Your ChatGPT recommendation is fetched by two scrapers you've never heard of
A researcher read ChatGPT's own network traffic and found it pulls the open web through two commercial scrapers, Bright Data and Oxylabs — reading your hotel's site for the facts, and someone else's page for the recommendation
The advice you keep hearing. Fix your website. Add schema markup, clean up the copy, publish more, and you'll show up when a traveler asks ChatGPT where to stay. It's the standard guidance, repeated on faith. It aims at the one page ChatGPT barely uses to decide.
What the traffic shows. Suganthan Mohanadasan, who runs an AI-search agency, spent a few days reading ChatGPT's raw network traffic — the JSON it sends the browser underneath each reply — and logged about 1,240 source records. Every fetched source carries a field called result_source, stamped with one of four values: serp, labrador, bright, oxylabs. Two of those are companies. bright is Bright Data; oxylabs is Oxylabs — the two largest commercial web-scraping firms, and direct rivals. A separate capture tool documents the same field independently, so this isn't one person's misread. When ChatGPT reaches the open web to answer a lodging question, it reaches through those two scrapers. The fourth value, labrador, is a licensed tier: Reuters, the Wall Street Journal, Wikipedia. National newspapers. A hotel doesn't get in there.
The split that decides you. ChatGPT reads your own page for the facts — if it can parse them — and everyone else's page for the opinion. Your rate, your address, your amenities can come from your site, as long as they sit in plain HTML and aren't locked inside an image or a script. The recommendation is different. The sentence that says this is where you should stay gets read off other people's pages: Reddit threads, review hubs, the "best hotels in [city]" lists. Those are the pages the scrapers bring back. Your site supplies the facts. The verdict comes from somewhere you don't own.
What it means for your hotel. You can build the best hotel page on the web and watch it move nothing, because the recommendation was never sourced from it. What moves the recommendation is your standing on the pages ChatGPT reads for opinion — how often you turn up in reviews, whether anyone makes your case on Reddit, whether the city listicles include you. That's a different job than tightening a homepage. Underneath it sits a smaller, sharper one: if the facts you do control live inside an image or a script the model can't parse, ChatGPT can't read the single thing on your own site it would have used. Check that before anything else.
The catch. One researcher, one logged-in account, a few days of traffic, a sample tilted toward tech queries. The field names are firm — you only need to see a field once to know it's real — but any percentage or ranking is directional, and the system changes week to week. Read it as the shape of the thing, not the measurement.
The pages other people write about you were always going to carry weight. Now there's a field in the traffic that names which ones ChatGPT read — and the two companies it sent to fetch them.
Enjoying this analysis? Hospitality.today delivers daily insights on hotel distribution, AI trends, and travel commerce — straight to your inbox. Subscribe for free at Hospitality.today →