What the assistants recommend to a Spanish hotelier

Before asking the first question we had written a list of 20 hotel software vendors. 17 of them came out: only 3 names on our list were never mentioned by any assistant, and in return the answers brought in 7 more that we had not put there.

There were 10 questions, in Spanish, the kind somebody running a hotel would ask: which program for a thirty room property, which one issues the Spanish electronic invoice. Each one went to 3 assistants, once. Of the 30 readings attempted, 29 went through, and the last one was lost to a dropped connection.

On who belongs at the top, the 3 assistants do not agree. We took the first three names from each, expecting at least one matching pair, and no group matches another. Only one name comes back in all three: Cloudbeds.

They read different places

The sources they cite barely touch either. Between two assistants the overlap runs from 0.08 to 0.18: of every hundred sites cited by either of a pair, only eight to eighteen are cited by both. The sites cited by all three come to 2.

The distinct addresses are 49 for the first assistant, 22 for the second and 94 for the third, which on its own cites more than four times as many as the second. ChatGPT mostly cites the vendors' own sites, while Perplexity goes to the Spanish software comparison sites. There is no single place to show up in.

The threshold we had written did not hold

Before spending we had declared that two assistants out of 3 agreeing would be enough. On the first run, though, one of the three answered none of the 10 questions: the credit on our account had run out.

With only two assistants that threshold does not get harder, it stops making sense, because the possible pairs drop from 3 to 1 and that single pair would have to match three names out of three. We topped the account up and redid those 10 readings. Only those.

We decided a second thing afterwards. We added the 7 extra vendors by reading answers we already had, so for two assistants out of 3 counting them proves nothing: they come out because that is where we took them from. The count that decides stays on the 20 written beforehand. On the assistant that had not answered that day, and that we therefore could not read, 6 of the 7 come out anyway.

What this does not say

We are not saying one vendor is better than another. We counted how many times a name shows up in an answer and nothing else, without trying any of these programs, and 10 questions asked once are not the Spanish market, nor one week of it. We chose the list ourselves too. Of its 20 names 3 never come out, and that says something either about them or about our list: from here you cannot tell which.