Try the Airbnb Markets Dataset Search API

- Grouping by `admin1`, `locality`, `room_type` or `property_type` returns enrichment-derived market cells (sub-country geography, or accommodation type). `admin1`/`locality` come from a reverse-geocode pass, `room_type` (`entire_place`, `private_room`, `hotel`, `shared_room`) is classified from each listing's search-card descriptor going forward, and `property_type` (Airbnb's finer canonical type) is read from the listing detail page. Because each covers a growing share of the census, these dimensions return an **empty `items` array until that coverage is high enough** for the cell counts to be reliable, then fill in on their own as coverage grows. `country` and `market` are always available. - `avg_person_capacity` is the average number of guests a listing sleeps, computed over the cell's detail-page-enriched sample; like the `*_enriched` fields it is reported only once that sample is large enough to be reliable. Combine it with `median_price_usd` for a rough price-per-guest read. - Pairing `group_by=room_type` with `median_price_usd` gives the median nightly USD price per accommodation type (e.g. entire homes vs private rooms), optionally filtered to one `country` or `market`. - `median_price_usd` is the cell's median nightly price converted to USD via an **approximate, dated FX snapshot**, so markets are price-comparable across countries despite Airbnb localizing the display currency. It is reported only for cells with enough USD-normalized priced listings and is a market-level approximation, not an exact conversion. - `avg_rating` is reported only for cells with enough rated listings, so a handful of reviews never swings a published market average; it is omitted otherwise. - `superhost_pct` is the Superhost share of the cell, from `0` through `100`. It is an **observed lower bound**: it counts listings seen carrying the Superhost badge on the search card, which renders inconsistently, so it under-counts. - `guest_favorite_pct` is the share of the cell's listings carrying the Guest Favorite badge, from `0` through `100`, with `guest_favorite_listings` the underlying count. Like `superhost_pct` it is an **observed lower bound** — the badge renders inconsistently on the search card, so it under-counts. - `superhost_pct_enriched` is the **reliable** Superhost share, computed over a **random sample** of the cell's listings enriched from the listing detail page (enrichment draws a representative sample, so the rate is unbiased even at partial coverage). It, along with `enriched_listings` (the size of that enriched sample), `distinct_hosts` (the number of distinct operators among the enriched listings) and `listings_per_host` (`enriched_listings / distinct_hosts`, an operator-concentration signal), is reported **only once the enriched sample is large enough to be reliable**; all four are omitted until then. Detail-page enrichment covers a growing subset of the census, so these fields appear for more markets over time. `distinct_hosts` is an aggregate count only — no host identity is ever returned. - `guest_favorite_pct_enriched` is the **reliable** Guest Favorite share over that same detail-page-enriched sample (the listing detail page carries an authoritative Guest Favorite flag, unlike the inconsistent search-card badge behind `guest_favorite_pct`). It is reported under the same gate as `superhost_pct_enriched` and omitted until the enriched sample is large enough. - The `min_listings` field in the response echoes the effective suppression floor that was applied. - The maximum result window is `10000`; `page * page_size` must not exceed `10000`. - Invalid enum values return the standard invalid params envelope. - Returns an empty `items` array (not an error) when no cell clears the suppression floor. - Does not trigger live scraping.