AI is Crashing the Search Party

I love a good underdog story: The Wright Brothers, Seabiscuit, Rocky Balboa. It’s inspiring when unlikely heroes punch above their weight class and defy all odds.

It extends to the business world too. One daring startup is stepping into the ring with a $2 trillion giant that enjoys a monopoly in its core market, a war chest of over $100 billion, and access to top-notch talent. It’s like volunteering for the Hunger Games.

Like Muhammad Ali, the 8-to-1 underdog who taunted his opponent and then-champ Liston, Perplexity is strutting into the arena to challenge Google at the game it plays best: Search.

Our reliance on Googling everything leads to over 8.5 billion searches every day, helping Google earn an impressive $175 billion from search ads last year alone. It's a beautiful business that grows steadily, fueled by our insatiable curiosity and rising ad spending to capture our attention. 

This ad revenue treadmill is Google's moneymaker. It has fueled Google's expansion and leadership in mobile operating systems with Android, internet browsers with Chrome, cloud computing with Google Cloud, and autonomous driving with Waymo.

Google Search is the foundation of Google’s empire, and Perplexity is plotting its downfall.

Why Now?

There's a graveyard full of companies that have failed to unseat Google over the past 20 years. But this time is different for two reasons. 

First, we're reaching a tipping point of consumer frustration. 

Charlie Munger put it well when he said, "show me the incentive and I'll show you the outcome". Google’s incentive to support advertisers with monetizable products like sponsored ads has compromised our search experience. Advertisers have also learned to manipulate Google's algorithms to get more clicks, lowering the quality of search results. 

The risks of an advertising-led model, as pointed out by Google’s founders Larry and Sergey in 1998, are becoming apparent. 

Second, we’re witnessing an AI renaissance. 

Advancements in large language models from OpenAI, Anthropic and Meta have made it easier to retrieve information at scale. These models have basic reasoning abilities and can understand the intent behind inputs to produce detailed, personalized, and multimodal results. 

AI has the potential to redefine Search, much like advancements in mobile in the 2000s created opportunities to build impactful companies like Instagram, Whatsapp, and Uber.

From Niche Data Tool To Giant-Killer

Perplexity burst onto the scene less than 2 years ago led by a rockstar group of machine learning researchers and engineers: Aravind Srinivas, Denis Yarats, Johnny Ho, and Andy Konwinski. Together, they’ve aggregated a wealth of experience from innovative companies, like OpenAI, Meta, Google, DeepMind, Microsoft, Quora, and Databricks.

Aside from possessing a winning bingo card of experience at leading AI companies, it is exactly the technically maniacal team that makes every VC trip over themselves with excitement. Will it be enough to unseat a giant? 

Perplexity’s vision has taken many forms. It started with the simple yet ambitious goal of helping developers query and code SQL in natural language.  

Their initial prototype, BirdSQL, was a search tool designed to find specific tweets from everyday life coaches to your favorite megalomaniac. But Aravind and his team realized it was too narrowly focused. They pivoted towards building an AI-native search engine capable of handling a broader spectrum of information.

And Perplexity was born.

Perplexity's mission is "to make searching for information online feel like you have a knowledgeable assistant guiding you." Unlike traditional search engines, Perplexity has a chat interface where users can ask questions to find relevant and accurate information. 

Perplexity vs Google: What causes a meteor shower?

The collection of blue links we’re so familiar with is replaced with paragraphs of clinically-worded, relevant prose aimed directly at answering our questions. It's a magical experience that eliminates the fatigue of clicking on links and digging for answers. 

Accompanying their answers are footnote links to websites, offering comfort that the AI models aren't hallucinating fiction over fact.  

Stealthy Intrusion

The trojan horse of Perplexity's approach is that it works beautifully for a special kind of search query.

Andrei Broder, a renowned scientist from IBM and Google, came up with a simple method to categorize search queries by intent. He identified three main types: 

1. Informational: seeking details

2. Navigational: looking for a specific website

3. Transactional: activities like making a purchase or using apps

Perplexity's approach excels at handling complex, research-based informational queries that are a challenge with traditional search engines. 

My refresher on what causes a meteor shower was more detailed and seamless with Perplexity than with Google. I also went back to Perplexity, not Google, to learn more, like “the best locations and times to view meteor showers”. I engaged more deeply. 

I’d argue that Perplexity's ability to handle complex queries and its frictionless user experience invites us to search more often, expanding the total pie of searches. In doing so, Perplexity is creating a unique wedge to crash the exclusive search party.

Now, not all search queries are created equal. Google’s most lucrative queries have transactional intent. A recent antitrust trial revealed that in a week in September 2018, Google made the most money from searches related to the iPhone 8, car insurance and cheap flights. Surprisingly, searches about meteor showers didn’t make the cut. 

Perplexity is just starting to tackle these lucrative transactional queries. Developing a mechanism and business model to tap into this search revenue will be key. Perplexity plans to let brands influence "related questions" under their answers, but it’s unclear how effective this will be. 

Where Are The Moats?

Perplexity is often described as a model “wrapper”, a thin layer of flair and functionality that spruces up the AI models. 

There’s appropriate tension around whether companies can build a differentiated and valuable “wrapper” business, as most of the “magic” comes from the underlying technology. Sure, the user experience is more engaging and friendly, but it can be replicated. In the context of moats around your castle, it's a narrow stream that’s easily crossed.

We’ve seen “wrappers” ascend quickly only to face a cliff-jump into the rocky waters. 

Jasper’s writing interface “wrapper” on top of OpenAI’s GPT-3 accelerated towards $100 million in ARR and achieved a whopping $1.5 billion valuation. But ChatGPT’s arrival, continuous innovation in AI models, and competition from other “wrappers” eradicated their early lead and any perceived moats from wrestling AI models into submission better than others.

The harsh reality is that most startups don't start with a competitive moat. Like underdogs who turn into champions, they develop it over time. Until then, one of the most impactful factors within a startup's control is how quickly they can innovate. 

Many believe that Perplexity’s relentless determination is fueling a rapid pace of innovation and execution, allowing them to outmaneuver competitors. They’re consistently rolling out new features and the latest AI models to improve their search experience. 

By innovating on a proprietary search index and custom models, Perplexity is seizing the chance to evolve beyond a “wrapper” and establish lasting moats. It’s still early days though, and they’ll need to demonstrate that this move will sustain their initial momentum. 

Google’s Dilemma

Google has controlled the search market for two decades, fending off fierce competitors. Remember Yahoo? 

Today, Google faces an innovator's dilemma. The fresh arrival of an agile, focused competitor, prompts an impossible question of whether to reinvent its user experience, even if it cannibalizes its existing business. 

Google’s response to AI-native search is a new experimental feature called Search Generative Experience, which provides AI-generated answers at the top of the search results. 

Google’s Search Generative Experience 

Despite its less catchy name and clunky user experience, it has become popular. Google revealed it has handled "billions of queries'' and noted "an increase in search usage among people who use the new AI overviews''. 

But Google’s reliance on advertisers limits its ability to innovate on user experience. It has to keep prioritizing search queries that are navigational and transactional. Sundar emphasized this balancing act in a recent earnings call, saying Google is "...being measured in how we do this, focusing on areas where gen AI can improve the search experience while also prioritizing traffic to websites and merchants."  

I don't expect Google to wow us with a better experience anytime soon.

And then there's the story of the unit economics of search. The search business lives and dies by the scale and profitability of its queries.

Vivek Goyal from Altimeter Capital knows a lot more about search than most of us do. He points out that traditional search generates remarkable revenue and profit on each query today. That's incredible for any business, especially one with Google's scale. In contrast, LLM-based queries cost substantially more and don’t generate as much revenue. 

With its upside-down economics, AI-native search is an awful pursuit. It's an impossible task for an underdog like Perplexity to build a profitable business with strong margins, and it's just as hard for Google to compete without cannibalizing its hefty profits from traditional search.

But I'm more interested in how the economics will shift over time.

Right now, the cost per search for Perplexity makes one question the point of trying. But the silver lining is that free, open-source models are improving rapidly. Perplexity is already using them to build its own custom models, which should help lower costs. 

Google has also seen significant cost reductions, thanks to advancements in hardware and engineering: ”...machine costs associated with SGE responses have decreased 80% from when first introduced in Labs driven by hardware, engineering, and technical breakthroughs."

But cost improvements don’t matter much if you can’t generate adequate revenue. And revenue on AI-native search is dismal today. My back-of-the-envelope math implies that Perplexity earns 1 to 1.5 cents per search, a meager amount compared to Google’s 4.5 cents per search and a fraction of the 3-30 cents cost of AI-native search. 

But there’s an opportunity for revenue uplift. Perplexity is rolling out a new, higher-priced Enterprise plan and introducing "sponsored questions" to open up new revenue streams. Google may also be considering subscriptions for its AI features. 

If these moves play out well, they can make AI-native search financially viable. Can Perplexity ramp up their game before Google finds a sustainable path to incorporating AI? It’s game on.

The Other Contenders

OpenAI is eager to join the battle. ChatGPT is a fantastic general-purpose text generator, but it lacks specific search capabilities that Perplexity and Google enjoy. OpenAI is reportedly building its own Search product, partly powered by Microsoft’s Bing search engine. 

Microsoft's decades-long history of competing with Google on search could be exactly the right investment and experience that Open AI needs. Understanding what doesn’t work can be useful when you’re taking aim at a giant.

One potential risk for OpenAI is their willingness - and even desire - to fight battles on multiple fronts - consumer, enterprise, hardware, and more. I’ve rarely seen companies emerge victorious when fighting multiple battles, but if anyone can manage it, it's the duo of Sam Altman and Satya Nadella.

Meta integrated its LlaMA 3-powered AI chatbot to the search bars of major apps: Facebook, Messenger, Instagram, and WhatsApp. It’s an exciting development, but it’s more similar to ChatGPT than Perplexity or Google. With ties to advertisers and over 3.2 billion daily users, I certainly wouldn't underestimate Meta’s challenge for the throne.

As LLMs improve and get cheaper, other well-resourced players like Apple may also charge the battlefield. Aravind has also mentioned Apple as a potential force to reckon with. 

I imagine the three-dimensional chess of competing for Search’s very own Game of Thrones is keeping Aravind and his team up at night.

Time will tell if Perplexity is able to build long-term moats and fight off the threat from well-resourced players, but it appears to be working for now.

Rapid Ascent

Perplexity has rocket shipped to over 50 million visits and more than 10 million active users each month. Its premium search subscription has generated over $20 million in ARR, more than doubling since the start of the year. 

In 2023, Perplexity handled over 500 million queries and is now serving around 169 million searches each month. It’s a far cry from Google's 8.5 billion searches *per day*, but quite impressive for a startup with fewer than 100 employees that didn’t exist a few years ago. 

I’m a big fan and power user of Perplexity, but I, like others, do wonder if it’s creating a loyal following amidst this rapid growth. 

AI apps are notorious for quickly attracting users as a result of their novelty and convincing demos, but struggle to keep users engaged and returning as often as more traditional consumer apps do. They’ve struggled to convert initial excitement into sustained habitual usage. 

Sequoia highlighted this challenge, pointing out that "users are not finding enough value in Generative AI products to use them every day yet. 

Despite questions around moats and other risks, Perplexity has quickly become the darling child within the tech and VC communities. While Google secured $25 million in its early days, it’s a different world today and Perplexity has raked in over $165 million from top VCs and successful founders/operators behind generational companies like Amazon, Meta, Y Combinator, Figma, GitHub, AngelList, YouTube, and HuggingFace. 

Word on the street is that Perplexity is raising another $250 million at a $2.5-3 billion valuation, which is certainly frothy. But unseating a giant will take significant capital and support. Even with this additional amount, Perplexity is arguably the least armored in a capital-intensive battle.

Search Wars Ahead

Perplexity has all the makings of a classic underdog contender, and you should try it next time you’re curious about the difference between cappuccinos and lattes, that biography on your idol, or the best time to visit Hawaii.  

Google's profits have never been more vulnerable since they entered the race back in 1998, and amidst all the fighting, it may never be the same.  

It's only a matter of time before an innovative and sustainable search experience emerges, and we may see several battles before a victor is declared in the greater war. 

We’re in the early innings of the Search Wars. But, one thing is clear: AI has the potential to revolutionize search, and it’s here to stay.