The search wars

Google and search have been synonymous for decades. In 2021, Google earned $209 billion in revenue from search. It accounted for 81% of their total revenue.

For the first time in a long time, someone has the ability to challenge the form. That someone is Microsoft (MSFT). A large part of this threat comes from MSFT’s investments in AI, and specifically OpenAI.

In this deep-dive, I’m going to dig into the dynamics of search and how MSFT might disrupt it.

The state of search

Google is clearly the market leader in search by a long way. The less obvious observation, which I hadn’t realised until writing this essay, was how that dominance has been gradually declining. It’s neatly correlated with the rise of Bing, Microsoft’s search engine.

image

How search works

Search engines have one purpose: to help you find the information you’re looking for.

They achieve this by “indexing” pages on the internet. Think of every page on the internet as a row in a spreadsheet. When you type something into Google, it goes into that spreadsheet and surfaces relevant rows.

Most search engines are free. They make their money by displaying ads alongside your search results. The better they understand what you’re looking for, the more relevant the ads. The more relevant the ads, the more likely you are to convert and make a purchase. Advertisers pay a cut every time you click on an ad.

Incidentally, some search engines are trying to flip this model. Neeva is trying to build an ad-free search engine using a monthly subscription fee.

The tech shift

To disrupt an incumbent with a large market share you need a shock of some kind. This shock is often a technology shift. The technology shift creates more value by improving something in the user journey.

The most recent tech shift was mobile. Facebook led the charge with social networking. Google also won with its Android initiative. Apple won with the iPhone. The value created in this case was access to information: before the shift, users needed to be in front of a computer to access information or connect with friends. Mobile completely changed this.

The next shift is artificial intelligence. Whilst many of us have been stunned by ChatGPT, Dall-E and Stable Diffusion, AI innovation started years ago. The real shift is that it’s become more accessible: it’s easier and cheaper to incorporate AI in your product.

What AI brings to search

OpenAI’s products have taken the world by storm. They released ChatGPT on 30 November 2022 and acquired 1 million users in a matter of days. For reference, Instagram is the only product that even comes close and they took months.

image

MSFT invested in $1 billion in OpenAI back in 2019. It will go down as one of the best investments in history. MSFT has the perfect stack to extract value from AI. They bought GitHub, which arguably provides the best training data for AI related to programming. Their cloud service, Azure, is the second largest cloud platform in the world. The biggest of them all is Microsoft Office - they have the distribution to deploy AI at scale for enterprises all over the world.

MSFT announced that it plans to integrate ChatGPT in Bing. Rumour has it that they will incorporate a chat like interface, and incorporate text to image generation using DALL-E. This should come as no surprise.

There are three clear improvements AI brings to the search experience:

  1. Reduces the time to answer: for the right use case, AI dramatically reduces the time needed to give the user a final answer. Every programmer spends time looking for answers to questions. I’ve tried this first hand and 9 times out of 10, ChatGPT gives me the answer faster than anything else.
  2. Memory: when I search on Google, I usually have to make several attempts. This is also the case with a conversational chatbot like ChatGPT. But the difference here is that you can guide your search through conversation. Because of the conversational interface, AI is able to use your previous query as an input.
  3. Generation: search engines provide help you discover existing content. With AI, users can create content — text, audio or video. I don’t even know if “search” engine is the right term anymore.

Despite all of the above, the road isn’t going to be easy for MSFT. Here’s why.

Search is Google’s battle to lose

Google has invested significant capital in AI over the last decade. It owns and operates AI products in more domains than OpenAI (see here for a full list). It’s just not as well known because they haven’t released these models publicly.

Most people put this down to Google being conservative. AI gets many things wrong, is biased and has baggage. Google is playing it safe because it wants to protect its revenue from search, public company status and keep regulators happy.

This is where OpenAI and Microsoft are playing the right card. Disrupting the mind share that Google has is non-trivial. Apart from the value that ChatGPT provides, it needed to plant itself as an innovation in users’ minds. And it did just that when it acquired 1 million users in a few days. Incidentally, Bing’s search revenue also doubled in 2022.

image

Winning in search, even with OpenAI’s capabilities, ain’t going to be easy for MSFT. Most of the problems I state below are AI problems. These problems are magnified when you use this technology for something as ubiquitous as a search engine:

  • The accuracy problem: AI sounds right even it’s wrong. This is dangerous with ChatGPT (e.g. I chose to medicate because ChatGPT said so). It’s even more dangerous when it’s part of a search engine.
  • Cost of querying: Querying with ChatGPT isn’t cheap. I’m not too concerned about this because this was also the case when Google launched search. The cost of querying will come down over time.
  • Losing advertisers: Microsoft has a sizeable search business ($11.59 billion in 2022). Introducing ChatGPT like technology is going to have an impact on advertising. I’ve not really processed what this impact is, but there’s a tradeoff here. Part of the reason ChatGPT is so good is because it reduces the “time to answer”. Intuitively, this would reduce the number of people who click on an ad. There’s probably some clever way to get around this (e.g. ChatGPT recommends products), but we are venturing into the unknown.
  • Bias: products like ChatGPT are based off algorithms that have been trained off public internet data: reddit posts, twitter, academic papers, blogs etc. They are subject to all kinds of bias. This has been proven by people prompting ChatGPT. OpenAI warns users about this when they use it’s APIs. This is likely to be magnified in a search engine setting. MSFT is familiar with this dynamic: in 2016, they released a conversational chatbot called Tay and had to take it down. Within a few hours of being released, trolls inundated Tay with abusive language. Tay responded by tweeting abusive content. They will need to think about how to get around this when they introduce conversational AI into search.

To close

Despite all of the risks and challenges, I remain optimistic about MSFT’s ability to challenge search. You cannot innovate without taking risks; it’s more important that you manage them well.

A technology shift is coming and I’m here for it. We could all benefit from a better search engine and have multiple options to choose from.