI file, decide which values count as a place a consumer would visit, then count only those on both sides.
• Ask every vendor what is inside the number. A global record count should come with a definition of a record, the categories included and the date counted. Our POI provider comparison makes the same point about record-count marketing across fourteen providers.

For the tests that matter once you have a clean place list, such as polygon fidelity, closure latency and category accuracy, see POI data quality and accuracy for foot traffic. GSDSI's own POI data is described on that page, and we recommend the same matched-sample test on your geography and categories for it that we recommend for any vendor. If a place file is paired with mobility data, the FTC sensitive location guide for buyers covers what to check.

How to count a place file honestly

The same discipline applies to any file you are offered, open or licensed. Before a record count goes into a spreadsheet or a vendor scorecard, write down four things next to it. First, the unit: is a record a business, a building, a brand location, a mapped feature or a point on a road? Second, the filter: which categories or tag values were included, named one by one. Third, the date: counts from a live source change daily, and counts from a vendor's marketing page may be years old. Fourth, the source of the number: the primary statistics page, a release note or a sales deck.

Then test the number against something you can see. Pick a neighborhood you know and walk it, or open it on a map. Count the restaurants, pharmacies, banks and shops you can name, then count how many records the file holds for the same blocks and how many of those are things nobody would call a business. That ten-minute exercise tells you more about fit than any global total, and it is the same test we describe in our guide to running a bake-off on one trade area.

None of this is a criticism of open data. A bench or a recycling point is a legitimate map feature, and for routing, accessibility or urban planning work those values may be exactly what you need. The mistake is only in the translation, when a map inventory is read as a business directory.

Method and sources

All figures come from the Taginfo API, read on October 15, 2026, with data current to that day. The total and the object-type split come from the key statistics request. The per-value counts come from the key values request, sorted by count with 40 rows per page. Every share is the value's object count divided by the 35,070,905 total, rounded to one decimal place. The grouping into non-business and business values is our judgment, stated in the table so you can regroup it differently. Nothing here is a statement about the accuracy of any individual OpenStreetMap object.