Most place searches inside an agent return prose. The model then has to extract the address, hours, and phone number from free text, and any mistake in that extraction becomes a wrong recommendation or a dial to the wrong number. On October 5, 2026, Exa addressed this directly: Exa Places adds 211 million places to its index, and results come back with structured fields rather than sentences to parse.
What you get in one call
The index covers restaurants, hotels, retailers, healthcare providers, cultural attractions, parks, and other local businesses. Each result includes names, categories, addresses, coordinates, opening hours and operating status, phone numbers, websites, and ratings. Exa also says it understands geolocation in the query, so “ramen in the SF Mission open Friday at 7pm” returns relevant businesses in that area instead of a generic list.
Two details matter for builders. First, Exa Places is available across all its endpoints with no additional charge. Second, the fields arrive in the same search call the agent already makes for web results, which means an assistant can check whether a pharmacy is actually open before recommending it — no second integration, no scraping pipeline.
The claimed quality gap
Exa ran a comparison across 400 queries, measuring F1 for businesses returned with correct metadata (hours, phone numbers, location). Per the company’s numbers, Exa Fast scored 0.800, versus 0.529 for Perplexity Search and 0.333 for Parallel Advanced — a 51% higher F1 than the closest competitor. These are vendor-run benchmarks, so treat them as a starting point rather than proof; the sensible move is to run a handful of your own queries through each provider before committing.
That said, the failure modes Exa describes are real regardless of who wins the benchmark. Missing a relevant place narrows what an agent can even consider, and a wrong address makes an otherwise good recommendation useless. For place queries, correctness of metadata is the product.
Where this fits your stack
Three use cases stand out from Exa’s own write-up:
- Personal assistants — verify hours and status before recommending, in a single call.
- Go-to-market teams — pull current lists like every dental practice in a metro, with phone and website fields ready to feed an outreach workflow. With Exa Agent Ultra, the team says a list can return every single match.
- Market research — count coffee shops within a mile of a site, or find towns with no urgent care clinic, without hand-building the dataset.
I’ve argued before that agents need search before they start guessing URLs — see give your agents search before they start guessing URLs. Exa Places extends that idea from documents to the physical world, and the structured fields are the real unlock: your agent gets data it can act on directly instead of a page it must read.
A grounded next step
The supplied material doesn’t cover pricing tiers beyond “no additional charge,” regional coverage depth, or how fresh the hours data is — all of which matter for a booking or logistics flow. Start small: wire a dozen representative queries through the developer dashboard, compare the structured output against ground truth for your cities, and only then decide whether it replaces whatever place lookup your assistant does today.