Search is just a tool for LLMs. Ask Bing Chat about tomorrow’s expected Starship launch and it will do a decent job.
Likewise with 3rd party extensions; a good LLM will make the best choice based on criteria.
In a way the LLM becomes the end user from a search/advertising perspective. The actual end user just expects an answer, not a list of 5 ads and 10 possible places to go look for the answer.
So yeah, search as an activity still exists, it’s just much less monetizable because the value prop of LLMs is distilling all that noise to a useful answer.
‘A decent job’ is really subjective here. You don’t get to see the pictures from the article it’s summarizing, which may be ok for something as routine as a Starlink launch, but doesn’t help for a hurricane forecast where you want to see its path on a map etc.
I rarely just want an answer when I am doing a search. But when I do I generally care more about accuracy than I am willing to trust a LLM.
This whole product category is barely a year old and you’re pointing out its (real) limitations?
If we’re talking about the viability of the search ads business in the next few months, sure, those are good reasons people will stay with traditional search and therefore be monetizable.
But as an industry trend, I think search will be 100% LLM intermediated pretty quickly. Pics of a subject, projections of hurricane routes… I’d be very surprised if both aren’t provided by LLMs within 12 months.
But won’t LLMs be monetized in the exact same ways search has been? If you’re asking Bing Chat instead of searching Google for the best hotels in a city, you bet people are gonna wanna pay to make sure their name comes up.
If anything, I can see this poses more of a risk to websites than to Google. You’re still querying a knowledge graph representation of the internet, it’s only really the last step (clicking a link to a third party website from the SERP) that disappears.
AI can clearly move past these limitations, but it won’t be an LLM it will be some other technique.
If anything in the next 12 months I expect LLM’s to get worse in many areas as people start doing the equivalent of SEO to get LLM’s to say whatever they want. It’s GPT 5.0+ vs companies trying to sell stuff etc.
LLM kind of suck doing web searches, asking for SpaceX launch date is faster with google. Asking it for a stock market summary, it will suck hard comparing to just loading the first page of wsj. Most real time news tasks will suck vs Google, where it excel is remassaging info that it has been trained on like coding and how tos
I am not sure about stock market summary and similar stuffs. I created a stock earnings report summariser [1] using the recently announced GPT from OpenAI. It is pretty handy and cool. This is not something Google can replicate for everybody’s niche needs.
Likewise with 3rd party extensions; a good LLM will make the best choice based on criteria.
In a way the LLM becomes the end user from a search/advertising perspective. The actual end user just expects an answer, not a list of 5 ads and 10 possible places to go look for the answer.
So yeah, search as an activity still exists, it’s just much less monetizable because the value prop of LLMs is distilling all that noise to a useful answer.