Rate parity check on Google Hotels by name or URL. One row per hotel and stay: every OTA and official site side by side (Booking.com, Expedia, Agoda), nightly rate, free cancellation, link; cheapest source, spread, direct booking rate, undercut by whom and how much. minSpreadPct: violations only.
A hotel name never bills a look-alike. Names are matched the way google-hotels-prices 1.18 matches them: only the words that say which property it is count, not the words that say what kind of lodging it is (hotel, villa, inn, suites, resort…) nor the town Google searched; a property Google lists must carry all of them, one Google resolves by itself most of them (a city missing from its name does not count against it), and a name that matches no listed property, or several, is a free not_found status row whose error says why and how to name the hotel exactly. On 1.3.1 the invented "Zorblax Villa Legian" was billed as "Wyanna Villa LEGIAN" and "Kuta Beach Zimzam Resort" as "Kuta Seaview Beach Resort" (run kaSjvdxOwyJvLnBEh), "Hilton Lisbon" as a Curio Collection hotel (8tjQ1RaqlcOSZNI8u) and "Abia Villa Legian" as Wyanna (hM4CPiCvWHqa5DeBq); "The Plaza New York" was not found (6W33jX6AiwYRZ2Itf) and now is. A run whose every name is not found says so on its run card (No rows — not found: …).
A hotel is one property however it is given: by Google Hotels URL or token, by its Google Maps link or place id, or by a name that resolves to it. 1.3.1 billed Altis Grand Hotel twice in one run, by its token and by its place id (hM4CPiCvWHqa5DeBq).
Input fields this actor does not know are listed in SUMMARY ignored_fields and at the end of the status message (; ignored unknown input fields: …), also when the run stops on an input problem (SUMMARY status: invalid_input); they were dropped silently. A checkOut date (google-hotels-prices' field, or checkOutDate, check_out_date, checkout, check_out) sets the stay when nights is left at 1 and is recorded in aliased_fields; until now an agent that sent one was priced a one-night stay. Runs on the Google Hotels module of google-hotels-prices 1.18 (offer rows carry price_basis internally; the parity row is unchanged). No prices, defaults or output fields changed.
1.3 — 2026-11-18
No more timed-out runs (5 of 204 external runs in the 30 days to 2026-11-18 ended TIMED-OUT: the platform stopped them with no SUMMARY and no status message, and the rows delivered until then were charged). About 10% before the run timeout (at least 20 s) no new property and stay starts, in the last ~3% no request is sent, and in the last seconds anything still waiting on Google is dropped uncharged; the run ends SUCCEEDED with every row it delivered, time_hit and timeout_skipped in SUMMARY, and the status message ends stopped before the run timeout (time_hit): N of M jobs not run. A job is never dropped between its charge and its push.
Faster where runs were slow: a property whose rates name no official site (vacation rentals, managed apartments) is named by one small call per run (65-160 KB) instead of the 2.2-2.7 MB entity page on each stay (3 rentals × 7 stays: 48 s → 20 s, 15.5 MB → 2.0 MB); names are looked up side by side instead of one at a time behind a lock that every name's stays waited on. Runs on the Google Hotels module of google-hotels-prices 1.17: a proxy session takes the next call as soon as it is free (a slow IP no longer holds a queue), and a dead proxy (407, network errors) moves the run to RESIDENTIAL like Google's rate limit does.
free_cancellation in sources is now true/false where Google states the site's own terms for its lowest rate (about one site in three; it was always null), terms dated before today are dropped, and an official site Google lists without a name is named by its booking domain. No prices, input fields, defaults or output fields changed.
1.2 — 2026-11-16
Never billed past maxTotalChargeUsd: each charge asks the platform only for the events that still fit (apify SDK 4 answers a larger request by billing one event past the cap; a $0.30 cap was billed $0.302 on 2026-11-16), nothing more is requested once the spending limit is reached, and the run ends — spending limit reached. Google's rate limit is handled the way the Maps Actors handle it: a 429 from an IP that was answering holds every session and slows the pace, and after 3 straight refusals the rest of the run moves to Apify RESIDENTIAL (only when the input uses Apify Proxy without RESIDENTIAL or custom proxy URLs); SUMMARY.rpc counts throttled and escalated. No prices, input fields, defaults or output fields changed.
1.1 — 2026-11-16
maxItems (sent by AI agents) now caps the primary rows delivered and charged; add-ons follow the delivered rows. Here the primary row is parity: maxItems is one total over every property and stay, properties past it are not fetched, and a property that yields no parity row (filtered, unpriced, not found) does not use it up. When the total stops the run the status message ends ; maxItems N reached, and SUMMARY records max_items. No prices, input fields, defaults or output fields changed.
2026-11-13 — Restore previous listing keywords
Restore title, description and SEO text to the November 5 baseline at the owner’s request. Preserve current runtime fixes, input/output contracts, billing, prices and defaults.
2026-11-07 — Consistent cap and comparison examples
Align remaining schema summary, cost and FAQ examples with the reviewed result-order and direct-undercut contract. Missing comparison evidence remains unknown. Documentation only; runtime, field names/defaults, ranking metadata and prices are unchanged.
2026-11-07 — Agent result-contract clarification
Clarify result ordering, row relationships, cost limits and interpretation of the existing outputs for agents. Input descriptions and README only; field names/defaults, ranking metadata and event prices are unchanged.
2026-11-06 — MCP discovery metadata
Clarify supported search intents and output nouns in the Store listing. Runtime, input contract, output and prices are unchanged.
Changelog
2026-10-16 — Keyword Planner and locale documentation
Validated search terminology against Google Ads Keyword Planner and Store shelves. Added localized usage guidance and bounded examples; retained existing English phrases. Clarified locale limitations where measured. No runtime, output fields, defaults, or pricing changed.
1.0 — listing and input form reworded, 2026-10-11
No behaviour, output or price changed; the same inputs return the same rows.
The input form is reordered: hotelNames first (agents have names, not entity URLs), then hotels, the stay (checkIn, nights, sweepDays), minSpreadPct moved up next to the stay as the violations-only control, then guests, currency and market; proxy and pacing last. Nothing was renamed or removed.
Every input now says what to pass, with an example, its default and what it costs.sweepDays states the arithmetic (5 hotels × sweepDays 6 = 35 rows, $0.35); minSpreadPct says what 0 means, that filtered rows are never billed, and that it filters on the cheapest-to-dearest spread, not on the undercut — "only where an OTA undercuts direct" is a filter on undercut_by; hotelNames states the name guard (two shared words and 60 % of yours), that a city Google leaves out of the name counts against the match, and what to do on not_found; hotels explains how to check the same comp set every morning; country says it is the booker's market, not the hotel's; adults says Google prices one room and nothing above 6 adults.
Store description rewritten for agents: one row per hotel and stay, every OTA and the official site side by side with nightly rate, free cancellation and link, the spread, the direct booking rate and who undercuts it. seoDescription carries the synonyms and the German, Spanish and French nouns.
The pricing text states the per-1,000 price, that a row costs the same however many sources it holds (4 to 36, about 18), the rows-per-hotel arithmetic, what is never charged, that there is no start fee, and that a run stops at the maximum cost you set (maxTotalChargeUsd).
Dropped claim: the market (country) does not decide whether the displayed rate includes taxes — the parent module measured that the display basis follows the country of the proxy IP (pre-tax plus mandatory fees from a US IP, all-in from most others). country changes which booking sites are listed.
README: a section mapping common requests to the exact input, the row fields and the empty-result contract; the measured 2026-08-30 audit numbers (71 of 100 independent US hotels undercut); that Google does not list a hotel's official site on every run, so a missing direct rate should be confirmed on a second run; the sources-per-row range corrected to the measured 4 to 36.
2026-08-30
Reliability: a run that cannot act on its input — a URL from the wrong site, a name nothing resolves to, an empty target list, a number outside the documented range — now ends cleanly with the reason on the run card instead of as a failed run. Nothing is charged in either case; a blocked or throttled source still fails loudly.
Store listing: moved into the categories buyers actually browse (the old Business category has no Store page).
1.0 — 2026-08-29
First public release.
One parity row per property and stay from Google Hotels: every booking source's lowest rate side by side (OTAs and the official site) with source ids, stay totals, free-cancellation flags and booking links.
Parity math in the row: cheapest and dearest source, spread in percent, median, the official-site rate, whether direct is the cheapest, which sources undercut it and by how much.
Properties by Google Hotels URL or entity token (exact, one offers call each) or by name (one place search; kept only when the listing's name matches most of the query's words, so a lookup never bills a stranger).
minSpreadPct delivers only properties whose spread reaches a threshold — the violations-only feed — and what it drops is never billed.
Relative dates ("30 days", "6 weeks", today) and sweepDays for a week of check-in dates per property in one run; occupancy with children's ages; any currency and market.
Billing: pay per delivered parity row only. Unpriced properties, unresolved names, duplicates, filtered rows and status rows are free; each property and stay is billed once per run.
status row per property and stay (ok, filtered, no_rates, not_found, duplicate, unpaid, error) and a SUMMARY record with the run's counts, so a failure is visible rather than a silent zero.
Same fetching module as the verified Google Hotels Prices Scraper, watched by a daily canary run.