Why ChatGPT is building its own search index, codenamed Labrador

For two years, the common assumption was that ChatGPT simply relied on external search engines such as Bing and Google. That assumption no longer holds. Research by Peec AI indicates that OpenAI has been building its own search index since 2023, internally codenamed “Labrador”. And by now, Labrador appears to play a meaningful role in determining what ChatGPT serves up in its answers.

What exactly is Labrador?

Before explaining Labrador, it helps to start with the basics. A search index is essentially a vast catalogue of the internet. Think of the index at the back of a large book, except it contains billions of web pages. A search engine crawls the web in advance, stores what it finds and searches that index when a query is made, rather than scanning the live web from scratch every time. That is one reason results can be returned so quickly.

Labrador is ChatGPT’s own version of such an index. According to the research, it is better understood as a family of indexes: one for the general web, alongside specialised indexes for PDFs, YouTube, news, academic papers, Wikipedia, local businesses, finance, legal and medical information, shopping and images. Sound familiar? In broad terms, it resembles the way search engines evolved: a central web index complemented by specialised retrieval layers.

This does not appear to be a side project. OpenAI job postings have referred to work on indexing systems, retrieval pipelines and serving layers, as well as approaches that combine traditional lexical ranking methods such as BM25 with modern embeddings.

Why OpenAI was effectively pushed in this direction

During the Google antitrust proceedings, ChatGPT head Nick Turley testified that OpenAI had proposed a partnership with Google and that Google declined. OpenAI therefore had an incentive to develop more of its own search infrastructure. Turley said the company had aimed to answer a large share of search queries internally by the end of 2023, while acknowledging that full independence from external sources could still take years.

Fast-forward to today and ChatGPT is reportedly testing whether its own index can outperform scraped search results, particularly for shopping-related queries. For some users, parts of that proprietary index are already involved in certain searches. ChatGPT also caches web pages rather than fetching every page live for every request, helping the system respond quickly at scale.

Google and Microsoft have not been cut out of the picture

ChatGPT has not simply abandoned its existing sources. It can still draw on external search and data providers, including Microsoft infrastructure and specialised partners for particular types of information. The shift is therefore not an overnight replacement. OpenAI appears to be adding its own indexing capabilities incrementally, testing them alongside existing sources and expanding their role where they perform well.

For brands, that means there may be little visible change in the short term: websites can still be discovered through established search and web ecosystems. But the direction matters. The stronger OpenAI’s own index becomes, the less dependent ChatGPT may need to be on third-party retrieval sources.

What does this mean for your brand?

Visibility in Google or Bing should no longer be treated as an automatic guarantee of visibility in ChatGPT. If you sell products, it is worth monitoring how your products appear in ChatGPT’s shopping experiences, where OpenAI has been actively developing its own retrieval capabilities. For local organisations, accurate and consistent information across authoritative local platforms remains important because ChatGPT can draw on specialised data sources.

The broader SEO implication is that discoverability is becoming more fragmented. Traditional search visibility still matters, but brands increasingly need to understand whether their content can also be crawled, interpreted, retrieved and cited by AI systems.

The rules for visibility in AI-generated answers are shifting again.

Want to read more?

This article is a concise summary of a much deeper investigation. Credit goes to Peec AI, whose analysis examines the underlying search infrastructure in considerably more technical detail, including observed server-side behaviour. For the full technical reasoning, source material and screenshots, read their original analysis: “ChatGPT built its own search index”.