With the rise of Large Language Models (LLMs) like ChatGPT and Bard, an increasing number of internet folk predict the end of search engines like Google and Bing. It is quite natural in any civilisation for newer technology to replace older technology.
But when it comes to search, things are not so straightforward. Neither are LLMs nor their implications linear. I would like to start with the prime reason behind the prediction that search is about to die due to LLMs.
How Are LLMs Killing Search?
Contrary to LLMs and their implications, the reason why internet-folk predict LLMs will kill search is quite simple: the proof is in the pudding.
Ever since ChatGPT 3, more and more users looking for information have been asking their questions to LLMs rather than search engines like Google.
Part of this issue arises form Search Engine Optimisation (SEO). Explained shortly, SEO is the process of manipulating the contents of a web page to “game” a search engine’s algorithm such that the web page is ranked higher by the search engine.
What happens when everybody tries to game a system? Not much good. Today’s Google search results, for instance, are filled with bloated and spammy web pages. I often find myself landing on pages that feature 1500-word articles filled with nonsense for a simple yes-or-no question!
Contrary to this ordeal, LLMs often provide relatively simple answers to questions. When you look at it this way, it feels that it is indeed quite natural that LLMs will eventually replace search. However, this is only part of the story; the facts make the playing field more complicated.
The Facts
For starters, ChatGPT set a growth-record for any consumer application by reaching 100 million monthly active users in just two months since launch.
For comparison, it took TikTok 9 months and Instagram 30 months to reach the same number.
This essay is supported by Generatebg
However, as of November 2023, ChatGPT’s entire site traffic was only 2% of Google’s (the world’s top search engine with over 91% market share) web traffic. Furthermore, ChatGPT’s month-over-month user growth seems to be flattening over the past months (source: Similarweb).
We have facts tugging on both sides of the argument. What now?
Well, here is the least objective yet most interesting fact of the lot: ChatGPT has replaced 90% of my own search engine queries. Yet here I am arguing that search is here to stay. Why would that be?
Why Search is Here to Stay — Argument by Contradiction
For the sake of argument, imagine a future where LLMs give you all the information you need.
This means you don’t need to visit a search engine anymore. All the information you will ever need will be available without the hassle of SEO-bloated, spammy websites. But we are missing something here.
You don’t use search engines just for raw information. You also use them to find “services”. Imagine that you had forgotten all about the internet (except access to an LLM) and are new to it.
How would you “find” a product/service you are looking for? Sure, you could ask the LLM. But the LLM would only give you indexed information. For the LLM to give you up-to-date information, it has to keep indexing regularly.
In other words, the LLM will have to serve as a wrapper for a search engine underneath it.
Disovery and Visibility are Part of Human Civilisation
When an LLM starts to serve as a wrapper for a search engine, things start to get even more complicated. What the LLM does is synthesize the indexed and ranked information provided by the search engine.
This means that all websites that currently work as “information as a service” will likely die off. Of course, they won’t approve of this. They will likely stop allowing any sort of indexing/crawling of their websites and advertise their services heavily in other niche channels (dominantly social media). Note the word “advertising”.
This is another phenomenon that CANNOT be killed. Moreover, as the “search” experience improves with LLMs, advertising will evolve into more targeted and efficient streams.
Short history of search: Local information brokers turned into yellow pages. Yellow pages turned into online directories. Online directories turned into search engines.
Short history of advertisement: Print ads turned into radio ads. Radio ads turned into TV ads. TV ads turned into internet ads.
My Prediction: At the limit, LLMs will not replace search engines; they will merge with them to evolve and improve the “search” experience. Discovery (search) and visibility (advertising) will continue to evolve and thrive as long as advanced civilisation continues to exist.
If you’d like to get notified when interesting content gets published here, consider subscribing.
Further reading that might interest you:
- The Fractal Tree
- Web Development: How To Do Animated Recursion
- Hardware: The Unsung Hero Of The Tech World
If you would like to support me as an author, consider contributing on Patreon.
This is where retrieval augmented generation (RAG) architectures in the LLM space may play a hand – LLM model is trained and implemented once, but then it augments inference queries via current indexes.