How we rank hotels

The order is decided by an AI model (Claude, by Anthropic). It receives your trip details and the data of the hotels and gives each hotel an overall score from 0 to 10. There is no fixed formula: the AI weighs the hotels against your wishes, and how much a point counts can differ from trip to trip.

What the AI receives about your trip: the place, your arrival station or airport (if you enter one), your activities (with distance, and whether you spend time there or visit once), your budget, your preferred district, the kind of your anchor (for example a beach) with the distance that counts as close, your wishes as text, whether you want breakfast or free cancellation, and how you get around (public transport or car). It does not receive the name of your anchor, and no hotel names.

What the AI receives about each hotel: type of property, stars, guest rating, number of reviews (rounded), price per night, the nearest station with its distance (not for trips by car), the distances to the transport hub, to your activities in town, to your anchor and to the city point, how far it is from the point of your preferred district (as the crow flies, only if we can place the district on the map), whether there are too few beds for your group, whether breakfast is offered, whether free cancellation is known, and which of the facilities you asked for are confirmed. Confirmed facilities are Wi-Fi from the search and, derived from Agoda's hotel data file, sauna, pool, spa, gym, parking and hot tub. Of these, the AI only sees the ones you asked for. If an entry is missing, it counts as unknown, not as absent.

We give the AI these points:

• Wishes and location: how well the property fits what you entered for the trip. This includes property type, stars as a level of facilities, confirmed facilities and breakfast, as far as a wish concerns them. For a relaxation wish or a nature anchor, full service counts.
• Price against budget: within the budget, cheaper counts as better.
• Guest rating.
• Number of reviews, only as a signal of trust: very few reviews dampen, many lift barely.

Our instruction on weighting them reads (translated from German): "These points matter. Weigh them so that the final order fits this user best. What they explicitly ask for is decisive. The general quality of a hotel should still not be neglected in the order." Below 30 reviews the rating counts as uncertain. Without a budget, price does not count unless a wish names it (such as "cheap").

On wishes and location we also tell the AI: places where you spend your time should be within walking distance: your preferred district, proximity to your anchor and activities we classify as a place to stay around. For places you visit once, the connection counts: a station within walking distance of the hotel and the distances as the crow flies. Distance to the city centre only counts if you explicitly ask for the centre (for example "central" without naming a place).

There is no fixed weighting. What you explicitly ask for is decisive. How much price, rating and number of reviews count next to it, the AI decides for each trip, so we cannot state a fixed relative weighting. Even with the same details, a new search can give a slightly different score. What is fixed: the AI receives only the data listed above. It rates only the criteria listed above.

Equal scores are decided by a fixed rule: first hotels with at least 30 reviews, then the better guest rating, then the lower price (only with a budget), and last the platform's hotel number.

Without the AI a fixed calculation rule ranks instead: if a stop has at most four candidates, if our check of your free text flags it (then it does not go to the AI), if the AI fails or returns no score for some of the hotels (then for the whole stop), for the hotels beyond the most promising ones (see below), and for properties that only fill free positions (above the budget, lower-rated within it, conspicuously cheap). This rule multiplies the fit to your wishes (calculated by us from the same location data), the price against the budget and the guest rating; hotels with fewer than 30 reviews are dampened a little. Fulfilled wishes (breakfast, free cancellation, confirmed facilities, a leisure-friendly property for nature anchors) give small bonuses, missing ones never lower the score. Unlike the AI, it has a fixed order of its levers: fit, then price, then rating, then number of reviews. Within a stop, every hotel scored by the AI gets a higher score than every hotel ranked by the rule.

Budget and filling up: hotels above the stop's budget share are not ranked, and neither are conspicuously cheap places, below roughly a quarter of the budget share. If too few well-rated hotels within budget remain to fill all positions, we fill the free positions with hotels above the budget share. They come after the ranked ones and are labelled as such: well-rated ones first, lower-rated ones after them. Within each group, hotels whose net price (before taxes and fees) does not exceed your budget per night come first, in the order of our ranking, then the others, the cheaper one first. A well-rated hotel above the budget therefore takes precedence in the selection over a lower-rated one within it. We only rank lower-rated hotels within budget alongside the others if positions would otherwise stay empty; they then stand among the ranked ones, ahead of the hotels above the budget. If there is no suitable hotel within budget at all, we show only hotels above it: first well-rated ones that fit the location and your preferred area, the cheapest of those, and rank them as described here. If hotels within budget and above it do not fill all positions, the conspicuously cheap places fill the rest last, provided their rating reaches the minimum rating. They come at the very end, are labelled as such, and the place closest to your budget comes first. When filling, the order of these groups takes precedence over the limit per hotel chain (see below): a third hotel of the same chain comes before any hotel of a later group. Only hotels above a minimum rating enter the pool in the first place (the threshold adapts to the candidate situation).

About the order of the list itself: the top positions follow the score, with additional gates (requested amenities confirmed, no hostels, campsites or private rooms, proximity to your anchor). The district wish influences the score through distance to its point; it is not an additional selection gate. We show at most two hotels of the same hotel chain, as long as others are available. Location and type of property take precedence over this limit, though: a third hotel of the same chain that is close to your anchor takes a top position before a hotel that does not, or a hostel, campsite or private room. If the pool is too small, we relax these gates step by step so no stop stays empty. If only a hostel, a campsite or a private room would then remain within budget, a place just above the budget comes first if there is one, labelled as such. Capsule hotels count as hostels. The last of the ranked positions is deliberately reserved: it shows the cheapest place rated 8 or higher that costs no more than the cheapest position above it rated 8 or higher, even if its score is lower; places with at least 30 reviews come first. A hotel, guesthouse or apartment takes this position ahead of a hostel, a campsite or a private room if it costs at most 20 % more. If there is no such place, another ranked position takes its spot. Only if no position above reaches a rating of 8 does it simply show the cheapest. Either way the choice is made among the properties that do not already hold one of the positions above. The limit per hotel chain applies to this position too. If you write “no hostel” or “no camping” in your wishes, we leave these places out, unless a stop has nothing else; then we show them with a note. Per stop we show at most one airport hotel and label it. A wish for luxury or five stars counts like a facility wish: places with 4.5 stars or more come first. When we show several booking platforms, this applies per platform group, and the groups appear one after another. Hotels far outside anchor or town limits, or without a valid price, are removed entirely up front.

The AI scores the most promising candidates per stop (currently up to 30, with deep research up to 60, and up to 90 when properties from our quality list are among them). We pick them by our rule-based fit and the guest rating; hotels with fewer than 30 reviews get a deduction. If properties from our quality list (see below) are available, some of the places come from the booking platform's search and the rest from that list; if one side has fewer, the other fills up. Within the quality list, instead of the plain guest rating we use a rating that also takes the number of reviews into account: the fewer reviews, the closer it moves to a middle value of 8. The AI also classifies whether an activity is a place where you spend your time or a destination you visit once, and it phrases the one-line reason on the ranked cards. Our score does not include hidden criteria such as commissions or platform affiliation.

Which properties enter the scoring at all is determined by the booking platform's sort order and by our own selection from the platform's hotel data file (a city's quality list and properties around your places). The standard search uses one sort order; deep research merges three, one of them selected by your travel party. For a place that Agoda lists as a city of its own, we also pick the best-rated properties of the whole city from Agoda's hotel data file (100 with the standard search, 200 with deep research, leaving out those the first sort order already delivered; a property that one of the two further deep research sort orders also delivers counts towards the quality list), by rating and number of reviews only, and query them for availability. This does not apply with a preferred neighborhood or a place under “Hotel should be close to”. If Agoda does not list a place as a city of its own (such as Coron or Ubud), there is no sort order there: we then query properties within about two kilometres of the place from Agoda's hotel data file for availability. If the search around the places where you spend your time finds properties, they take the place of the quality list, unless the stop would then have nothing suitable. In addition, we search ourselves around the places where you spend your time (district, anchor, activities classified as such): hotels within about one kilometre from Agoda's hotel data file, hotels with at least 30 reviews first, then by rating, which we then query for availability. They are scored exactly like all others. We cannot warrant that the booking platform's pre-selection is free of commission influence: for the order of its own results, Agoda itself discloses that the money it earns is one of the factors. We pick the quality list ourselves, by rating and number of reviews only, without any link to commissions. Our score only takes effect afterwards, on the properties that reach us this way.