Frequently Asked Questions

We have provided 3 layers of demographics data from the 2021 U.S. census, and we have 11.2mm Points of Interest (POI) data in the U.S. released by Overture Maps Foundation.

Here are the fields used in querying the data for visualization or summary: names, categories, and addresses.

Here are the properties we display for each point of interest: names, categories, addresses, and confidence (i.e. the confidence of the existence of the place).

In the U.S., 80% of the POIs have categories filled out. Globally, 88% of the POIs have categories filled out.

We ask ChatGPT to interpret the user's requests and support the following results for POIs that fit the search criteria:

• POI layer visualization
• POI count

The query against the Overture Places database includes the following:

• Geography: We support granularities on a city, state, or zip code level. If not specified by the user, the default is nationwide.
• POI category: We return all features with Overture Places "categories" property that most closely match the user's input. Note that quality of the category match is limited by the comprehensiveness of the "categories" property in the dataset.
• POI name: We return all features with Overture Places "names" property that match the user's input and are in the most commonly used category (e.g. "McDonald's Law Firm" is not returned in a search for "Show me all McDonald's in Colorado"). Note that we only match your query to a category OR a name and don't support the ability to specify both currently.

We use Mapbox Street map for our base map. The Points of Interest data from Mapbox may not match with the Overture dataset (e.g. businesses, or locations of the pin). The basemap is only there to provide some context on the area.

• Incorrect coordinates: There are features in the Overture Places dataset that have the wrong coordinates (e.g. a restaurant in Florida may show up in Mexico).
• Missing POIs or returning POIs from wrong categories: Because we are only returning the most commonly used category for each keyword search and there are ~20% of features that don't have a category tagged, this method may miss certain POIs or return the wrong POIs that belong to a more popular category. We are working on improving this experience in future iterations.
• Incorrect interpretation of category vs. brand: Sometimes ChatGPT doesn't do a good job at interpreting whether you are searching for a name or a category. If you want to force it to search for a specific category, you can say something like Find category Afghan Restaurants to force it to use a specific category.