It’s likely OpenAI will build a search engine

Full disclosure: this is just a prediction – I don’t have insider knowledge on this – but I’m more and more convinced that OpenAI is going to build its own search engine. Probably not a public one, but something internal that gives them reliable, up-to-date search results.

In our testing, we’ve found that ChatGPT very, very frequently pulls in search results from Google and Bing. The queries that it searches can be seen in DevTools:

When I prompted ChatGPT for “What are some good short haul destinations in October that will still be sunny?”, we can see that it searched the internet for “Europe sunny destinations October weather”.

This is how ChatGPT gets around the fact that their models are trained on data that can be a year or more out of date. If it needs fresh information to answer a question, it’ll have to get that information from the internet by searching.

And if you look in the citations – the list of pages that it used to inform it’s answer – we can also see just how strongly ranking well in Google helps. For this prompt at least, it’s incredibly likely that it pulled all of the answers from Google’s search results (but I have seen it do this with Bing too).

Most of the pages it used to answer the prompt rank on page 1 of Google for that query. All of them rank in the top 100.

It’s easy to imagine OpenAI being nervous about that dependency on Google and Bing. Companies really don’t like relying on third parties for core product functionality. If there’s an outage, or if access is removed, it creates a downstream impact you can’t control. And when your product is being used by hundreds of millions of people, that’s a significant risk.

And that’s especially true when the third party is Google – a direct competitor to ChatGPT. Google’s actively trying to prevent scraping, they’ve disabled access to certain search parameters (like num=100), and they’re hiring engineers to protect their results. They’ve been heading in this direction for a while. It’s not hard to imagine a future where they manage to cut off OpenAI entirely. And, possibly – although hopefully not – all of the SEO ranking providers might be collateral damage here.

If ChatGPT can’t access live search results, it falls back on a model with a knowledge cut-off that’s often a year behind, which makes the product significantly worse. It’s less risky with Bing, and I can’t imagine they will cut them off – they have a partnership with Microsoft – but I imagine they’d still prefer to not have to rely on any big third-party.

OpenAI has been hiring for search infrastructure roles, which suggests that’s the direction they’re going. And it makes sense – I bet it makes them nervous how reliant they are on Google/Bing. Search is a critical dependency, and it feels like it should be something they own.