OpenAI leans on Google Search data to power ChatGPT

There is nothing quite like using your biggest rival’s homework to beat them at their own game.

According to a report from The Information, OpenAI has been quietly relying on scraped Google Search data to train and power ChatGPT. The AI giant is reportedly pulling this information through SerpApi, a Texas-based web-scraping service, specifically to handle real-time queries around breaking news, live sports, and financial markets.

Bypassing Google's front door

This backchannel setup is especially ironic given recent history. Documents from the Department of Justice's ongoing antitrust trial revealed that Google flatly rejected OpenAI's direct request to license its search index last year.

Since OpenAI could not walk through the front door, it took the side entrance. SerpApi acts as a proxy, scraping search engines at scale. OpenAI was even briefly listed as a featured customer on SerpApi’s website before the reference quietly vanished. Other tech heavyweights, including Meta, Apple, and search rival Perplexity, reportedly use the service too.

Outside developers had already caught on. Former Google engineer Abhishek Iyer ran an experiment by creating dummy webpages that only existed within Google’s index. When he asked ChatGPT about them, the chatbot returned information pulled straight from those pages, proving that Google's proprietary search snippets were feeding ChatGPT responses.

OpenAI admits this internally. During court testimony earlier this year, OpenAI’s Nick Turley conceded that serving even 80% of their traffic from an in-house index was a lofty goal they were nowhere near reaching, calling a 100% self-reliant index not operationalizable.

Why it matters

For everyday users, this explains why ChatGPT Search often feels surprisingly competent on live topics. It is essentially piggybacking on the best web index in human history. That said, if you are conducting high-stakes research or digging into obscure queries, we think it is still safer to go straight to Google. An AI summarizing a third-party scrape is simply another layer where details can get lost.

For the industry, this is a glaring reality check. ChatGPT may have 700 million weekly active users and an estimated $10 billion in annual revenue, but Google handles over 5 trillion searches a year. Crawling, indexing, and ranking the entire web across decades is a moat that raw transformer compute cannot easily bridge.

We think OpenAI’s move is ruthlessly practical, yet undeniably fragile. Google is stuck in a corner: suing to shut down scraping APIs like SerpApi right now would give antitrust regulators ammunition to argue that Google is monopolizing search access. But relying on your competitor's proprietary snippets to build your rival product is a massive vulnerability. If Google tightens its technical defenses or licensing rules, OpenAI’s real-time answers could take an immediate hit.