Hey there, Am I the only one experiencing that the GPT- 4o web search model is way worse than what OpenAI is offering in ChatGPT search? I mean, it’s not even close. For example, this prompt: “Write me an up-to-date assessment of the DAX high on May 21, 2025.”
ChatGPT result: 
API result: 
Is ther e a way to improve this?
1 Like
Any idea? Would be grateful for some help. Results are not even close.
Have you searched the forums. We’ve had this conversation a few times! 
Yeah, I did. But mine is specifically about search results from GPT-4o search preview.
1 Like
Sorry, it’s not something I’m super familiar with. Apologies if i came off a bit cheeky. It wasn’t my intention!
Hopefully another developer with more experience with search stops by.
That said, I’m fairly certain there’s a bit of “secret sauce” with ChatGPT that they keep under-wraps. One of the reasons their consumer product went from 0 to 100 million users in a month, I think?
Maybe it would be better to phrase the question of how you can improve your API results rather than lamenting the fact ChatGPT does something a bit extra?
Also, did something happen to your uploaded images? I can’t see them.
I don’t know how true this is but the rumor is that ChatGPT search has access to more websites due to contractual reasons. In any case, they’re not the same tool under the hood and prompt the LLM differently. It isn’t possible to get an exact clone of ChatGPT. As Paul mentioned, they like to use secret sauce that we don’t have.
After several “so-so” attempts (because domain is kind of precise) with search via API, I ended up hooking custom GPT + bright data API + jina.ai API.
Bonus, supports Google maps, reviews, related queries, etc.
Had to set-up cloudflare worker with express js server to wrap the bright data API (has to use proxy), as their REST API is not optimal, to access it via REST API, but definitely worth it.
1 Like