Google Hotels Data Scraping

Query context is part of the record, or nothing reproduces

A result here is produced by a search, and the search carries context beyond the words typed.

• The query itself, exactly as issued.
• The location it was issued from, which affects both results and currency.
• Device type, which affects layout and sometimes the offer set.
• Language and market settings.

Without all of it, an observation cannot be reproduced — and a competitor's absence from a result set is meaningless unless you know what was searched.

So query_text, search_location, device_type and market_settings travel with every record. This is the same discipline our Google Shopping page applies to product results, and the reason is identical: position without its query is not data.

And personalisation is a real limit

Results can vary with signals we neither observe nor want to. We collect as an anonymous visitor and state that the result set is one anonymous observation rather than a canonical ranking. personalisation_note ships with the batch.

Referred offers, seller mix, and free versus paid

The offers are referred

The module does not sell rooms. It displays offers from OTAs and hotel direct channels and refers the booking. rate_is_referred is a constant true, and the seller is recorded on every offer.

So a rate here is a seller's rate as the module displayed it, which is not always what the seller's own site shows at the same moment. Where you collect the seller directly, the difference is computed from paired records — never asserted.

Seller mix is the actual finding

Which sellers appear for a property, in what order, and whether the hotel's own channel appears at all. For a hotel group that is the distribution question, and it is more useful than the rates themselves.

seller_name, position and placement_type travel with every offer.

Free and paid placements both appear

The module carries both paid placements and free booking links. Where the distinction is visibly marked we record it; where it is not, placement_type is undetermined rather than guessed.

Inferring paid placement from position would be a model presented as an observation.

What we do not produce

• A canonical ranking. Every observation is one anonymous search from one context.
• A share-of-voice figure across queries we did not define. Share is only meaningful against a stated query set, and that set is yours to define.
• A paid-versus-free classification we inferred from position.
• Bookings, referral volumes or commission. Not published.
• Signed-in or personalised results. No accounts created.

On share of voice

This is the most requested derived figure here and it is only as good as its query set. We compute it against a query set you define and we record, with query_set_id on every figure.

A share-of-voice number without its query set is a number about an unspecified population — which is the same failure our coverage page describes about shares without denominators.