Creator Profile Avatar for Malte Landwehr
Malte Landwehr
Creator Profile Avatar for Malte Landwehr

Malte Landwehr

17.4K total followers

Latest posts

ChatGPT uses Bing. That is what I read on X all the time. And it makes me cringe. Technically it is not wrong. But it is also not the full truth either. ChatGPT is using Bing, yes. But also Google. And their own index. And it is actually not a single index. They have Labrador for general web documents. And a couple of specific ones. One for PDFs. For Youtube. For Wikipedia. For arXiv. For local, finance, and a few other verticals. 1️⃣ How do we know? ChatGPT exposes this in their server-sent events (SSE). Similar to how they tell us they use BrightData, OxyLabs, and SerpApi to scrape Google Search, Shopping, and Maps. Bing Search and Bing News are used as well. But directly without scraping providers. There is other evidence. Nick Turley (Head of ChatGPT) testified under oath that OpenAI started working on their own index in 2023. For HIPAA-accounts, ChatGPT explicitly mentions that all grounding is done via their own index. (ht to Ryan Edwards) Job postings from ChatGPT regularly include experience in running indexing systems and retrieval pipelines. This is not needed for a simple Bing-wrapper. 2️⃣ What does this mean? Good news: you can be visible in ChatGPT without being visible in Google and Bing. Bad news: for maximum visibility in ChatGPT you need to figure out Google, Bing & Labrador. 3️⃣ One last thing. I use the term “index” loosely here. You could also call them “retrieval systems with very long cache times”. But the caching is severe and goes far beyond a simple scraping wrapper. I have seen ChatGPT cite Wikipedia articles and PDF files 6 months after they have been deleted. To me, that qualifies as an index. You can find all the details in in the Peec AI blog.

X (Twitter) · 3 wks