Skip to content
All Resources
Location Intelligence 7 min read

Prepared by GSDSI Regulatory Content Team

Last updated

Editorial standardsTrust Center

Why POI counts differ between providers, and how to compare them

Point of interest (POI) counts differ between providers because each one defines a place differently, draws on different sources, includes different feature types and applies different filters for closed, duplicate or low-confidence records. Totals also change with every release. Compare counts only after matching the definition, geography, filters and release date.

How to use this article

Read the checklist here, then use the linked hub and product pages for procurement citations.

Ask two providers how many places they hold and the answers rarely agree. That gap does not mean either number is wrong. The totals are built on different rules. This guide leans on two public projects, OpenStreetMap and the Overture Maps Foundation's places theme. Both publish their definitions, schemas and release notes for anyone to read, so every claim below can be checked.

Why do POI counts differ between providers?

POI counts differ for five reasons: definition, coverage, feature types, counting method and release timing. The table pairs each reason with what it changes and what to ask. Later sections walk through the public evidence for each one.

Five reasons POI counts differ, what each one changes and what to ask, as of November 2026. Source: GSDSI editorial analysis of public OpenStreetMap and Overture Maps documentation.
ReasonWhat it changesWhat to ask the provider
Definition of a placeWhether parks, landmarks, street furniture or peaks sit in the file beside businesses.How do you define a point of interest, and which categories are excluded?
Geographic coverageWhere the records sit. A global total can hide thin countries and cities.What is the count for my countries, cities and categories?
Feature typesThe share of rows that are not commercial locations at all.Which non-commercial types are in the count, and can I filter them out?
Counting methodWhether closed, duplicate, doubtful and nested records are counted.Does the count include closed or low-confidence records, and how are malls and chains counted?
Release timingWhich snapshot the number describes.What release date and refresh schedule sit behind the number?

What counts as a point of interest?

A point of interest has no single agreed definition. The Humanitarian OpenStreetMap Team's tracing guide calls it a general term for a point-like feature noteworthy enough to include on a map. Overture's places guide defines a place as a concrete, physically identifiable, stationary destination in a publicly observable space. Its examples run from schools and businesses to landmarks and mountain peaks.

OpenStreetMap is the clearest case of a moving definition. The OpenStreetMap Wiki says points of interest are often tagged with the amenity key. It also says map nodes are not usually points of interest. Mappers describe places with many keys, so an analyst has to choose which keys to include. That choice sets the total.

One published extract shows the effect. The Layercake POIs layer from OpenStreetMap US includes a feature when it carries at least one of twelve top-level tags. Amenity, shop, tourism, leisure and office are among them. An extract built on fewer tags returns a smaller file from the same map.

Which feature types inflate a POI count?

Feature types that few buyers would call a business can make up much of an open-data total. On November 10, 2026, Taginfo's key statistics reported 35,325,236 OpenStreetMap objects carrying the amenity key. The per-value counts for the same day show 6,950,500 tagged parking, 5,149,017 tagged parking_space and 3,466,558 tagged bench.

Ten non-business values summed to 20,406,934 objects in the Taginfo data for November 10, 2026, or 57.8% of the key. The ten are parking, parking spaces, benches, waste baskets, bicycle parking, shelters, recycling, toilets, post boxes and drinking water. Restaurants, fast food, cafes, pharmacies and banks summed to 3,817,449, roughly one object in nine. Our earlier recount, OpenStreetMap amenity tags are not businesses, explains the method.

Map rendering tells the same story. The OpenStreetMap Carto symbol list includes icons for benches, picnic tables, drinking water and decorative fountains. Each is a real mapped feature. None is a storefront. A count that keeps them is honest about the map and misleading about commerce.

How does geographic coverage change the total?

Geographic coverage is uneven because every dataset inherits the reach of its sources. Overture's places guide, in its October 2026 source table, describes about 88 million place records in total. One contributor supplies roughly 60 million of them. A dataset shaped by one large source will be dense where that source is strong and thin where it is not.

Overture is candid about the limits. The same guide says the theme provides global coverage and that country-level figures will be published once measurements are confirmed. It also says the data is known to contain duplicates, a high junk rate and low property completeness. A global total tells a buyer nothing about one city or one category. For a country-by-country view of location data, see the global location data coverage matrix.

How do closed, duplicate and low-confidence records change a count?

Closed, duplicate and doubtful records are where counting methods diverge most. Overture's place schema gives each record a confidence score, from zero to one, for whether the place exists. The schema also carries an operating status: open, temporarily closed or permanently closed. A permanently closed place gets a confidence of zero, yet it can still sit in the file.

Status fields are only as good as the data behind them. When Overture added operating status in its September 2025 release, every value was set to open. Real values were planned for later releases. A buyer who counted open places in that release counted every record.

Review results show how far a raw batch can shrink. A slide from Overture's April 2026 member summit reports on a batch of about 350,000 new United States food and retail records slated for its May release. Roughly 57% were open, 31% were closed and 12% were junk.

Duplicates depend on matching rules. Overture's guide describes a pipeline that decides which source records describe the same real-world place, then keeps one record per match. A matcher that merges readily treats more records as duplicates and keeps fewer rows. A cautious matcher leaves more near-duplicates in the file, so its total is higher. Nested places add another choice. Counting a mall and each store inside it gives a different total than counting storefronts alone.

Why do POI totals change between releases?

Release timing matters because a count is a snapshot. Overture publishes its places theme monthly, and each release can change scores as well as rows. Its April 2026 release notes describe rolling back a confidence change made in March. The same notes say about 10 million places received updated scores.

Taginfo shows the same drift in open map data. The amenity key held 35,325,236 objects on November 10, 2026, per the key statistics request, and it was smaller when we read it for our earlier post on amenity tags. Comparing a number from one release with a number from another compares two different datasets.

How to compare POI providers fairly

A fair comparison fixes the rules before anyone reads a total. Send each provider the same five questions.

  1. How do you define a point of interest, and which categories do you exclude?
  2. Does the count include closed, temporarily closed or low-confidence records?
  3. How do you count chains, malls and other places that share an address?
  4. What is the count for my countries, cities and categories, not the global total?
  5. What release date and refresh schedule sit behind the number?

Then test a sample. Pick one city and one category you know well, apply the same filters to every file, and check the results against the street. The POI accuracy checks in our data quality guide and the POI provider comparison guide go deeper on both steps. A sample file costs less than a wrong assumption.

GSDSI describes its own POI product on the POI data page and the POI and geofencing product page. Bring the five questions to a scoping call, and ask for a sample in the markets you care about.

Every source on this page was read on November 10, 2026, except the earlier Taginfo reading dated October 15, 2026. Research by Catarina Costa Carvalho.

Frequently Asked Questions

Why do two POI providers report different totals?
Each provider decides what a place is, which sources to use, which feature types to keep and how to treat closed or duplicate records. Those choices differ, so the totals differ. The numbers also describe different release dates. Two totals are comparable only after the definition, geography, filters and date are matched.
Is a bigger POI count better?
Not on its own. A larger total can include street furniture, closed businesses, duplicates or records with a low confidence that the place exists. A smaller total can reflect stricter filtering. The useful measure is how many open, correctly located places a dataset holds in the countries and categories you need.
Does OpenStreetMap have an official POI count?
No. OpenStreetMap has no single tag that marks a point of interest. Mappers use keys such as amenity, shop and tourism, and each analyst chooses which keys to include. Any OpenStreetMap POI count is the result of that choice, so two extracts of the same map can report different totals.
How should I compare POI counts from two vendors?
Ask both vendors the same questions about definition, excluded categories, closed and low-confidence records, nested places and release date. Request counts for your own countries and categories. Then pull a sample for one city you know well, apply identical filters to each file and check the results against what is on the street.
How often do POI counts change?
Counts change with every release. Open datasets publish on their own schedules, and a release can add records, remove them or change the scores used for filtering. Always record the release date next to a count, and compare providers using releases from the same period where you can.

✓ Opt-Out Request Honored via Global Privacy Control