I paid Claude US$1,070 to read the worst reviews of 30k places to stay

关于

Hi HN, I'm Oriol, a solo developer in Barcelona. Before booking anything, I used to do the same thing every time: find a place with a great rating, open the lowest-rated reviews and read them one by one. Broken AC, thin walls, a bed as hard as a board, a shower that barely works. Very often the deal-breaker was right there, written by guest after guest, under a lovely 8.9 global score. OKstays does that reading for you. It scores 33,000+ places to stay in 95 destinations on how well you'll actually sleep and stay there, and ranks them for how you travel (light sleeper, remote work, family...). This month I re-scored 30,500 places from what guests wrote. For each one, Claude reads its newest and lowest-rated reviews, then a second search collects what about 7 other review and booking sites say. It scores 67 comfort signals from 0 to 10, each with a short reason you can read on the place's page. If a deal-breaker (dirty rooms, an uncomfortable bed, a poor shower, bugs) ends up at 6.0 or below, the place is left out of every ranking. One angry guest can't do that alone: it has to be a pattern. What came out (method and charts: https://okstays.com/stats/deal-breakers ): - 1 in 39 places to stay has a deal-breaker guests keep reporting. - Half of them come down to a poor shower (51%). Bugs are next (35%). - 92% of them still have a guest rating of 8+. Across all places, it's 97%. - Under US$100 a night, deal-breakers are nearly 4.5x as common as at US$400+. - Among hotels: about 1 in 5 in New Delhi, 1 in 19 in Mexico City, and 1 of 698 in Tokyo. The cost, because HN always asks: about US$1,070 in API calls, US$3.7 cents a place, all batched (first search ~US$1.3c, second search ~US$1.2c, scoring ~US$1.2c). Before spending it, I tested methods against places I had researched in depth. My old scores were off by about 1.1 points on average; reading reviews brought that to about 0.5 (on a different, larger reference set, so read it as a direction, not a precise gain). It's free, no sign-up, and it also works inside ChatGPT, Claude and Grok as a connector (okstays.com/mcp). It earns booking commissions, and hotels can't pay to rank or to be removed. Last month it earned its first US$8.81, 18 months after I started. That was also the first money I ever made online. Two things I'd love from you: pick a city you know well and tell me if the top 5 feels right. Thank you!

收录记录

1. Show HN

   2026年11月5日 · 1 次收录