ChatGPT has built its own search index

For two years the assumption was that ranking on Bing got you into ChatGPT. New research shows OpenAI has built its own index, called Labrador, and it's already serving most free users. Here's what it means for your visibility.

In this post

ChatGPT was widely thought to be using Bing for the majority of it’s web retrieval and searches. Ranking on Bing meant showing up in ChatGPT.

That’s now been thrown into question.

What's been found

Between May and July 2026, researchers at RESONEO and Peec AI captured the data ChatGPT sends to the browser and found a field naming where each result came from.

One value was "Labrador": OpenAI's own index, its private catalogue of the web. Others were Bright Data and Oxylabs, which scrape Google results on demand.

The numbers rule out Labrador being Bing with a new badge. Only 1.5% of Labrador URLs appeared in Bing's top 20 for the same queries, and where Bing cuts titles at 75 characters, a quarter of Labrador's run longer.

Peec also found it's a family of indexes covering the web, news, PDFs, YouTube, shopping and local.

OpenAI won't be surprised. In April 2025, ChatGPT's head Nick Turley told the Google antitrust trial that OpenAI had hoped its own index would handle 80% of queries by year end, citing "significant quality issues" with Bing.

So the timeline slipped, but the research seems to confirm that statement of intent from OpenAI.

Who sees it

This matters most for free users, the bulk of ChatGPT's audience. In free instant mode, Labrador supplies almost every citation on settled questions.

Paid "thinking" mode still pulls around three quarters of its results from scraped Google. Two very different systems are finding your pages, and the one most people use is OpenAI's own.

How Labrador reads a page

Per Search Engine Land's breakdown, the index stores your URL, your full title, and a snippet of roughly 200 characters lifted from the page body, starting at the H1.

Your meta description is ignored. A separate cache of full pages skips JavaScript and rejects anything over 4MB.

Encouragingly, small sites get the same treatment as OpenAI's licensed publishing partners. There's no velvet rope.

What to do

Put the answer in the first 200 characters after your H1. Write titles that make sense on their own. Keep image alt text short near the top, because it eats into that snippet.

Make sure the page reads without JavaScript, and allow OAI-SearchBot in robots.txt (it's separate from GPTBot, so you can welcome search while blocking training).

The wider point holds across ChatGPT, Claude, Perplexity, Gemini and Google AI Overviews: each engine is building its own view of the web, and ranking on Google or Bing no longer carries you in automatically.

If you'd like to see which of your pages ChatGPT cites, and which it skips, we can show you the data on a 20-minute call.

FAQ

Does ChatGPT still use Bing?

Much less than it did. Free instant answers come almost entirely from OpenAI's own Labrador index, while paid thinking mode leans on scraped Google results.

Will ranking well on Google get me cited in ChatGPT?

Not automatically. Labrador has its own crawl and snippet, so a page that buries its answer below the H1 can still be passed over.

How do I get into OpenAI's index?

Allow OAI-SearchBot in robots.txt, keep pages readable without JavaScript and under 4MB, and lead with a clear H1 followed by the answer. There's no submission form.

Back to blog