Learning Library

← Back to Library

LLM Search Disrupts Google’s Business

Key Points

  • Large language models (LLMs) are now delivering search experiences that can shift substantial value away from Google, offering ad‑free, highly actionable results.
  • Demonstrations with an LLM (referred to as “O3”) showed it can instantly provide detailed ticket information, flight options, booking strategies, and logistical tips—features that Google’s standard search and services don’t bundle together.
  • The “bias for action” of this model is unusually strong, delivering step‑by‑step recommendations that go beyond simple answers, making it more useful for planning tasks.
  • Although currently accessed via a paid Pro account, the presenter predicts that similar LLM capabilities will become widely available within months, threatening Google’s ad‑driven revenue model.
  • By handling diverse queries (sports tickets, travel, health advice) in a unified, strategy‑focused interface, LLM‑based search could fundamentally disrupt many of Google’s core search verticals.

Full Transcript

# LLM Search Disrupts Google’s Business **Source:** [https://www.youtube.com/watch?v=dlwODfg4eE0](https://www.youtube.com/watch?v=dlwODfg4eE0) **Duration:** 00:08:14 ## Summary - Large language models (LLMs) are now delivering search experiences that can shift substantial value away from Google, offering ad‑free, highly actionable results. - Demonstrations with an LLM (referred to as “O3”) showed it can instantly provide detailed ticket information, flight options, booking strategies, and logistical tips—features that Google’s standard search and services don’t bundle together. - The “bias for action” of this model is unusually strong, delivering step‑by‑step recommendations that go beyond simple answers, making it more useful for planning tasks. - Although currently accessed via a paid Pro account, the presenter predicts that similar LLM capabilities will become widely available within months, threatening Google’s ad‑driven revenue model. - By handling diverse queries (sports tickets, travel, health advice) in a unified, strategy‑focused interface, LLM‑based search could fundamentally disrupt many of Google’s core search verticals. ## Sections - [00:00:00](https://www.youtube.com/watch?v=dlwODfg4eE0&t=0s) **LLM Search Challenges Google** - The speaker demonstrates an ad‑free, highly actionable LLM‑powered search experience that provides detailed results like ticket information, arguing it could shift significant search value away from Google’s traditional model. - [00:03:56](https://www.youtube.com/watch?v=dlwODfg4eE0&t=236s) **LLMs vs Google for Complex Queries** - The speaker argues that large language models retrieve obscure, nuanced information—like identifying an 80s movie with a robot kid and SR‑71—more reliably than traditional Google search, and reflects on using longer, natural‑language prompts for AI learning and search integration. - [00:07:06](https://www.youtube.com/watch?v=dlwODfg4eE0&t=426s) **LLM Search Disruption Threat** - The speaker warns that advanced LLM-powered search could rapidly erode traditional search market share, posing a looming competitive threat to Google. ## Full Transcript
0:00All right, I want to do one of those 0:02rare Nate shows the screen videos 0:04because I want to talk about something 0:05that I think is getting solved that 0:07hasn't really been properly solved until 0:10now. We've talked about chat GPT and 0:14other large language models disrupting 0:16search, but I feel like we are at the 0:18point where you can reasonably shift a 0:22lot of the search value out of Google 0:24and that is a big big deal. I'm not 0:27saying tomorrow that Google has no value 0:29and I'm not saying Google doesn't 0:30respond. I'm well aware that the cost 0:33curve on serving LLMs is dominated by 0:36Google. Google has some very sharp minds 0:38working on LLMs, but the cost to disrupt 0:41their own core search experience is not 0:43something I'm sure they're willing to 0:44bear because as you can see here, this 0:46is an LLM search results page that has 0:49no ads. That does not suit Google's 0:51business model and yet it's very useful. 0:54I just threw a few uh queries we're 0:56going to run through. I'm a Michigan 0:57fan. Can you find me Michigan tickets 0:59for next season? I'd love to see the 1:01Ohio State game. So, I see the game 1:03time, the date, the venue. It's going to 1:05be at home at Michigan Stadium. It gives 1:08me the average price. It then sort of 1:11goes beyond that. This is a very 03 1:14thing to do. Tells me how to lock in the 1:16tickets. Gives me different 1:18strategies. Gives me tips to go to the 1:20game in Ann Arbor. gives me a pick list 1:23of next steps. I tell you what, 03 has 1:25the highest bias for action of any model 1:28I've seen. It's higher than my bias for 1:29action. And that is saying 1:31something. And I say, great. I love 1:34this. Uh, can you get me Smart Flights 1:35from Seattle for this game? What are my 1:37options? Now, this is really interesting 1:39because I'm deliberately trying to push 1:40it. I would not say this is as good as 1:42Google Flights by any means, but I think 1:45it's enough that it is going to disrupt 1:47some of the top offunnel search as chat 1:49GPT awareness begins to spread. You have 1:51to realize, you know, I get 03 now 1:54because I'm paying for pro, but in 6 1:56months, everyone's going to have lots 1:57and lots of O3. And this kind of 1:59experience is going to be possible for 2:01everybody at 2:03scale. And with that in mind, this kind 2:06of strategic here's your options to 2:09think about is super useful and frankly 2:11in some ways easier to sort through than 2:13the wall of flights that Google 2:16provides. Again, no ads here. deeply 2:19disruptive of Google's business. So then 2:22it gives me the smart booking moves. It 2:23gives me the strategy. Google doesn't do 2:25any of this. It gives me ground and 2:28logistics tips. Again, Google's just 2:30trying to sell me rental cars to to buy 2:33at this 2:34point. Um, and then I I deliberately 2:37switch it up. I'm like, "Okay, let's try 2:39a health one. What are some quick 2:40exercises I can do to improve my back 2:43health?" Because this is one of those 2:44things where like I don't just want to 2:46fall into the trap of being like, "Oh, 2:47yeah. booking flights. That's why people 2:49use Google. People use Google for lots 2:51of other stuff. So, I wanted to give a 2:52real rant here. So, I think the the 2:55weakness here where I think that OpenAI 2:57can and probably will improve is there's 2:59no like demo video, no diagram. Now, 3:03Google's a bit disorganized, but I will 3:05say their image search and so on will 3:07get you diagrams of all these exercises 3:09in the first click. And that's something 3:12that's missing here that I anticipate 3:14OpenAI is going to close, which is the 3:17other reason to do an exercise like this 3:18periodically. It basically gives you a 3:20cheat sheet into what OpenAI is going to 3:23be thinking about from a product 3:24perspective because Sam is very vocally 3:26going after Google's cake um and wants 3:29to grab that search 3:31spot. But even without the without the 3:34diagrams, this is already very useful. I 3:36have clear names. I have reps. I have 3:39quick form cues, which is actually very 3:41difficult to get on Google. I have a how 3:44to use it in one clear place. It's that 3:47clarity of like a single answer that's 3:49opinionated that I think that Google is 3:51really missing. And I know they're 3:52trying it with the Google summaries. 3:54It's just not as 3:56effective. All right, I now go to the I 3:59can't remember this and it's complete my 4:01knowledge kind of query. This is my 4:03favorite version. Uh, I can't remember a 4:05movie I was watching. It had a kid in it 4:07that was like a robot. He flew in like 4:09an SR71 or something across the US and 4:11the kid was part human. It was really 4:13old from the 80s. It's Daryl. D A R Y L. 4:18Um, and yes, I did remember the plane 4:20correctly, which shows you how much of a 4:21nerd I am. Uh, it just retrieves it like 4:24that. This kind of query is kind of 4:27funky for Google to retrieve. Sometimes 4:29you get it, sometimes you don't. I do 4:31get better results reliably with LLMs 4:33because it just pattern matches across 4:35their pre-training data with natural 4:37language in a way that like the standard 4:39Google search just doesn't do as well 4:41typically. Then moving on, last query I 4:45think I threw up here. What's my best 4:48approach to getting into the whole AI 4:50vibe coding space if I am a 4:52beginner? Now, you'll notice as I go 4:55through here that I am using what I 4:57would call Nate prompts. I'm doing a 4:59little bit more language here. I It's 5:02hard for me to stop myself doing that at 5:04this point. Like I know how LLMs work. I 5:06know that works better as a query. The 5:08jury is very much out on whether people 5:12are going to 5:13tolerate longer queries to get results 5:16that are more meaningful to 5:18them. I know that people are using Chad 5:21GPT a lot. My guess is that is going to 5:24bleed into search as openAI is able to 5:26make search more relevant and meaningful 5:28and seamless a part of the LLM 5:30experience. As an example, I get people 5:32who are asking 5:34me why does chat GPT not understand chat 5:37GPT's own plans? Well, the answer is the 5:40pre-training data doesn't include 5:42current models and it would have to 5:43search. But that's a very easy search 5:45thing and it would feel really seamless 5:47and it would be one of those examples of 5:49search being really useful in the model 5:51for people who are typing very quick 5:53queries like what's in chat gpt 5:56free. All right, moving on. Uh it gives 5:59me a practical path to coding. It talks 6:02about the stack. This gets in over my 6:04head. So let's just assume I am much 6:08more of a beginner than all of this. 6:11Um, so I'm going to actually do this 6:13live here. Let's assume I am 9 years 6:17old. Uh, please dial back and give 6:22me a 6:24beginner friendly approach. Now, this 6:28gets at the one weakness that I think 6:30they're going to have to 6:31fix. This is slow. It's good search, but 6:35it tends to be slow. This is the fastest 6:37response I had in this whole query 6:38string. And I think it's because it's 6:40reasoning off of a pre prior response. 6:41So it's 6:43quicker. So it gives me like specific 6:47recommendations. 6:49Um which I think are really great 6:51actually. Like it's super helpful. Might 6:52use this with my kid. Um gives me 6:55support and safety. Encourages me to 6:57have a grown-up, which sounds great. Um 7:00I love this tip. If something feels 7:02confusing, say pause and ask for help. 7:04That's what real engineers do every day. 7:06I think that's just a fantastic little 7:07moment. 7:09And uh it gives me encouragement that 7:11spending just 15 minutes really 7:13helps. So my point here is that search 7:18is at the point in LLMs where it 7:23is potentially disruptible if they can 7:26bring this kind of quality of search 7:28down across the free tier. And I want to 7:30flag it now. I want to get what your 7:32guys' thoughts are. I would be worried 7:35if I were Google. I don't think it's 7:36going to be easy to compete with OpenAI 7:38at this point. They have all the 7:41momentum. They have the ability to 7:43distribute to their whatever 6 7 800 7:46million I lose track. It's going to hit 7:48a billion by the end of the year active 7:50users. And that search experience if it 7:53starts to peel away search volume, it 7:55could look like a sigmoid curve where it 7:57peels away a little bit initially and 7:59then it just falls off a cliff. uh if 8:02they can actually crack the nut on 8:04product adoption for basic everyday 8:05searches which I guarantee you there are 8:07teams at OpenAI working on. What's your 8:09thought? Are we at a point where we are 8:11close to a tipping point on