Exa Places Adds 211 Million Locations to Beat Perplexity on Local Search
Exa added 211 million restaurants, museums, parks, and local businesses to its search index with structured metadata for AI agents.
- Exa added 211M places to its search index with structured metadata for agents.
- Each result returns name, address, coordinates, hours, phone, website, and ratings as JSON.
- Exa Fast scored 0.800 F1 on places queries vs Perplexity 0.529 and Parallel 0.333.
- Natural language geolocation queries supported, like ramen in SF Mission open Friday 7pm.
- Available across all Exa search endpoints at no additional cost.
- Targets personal assistants, local GTM list building, and site-selection market research.
Exa adds 211 million places to its search index
Exa has added 211 million places to its search index, covering restaurants, museums, parks, shops, and other local businesses. The company calls the dataset Exa Places and makes it available through every Exa search endpoint.
The release gives AI agents structured local-business data through the same interface they use for web search. Existing search-call pricing still applies, but Exa charges no separate fee for Places results.
Place records ready for code
Each Places result uses a predictable schema with fields that an application can filter, rank, or pass into another workflow. Available data includes:
- Business name, category, address, and coordinates
- Opening hours and current operating status
- Phone number and website
- Ratings and other location metadata
Structured hours and status can help an agent check whether a business is open before recommending it. Coordinates support distance calculations, while phone numbers and websites can feed contact or follow-up steps without extracting values from page markup.
Natural-language queries become filters
Developers can describe location, time, category, and qualitative requirements in a single query. Examples from Exa include ramen restaurants in San Francisco’s Mission District that are open Friday at 7 p.m., Mexico City museums with reviews mentioning children’s activities, and highly rated independent bookstores in Brooklyn.
Exa interprets geographic terms in the query and returns matching businesses from the relevant area. That interface can reduce the application code needed to translate a user’s request into separate category, location, hours, and keyword filters.
Exa’s benchmark needs context
In a vendor-run evaluation of 400 place-search queries, Exa compared its Fast search mode with Perplexity Search and Parallel Advanced. The test measured whether each service found relevant businesses and returned correct metadata, including opening hours, phone numbers, and locations.
| Provider | F1 score |
|---|---|
| Exa Fast | 0.800 |
| Perplexity Search | 0.529 |
| Parallel Advanced | 0.333 |
F1 combines precision and recall into one score. Precision measures how many returned results were correct, while recall measures how many expected results the system found. A value of 0.800 summarizes the reported balance between those measures; exact-match accuracy requires a separate metric.
Exa Fast’s score was about 51% higher than Perplexity Search’s and 140% higher than Parallel Advanced’s on this test. Both competitors provide general web search rather than dedicated place databases, which limits the comparison. Evaluating production suitability also requires details about the query set, ground-truth construction, geographic coverage, data freshness, and test reproducibility.
One endpoint, several workflows
Exa positions Places for applications that already combine web research with local discovery:
- Personal assistants: Find restaurants, pharmacies, shops, or attractions while checking hours, status, and contact details.
- Sales research: Build lists of businesses by category and region, then pass available contact data into qualification or outreach systems.
- Market analysis: Count nearby competitors, map service coverage, or identify areas that lack a particular type of business.
A shared search endpoint reduces tool-selection logic inside an agent and removes a separate place-data adapter from the application stack. Teams can inspect responses in Exa’s search playground and use the existing API documentation for integration details.
Google Places and Yelp Fusion remain established sources for local-business data. Exa’s practical distinction is the combination of place records and general web search behind one interface. Applications that need reservations, transactions, provider-specific identifiers, full review content, or specialized geographic coverage may still require additional APIs.
Checks to run before production
Production adoption depends on behavior that a headline index size cannot capture. Developers should validate the following against their own markets and workloads:
| Area | What to verify |
|---|---|
| Freshness | How often hours, closures, phone numbers, and ratings are updated |
| Coverage | Result quality across required countries, cities, languages, and business categories |
| Identity | Whether records have stable identifiers and how duplicate locations are handled |
| Missing fields | Which properties may be null and how the API represents uncertain data |
| Usage terms | Rules for storage, display, attribution, and downstream use |
| Operations | Rate limits, latency, pagination, retries, and total cost under expected traffic |
For teams already using Exa, Places offers a low-friction way to test local search without adding another provider immediately. New integrations should compare coverage, freshness, terms, and total cost with dedicated place APIs before committing to a production architecture.
Comments
