24 comments

[ 2.4 ms ] story [ 23.1 ms ] thread
https://developers.google.com/custom-search/v1/overview

  Note: The Custom Search JSON API is closed to new customers. Vertex AI Search is a favorable alternative for searching up to 50 domains. Alternatively, if your use case necessitates full web search, contact us to express your interest in and get more information about our full web search solution.

  Existing Custom Search JSON API customers have until January 1, 2027 to transition to an alternative solution.
I was using it for an ai agent i built for my use as home assistant. It seems like many competing companies might be doing same.
Microsoft already did this with the Bing Search API back in August 2025, killed both the free and paid tiers with nothing that drops in to replace it.
Not affiliated - would Kagis search API be a drop in replacement?
All I'm sayin' is that Searx exists and can provide JSON.
SearX isnt developed anymore. The main dev created Hister to replace it.
SearXNG is though, and Hister is a different thing, not really a replacement.
(comment deleted)
Why would they do this as AI companies and agents have driven demand for other companies to want to integrate search into their products?
I commented the exact same thing. Couldn't Google save itself by becoming the search engine all of the AIs use? That's an area where I would imagine they have a massive moat & would be extremely difficult for startups to compete.
Why would you use AI to write this post? If you can’t bother to write it, why should I bother to read it?
We really are speed running any percent to the destruction of decent quality technology.
That's surprising to me that they would kill something like this off, because LLMs are going to need search APIs. Why wouldn't Google want to be the search engine all of the AIs use? While we (people) may not need search engines as much as we once did, LLMs probably do need search capabilities. I could be wrong, maybe reply if I am, but don't LLMs typically need to search for information? For example, when I ask ChatGPT some obscure question about my Behringer RD-8 drum machine, an LLM isn't going to know the answer based on its original training will it? (even if the manual was out in the public internet when the model was trained)? Under the hood its searching for information and then processing it extremelly quickly right? These LLMs don't just "know" the answer do they?
Anyone know if that will affect stuff like Duckduckgo search for its !g fallback or do they have custom deals?
Isn't that just a redirect to google.com? Shouldn't be affected
it won't, it appends your query like this: google.com/search?q={query}

you could code this yourself

Plenty of alternatives that scrape Google: SearchApi, Serper, cloro, etc.