Whoever Controls the Standard Wins

Here’s what comes up on Perplexity AI when you search for “Bard Rebrands as Gemini”:

Gemini Ultra (the higher-end version above Pro and Nano) is Google’s most powerful large language model (LLM), and it supports multimodality (text, images, video, and more). When it was announced, Google even claimed it outperformed OpenAI’s GPT-4, which is widely considered the state of the art, so it has drawn a great deal of attention.

ChatGPT is known as the app that reached 1 million users in the shortest time (that title has since gone to Meta’s Threads, but I personally think that’s an unfair comparison because it piggybacked on Facebook and Instagram’s network effects). I believe ChatGPT’s explosive growth came from delivering the excellent GPT-3.5 LLM through a chatbot—an easy-to-use interface for everyday people.

Bard’s user adoption was so low that you could hardly find anyone who actually used it. Google’s decision to replace its AI brand Bard with Gemini, the model that underpins its LLM stack, is widely seen as an admission that Bard’s image—long perceived as weaker than ChatGPT—would be hard to rehabilitate. But when you swap out the AI model, something else changes along with it: a new user interface “designed to reduce visual clutter, improve readability, and simplify navigation.”

Google’s search engine, which today still boasts the highest market share, succeeded because users could type a query into a simple text box and get back the most relevant results. The PageRank algorithm played a major role in this. PageRank was a new approach that ranked web pages based on their importance, inferred from the number and quality of links pointing to them. The basic assumption was that more important websites are more likely to receive more links from other sites. This link-analysis algorithm was named after Google cofounder Larry Page, and it was the first algorithm Google used.

Today, many complex algorithms beyond PageRank influence search results, but Google Search no longer has the overwhelming superiority it showed in its early days. I mentioned Perplexity AI—a service that integrates LLMs with search to directly answer questions—right from the first paragraph to underscore that Google’s search experience is no longer the best available. (For more on the decline in Google’s search quality, “Google Search Is Dying” is a good read.) Even so, Google still commands around 90% of the global search engine market.

In 2023, Alphabet (Google’s parent company) generated $162.4 billion in ad revenue from Google Search alone—over 200 trillion won in Korean currency. Given that digital advertising is extremely high-margin (because the variable cost per search result is very low, the marginal profit is very high), it might seem strange that other tech companies haven’t rolled out competitive solutions to challenge what looks like Google’s “easy money” search business. In reality, Microsoft integrated OpenAI’s GPT-4 model into its Bing search engine and offered it for free, yet failed to gain any meaningful search market share. The only players that have managed to compete effectively in Google’s near-monopoly are Amazon (capturing shopping-related searches within its own platform) and OpenAI, which offers ChatGPT—an interface layered on top of GPT-3.5 and GPT-4. On the latest earnings call, Alphabet CEO Sundar Pichai declined to answer an analyst’s question about whether ChatGPT had reduced Google’s search query volume. Even so, as of the fourth quarter of 2023, Google Search revenue was up 11% year over year, underscoring that Google’s grip on the search market remains strong.

So why does Google Search still command a near-monopolistic 90% share of the search market, despite its relative decline in quality?

The year was 2004. The idea of “mobile” barely existed, and PCs were effectively the only way to access the internet. Microsoft had just lost a major antitrust case, but it still held a de facto monopoly in operating systems with Windows, and Internet Explorer (IE), which shipped with Windows, enjoyed a 95% share of the browser market. Around this time, Sundar Pichai, an engineer from India, joined Google as a product manager and took charge of the Google Toolbar, which let users access Google services directly from within IE. While working on the toolbar, around 2006, he had this thought: “What if Google built its own browser?” Then-CEO Eric Schmidt strongly opposed the idea, but Pichai managed to persuade Google cofounders Larry Page and Sergey Brin, and in 2008 Google released the browser we all know as Chrome. The rest is history. Chrome became one of the most successful products Google has ever launched, and it played a central role in Pichai’s rise to Google CEO in 2015.

Today, most of us (even if “we” in Korea still use Naver a lot) don’t type a search engine’s URL and then enter a query into its search box. More often than not, we just type the query directly into the address bar. And most people never bother to change the default search engine (unless you’re that nerd using DuckDuckGo). Chrome’s default search engine is—unsurprisingly—Google Search. And Chrome is the most widely used browser in the world.

Even though Google Search has relatively declined in quality, as long as it returns results that are “good enough,” people don’t feel compelled to switch. That’s because the switching cost involved is higher than you might think. At this point, the importance of standards should be obvious without further explanation.

Standards are so important that in 2023 Alphabet paid a staggering $48.9 billion to make Google Search the default on browsers other than Chrome, such as Safari. That’s about 65 trillion won. Put differently, from Google’s perspective, the costs it saves each year by owning the standard through Chrome amount to tens of trillions, if not hundreds of trillions, of won.

The United States, arguably the most influential country in today’s global society, provides the US dollar—the currency at the root of all economic activity—as the world’s standard. Entrepreneurs who have solved most of the problems they set out to tackle in business often end up entering politics, because aside from money, it is the law that sets the standards for a society.

Right now, LLMs like GPT-4, Claude, and Gemini Ultra—each developed with investments running into the billions—are competing and rapidly improving in performance. But over time, we can assume there will come a point when, from an ordinary user’s perspective, the differences in utility between models are no longer that significant. Much like how Google Search today may not deliver the absolute best results, yet most users don’t feel the need to change their default settings and incur switching costs.

OpenAI initially operated as a B2B company that provided large language models via API. But when it launched ChatGPT, a true B2C consumer product, its valuation jumped roughly 100-fold—from around $1 billion to $100 billion. Investors clearly factored in the possibility that OpenAI could end up controlling the standard interface through which people interact with AI.

In the coming AI era, the most important thing will again be to seize the standard. To do that, we’ll need a combination of a sufficiently powerful AI model on the backend and a friendly, intuitive interface on the frontend. Google’s decision to rebrand Bard as Gemini and roll out a new user interface “designed to reduce visual clutter, improve readability, and simplify navigation” is precisely about capturing that standard. Whoever controls the standard wins.