Has Anyone Noticed ChatGPT’s Search is Getting Worse and Dumber?

I’ve noticed a serious drop in response quality since the new ChatGPT Search feature was introduced. Previously, I could request English-only information and get valuable, relevant answers. But now, even when I specify “search in English,” the responses are often in mixed languages or based on low-quality sources.

Instead of insightful, well-sourced information, the answers feel like generic, “content farm” material, without depth or authority. Many times, ChatGPT doesn’t pull from reputable or authoritative websites, and instead provides shallow SEO-style responses that don’t really help.

This is a major downgrade from what it used to be, making it extremely frustrating to get precise and valuable information. It genuinely feels like a bug, and I hope it can be fixed soon.

By the way, I am primarily a Chinese user and frequently switch between English and Chinese in my conversations. However, I’ve noticed that the search function now seems to limit English-based searches specifically, making it harder to get consistent English-only responses even when I explicitly request them.

4 Likes

It’s not just you. i am exclusively English speaking with paid subscription, and the moment they turned on web search functionality, ChatGPT response quality fell off a cliff. I assume it is much cheaper to run a web search than it is to generate an answer (meaning, it takes less tokens).

1 Like

Totally agree. The quality of responses when using web search is terrible. I used to use it all the time to replace google, but now I’m going back to google. Hope it gets better soon and the product managers are reading this :smiley:

1 Like

Google has indeed become faster to use. It’s a disaster. It’s been months since you wrote that and web search has only gotten worse. It’s not possible to use any longer. The search model that hijacks the conversation has no context, doesn’t understand the question, it’s unable to form a good search query and it’s unable to understand the results it get, all because it’s such a simple model it can hardly be called a model. ChatGPT is ruined because of it.