All the Answers

Humans will travel a long way for a good answer. 

The Lydian king Croesus, for example, sent seven emissaries to different corners of the region in search of the most accurate oracle. When the oracle of Delphi correctly surmised Croesus was making a lamb-and-turtle soup at the time she was consulted, the king lavished her with his formidable wealth. She was both deity and advisor, a possessor of mystical powers but capable of weighing in on such practicalities as military strategy. Even for the ancient world’s richest man, gathering the best information required a journey of many days. 

Such was the predicament of almost all of our ancestors. Our forebearers crossed oceans, climbed mountains, and braved disease to visit places of illumination. To perceive the wisdom held in the Library of Alexandria, to stare into the coy eyes of a guru, to attend an institution of higher learning – these were privileges paid for in distance traveled. 

Because of our drive for knowledge, the business of providing answers is often a good one. Though their names sound fusty today – redolent of mothballs and a lint-covered lozenge – the Yellow Pages and Encyclopedia Britannica were once famous enterprises. For several decades in the 20th century, both were household staples. 

For both companies, the fundamental value was the same: faster access to information. They may not have offered the numinous wisdom of a bearded prophet or monkish scholar, but for most matters, practical and transactional consumers were well-served. It represented a kind of compression: bundling the answers to our most common questions into convenient packages. Inquiries that once required a visit to the library or a walk to town could suddenly be resolved in minutes. 

From one vantage, this is the history of human information: the progressive shortening of the distance between question and answer. 

This process reached what looked to be its logical conclusion in 1996. That was the year Stanford doctoral candidates Sergey Brin and Larry Page released the first version of “BackRub.” Though other internet search engines had come before, none had sought to organize information by the number of backlinks a source received.

We know how monumental Brin and Page’s innovation proved to be. Today, Google’s market cap sits at $1.4 trillion – a figure that would make Croesus blush – even if it is down from 2021’s high of $2 trillion. The company generated $283 billion in revenue in 2022, up 10% from the previous year, with search contributing 57% of that figure.

While its financial might cannot be ignored, the company’s civilization-scale effect is more striking. Google search is an irreligious omniscience capable of producing answers in fractions of a second. The oracle would blush, too.

Indeed, never before have we relied on one source of answers so completely. It is the place we go to ask virtually every question we can think of, from sacred to secular to profane. (The third-most common search term in the world? Pornhub.) It is the destination for nearly every one of our assorted wonders: What should I watch? Where should I work? Does my mother love me? Am I paid too little? Why is Russia invading? Is my husband having an affair? Why do I have a headache? Am I a good parent? Am I a good son? What happened in 1813? How do I recover from a stroke? Did the Knicks win? 

Google’s strength is that it has all the answers. For the first time in recent memory, however, the tech community is looking at the company’s search business and asking questions. 

The renewed interest in search as a category stems from the emergence of ChatGPT and other AI models. Since it launched late last year, the OpenAI product has been heralded as the future of search and a potential Google killer. Former Coinbase CTO Balaji Srinivasan hailed it as a “historic accomplishment,” and “already better than Google for many queries.” Paul Buchheit, a partner at Y Combinator and the creator of Gmail, pronounced his old employer “a year or two away from total disruption.”

Could Buchheit be right? Will Google follow the Encyclopedia Britannica and Yellow Pages into quasi-obsolescence? Or is this merely an information empire in transition, willing to unbalance itself to survive?

Today, we’ll explore both sides of the argument, outlining why Google cannot get disrupted – and why it might happen anyway.

Is Google search undisruptable?

From many angles, Google search looks like an undisruptable product. It produces an extraordinary amount of revenue, commands a market-leading position, inspires consumers’ affection, and is supported by one of the world’s most technologically innovative companies on earth.

Before we outline these arguments (and quibble with them), let’s take a moment to understand how Google search reached its pinnacle. 

Though it may feel this way now, Google was not pre-ordained to become a trillion-dollar business. In 2002, for example, the budding search engine company was treated with some skepticism. The New York Times titled a piece from that year, “Google’s Toughest Search Is for a Business Model,” summarizing the market’s concerns. One Salomon Smith Barney analyst more directly articulated Wall Street’s worries. “People rave about Google,” the banker said. “But as a business, it will take an awful lot for them to catch up to Overture.”

Overture – the GoTo search engine creator – was not Google’s only competitor. Companies like Lycos, Excite, AltaVista, and AskJeeves vied for share in the aftermath of the dotcom crash. Why did Google win? 

For one thing, Google had the better tech. Brin and Page’s fundamental insight that “link popularity” was an indication of source quality proved to be profound. The “PageRank” algorithm it inspired allowed Google to provide better, more useful results than its competitors. Though a single example, it’s indicative of the technological power that defines Google – and that it has leveraged throughout its life. 

Google’s sharper technology resulted in a superior product. Google was not only the most useful search engine, it was also fast and intuitive. Brin and Page protected this strength in the face of temptation. Other search engines, for example, allowed advertisers to use pictures in their postings. Google didn’t. Why? Because images slowed the speed at which pages loaded, corroding the user experience.

Finally, Google rolled out its advertising model savvily. It eschewed pictures, as mentioned, allowing only small, text-based spots. Critically, Brin and Page only offered advertising space above and to the side of organic results. Compared to a player like Overture, this was a significant shift. The firm the Solomon analyst saw as an undefeatable Goliath surfaced only with sponsored results; companies paid to rank on certain search terms, with the highest bidder receiving the top spot. Google was shrewd enough to “borrow” this auction-based system (inviting a patent infringement case from Overture) but used it to augment organic sources.

Google’s success relied on more than just better technology, a superior product, and a smarter business model – it would be a different business had Eric Schmidt not stepped in as CEO in 2001 – but these elements were foundational. Through the lens of the other information innovations we’ve mentioned, Google succeeded for a simple reason: it shortened the distance from question to answer and found a way to monetize it.

Not even Brin or Page could have imagined just how monetizable their search engine would be. As alluded to, Google search is not just a large business but one of the world’s greatest, most monumental businesses. It is the economic engine behind the entirety of the Alphabet empire – the organizing principle, raison d’etre, alpha and omega. Other businesses like Chrome, Maps, and Android exist, at least in significant part, to bolster its power; moonshots in biotech, autonomous vehicles, and satellites can only be afforded because of its productivity. 

How large is it, exactly? In 2022, search generated $163 billion in revenue, 57% of Google’s total turnover. The entirety of Google’s advertising divisions produced $224 billion, 79% of all revenue. Here, perhaps, we see the first chink in the armor. Is it healthy for a business to be so reliant on one product? Should a $1.4 trillion enterprise balance the majority of its weight on a single source of revenue?

While a fair concern, historically, search has looked like a bulletproof business. Every internet user in the world requires it, and as information proliferates, one might argue its value should, too. Moreover, Google has moved to reduce its reliance on search over time. For example, in 2017, search made up 63% of revenue, which has declined in recent years. Though still Google’s most important business line, the growth of the company’s cloud services and YouTube ad division have helped diversify revenue.

Google search’s market share is equally impressive. Varying estimates suggest Sundar Pichai’s firm commands between 84% and 91% of the space, with Microsoft Bing and Yahoo stuck in the single digits. Depending on the data source, this share may have dipped slightly over the last few years, but not by much. Google has done an extremely sound job of building distribution advantages around search. Every time a user downloads the Chrome Browser or buys an Android phone, Google has essentially set them on the path toward its search engine. 

Perhaps because services like Google search are free for consumers, it’s easy to forget how much we value them. In 2018, three economists sought to quantify how consumers valued various free internet services by asking how much they would need to be paid to abstain from them for a year. At least in theory, consumers showed a high willingness to pay for modern life’s digital comforts, requiring $3,600 to give up online maps, $8,400 for email, and $17,500 for search engines. 

‍

Though these figures refer to a product category rather than a specific provider, they nevertheless serve as a strong proxy for the value we derive from services like Google search.

An interesting thought experiment: would that figure be higher or lower today? Five years after the research was published, has Google search become more or less valuable to us?

Here, too, we may spy another weakness. Though it remains the world’s search engine of choice, there’s a sense among many that Google’s product has degraded. Much of its deterioration seems purposeful, either as a way to extract more revenue or obstruct competitors.

The most obvious example is the increased ad load on search. Today, users might have to scroll through 42% of the results page before reaching an organic result. It bears only a passing resemblance to Google’s turn-of-the-millennium product. Minimalist efficiency has morphed into a page stuffed with sponsored links, ornamented with “knowledge panels,” and filigreed with distractions. This is baroque search, rococo search, search designed by Louis XIV reincarnated as a Mountain View PM. 

Google has not only added more advertisements; it has camouflaged those it serves. These trends go together: if users realized how much of the page is consumed by commercial activities, they might pick a more restrained provider. One Twitter user chronicled this slide toward the furtive, outlining the “color fade” of Google’s ad labeling over the years. 

More cruft comes from links to in-house products. For example, Google Hotels, Flights, and Shopping are often prioritized in relevant queries. While this may improve user experience sometimes, it also obstructs discovery. It also means that Google acts in competition with the advertisers it serves. 

All of this is a long way from Brin and Page’s willingness to put consumer needs first, forgoing the easy money of ubiquitous ads. Google has found ways to extract more money from the distance between question and answer, but it has made that journey longer and more convoluted. 

The final point to make when assessing the strength of Google search is that it is backed by a business with deep pockets that spends lavishly on research and development (R & D). Google invests tens of billions every year into frontier technology. In 2022, it deployed $39.5 billion, up nearly $8 billion from the prior year. While this investment is spread across several moonshot technologies, much is focused on artificial intelligence. Google has made AI a priority for some time, launching its Google Brain division in 2011 and purchasing startup DeepMind for $500 million in 2014. Though it has taken a tell-not-show approach to its AI capabilities, there is no doubt that Google is home to many of the sector’s finest minds and most cutting-edge technologies. That could prove pivotal in the years to come.

Is ChatGPT a contender?

Do you know what happened to Croesus? It did not go well. As the story goes, when he asked the oracle of Delphi if he should attack Persia, he was told doing so would “destroy a great empire.” Croesus took that pronouncement as good news, a sign of Persia’s impending decline. 

As it turned out, Croesus might have benefitted from a clarifying question: sorry, oracle, which empire exactly?

Persian forces crushed the Lydian king’s army over a series of battles. Eventually, Croesus’s dominion dissolved in his hands. As he prepared to be burned at the stake, the once-great ruler would perhaps have had a moment to reflect on the value of good, clear information. He would also have felt the sentiment later articulated by sources as varied as Ralph Waldo Emerson and Omar Little from The Wire: if you come for the king, you best not miss.

Over the years, Google Rex has entertained plenty of foes – and vanquished them all. Bing’s warships patrolled the kingdom’s coastline for a time but never came to shore. Yahoo pushed an old bathtub and wooden planks into the water and started paddling. The vast infantries of Yandex and Baidu conquered territories Google will not or cannot reach. Somewhere, a trio of ghillie-suited DuckDuckGo scouts sits in a tree.

To be clear, these are all viable enterprises in their own right. Though often treated as the butt of a joke, Bing earns $11.6 billion in revenue – who’s laughing now? Yahoo is…Yahoo, a business purchased in 2021 for $5 billion thanks to its finance, sports, and news media properties. Baidu and Yandex are regional giants, multi-faceted in their revenue-generating activities. And even DuckDuckGo, though smaller, is not insignificant. The Boston-based company has been profitable for nearly a decade and reported more than $100 million in revenue in 2021. In the matter of search, though, none come close to Google. 

Last November, a new adversary appeared: ChatGPT. Though it looked like a novelty at first glance, the power of its AI rightfully earned it more than 100 million users and much fanfare. It is, by some accounts, the fastest-growing consumer product of all time, besting the likes of TikTok and Spotify. As we’ve outlined in “Endless Media,” OpenAI’s product and the broader generative AI movement may change how we create and consume content. Will it be equally revolutionary when it comes to search?

Some believe it has already upended the sector. As mentioned, even those pivotal to building Google’s empire, like Paul Buchheit, see ChatGPT as a threat – a slippery assassin loosed in the king’s court. Many others agree. In the days and weeks following its release, Twitter was awash with examples of ChatGPT’s capabilities as a poet, historian, coder, teacher, and mathematician. Were this your only source of news, your only connection to the internet, you’d be forgiven for thinking Google was dead already. 

When faced with a technological marvel, astonishment is the natural response. It is not the most useful one, however. Astonishment numbs, overwhelms, flattens. It saturates our sensitivities such that finer thought becomes difficult, like taking a shot of lidocaine to the palm and then trying to tie your shoes. To gauge how significant a threat ChatGPT is to Google, we must examine its strengths and weaknesses more closely. 

It’s notable how different using ChatGPT feels. The simplicity of typing a request into a box and receiving a thoughtful, written response – unencumbered by images, advertisements, and other links – is compelling. From a pure interface design standpoint, OpenAI’s platform feels like a liberating throwback, a return to simplicity and clarity.

Beyond a more elegant interface, ChatGPT can answer certain queries more effectively than Google. Given the newness of the technology – and how rapidly it is developing – it’s unlikely we’ve comprehensively mapped exactly where it excels. However, there do seem to be places in which there are clear superiorities: 

• Creating original content. ChatGPT is not scraping the web and returning a response. Its AI is “thinking” and creating content it believes answers your question. Whether this is the job of a search engine is another matter, but it is undeniably useful for circumstances in which original content like a code snippet, blog post, email draft, or social media post is valuable.  
• Modifying delivery style. Related to the above, ChatGPT allows users to control how they would like an answer delivered. Whether that’s cosmetic or consequential depends on what the answer is being used for. It is fun to write a trademark application in the style of the Old Testament, but not especially valuable. However, producing one that resembles a legal document might be extremely useful. 
• Asking contextual questions. ChatGPT is especially good at answering questions that rely on immediate context. For example, if I post an essay into ChatGPT, I can ask questions about that piece of work. What mistakes have I made? Does it sound upbeat or depressive? Can you create a summary for me? Though ChatGPT might not do all of these things perfectly, Google search does not begin to play in this space. The same principle works for math equations and code.

Talia Goldberg, a GP at Bessemer Venture Partners that has written a compelling piece on the future of search, identified several ChatGPT features that overlap and shade the strengths mentioned above, including “specific, non-subjective queries,” “queries that require pattern recognition,” and “NLP-based queries like translation, summarization, sentiment assessment.” In time, we may find ChatGPT coalescing around particular use cases.

Like any intelligence, ChatGPT has its limitations. For one thing, it’s not clear that it’s a search engine. “I think a part of ChatGPT will be search, and it may become a primary destination for certain types of search behavior,” Goldberg noted. “But I don’t believe search will be its dominant use case.” 

This is not mere semantics. From Google’s perspective, it matters a great deal whether ChatGPT sees its future as a search engine. If it does, it’s likely to build its product to serve that use case – and compete head-to-head. If its priorities lay elsewhere, it may neglect obvious features in favor of others. As ChatGPT evolves from experiment to business, Google will closely watch its direction of travel. If you want to know what ChatGPT thinks of its positioning, here you go:

If one does see ChatGPT as a new type of search engine, there are more immediate deficiencies to worry about. Chief among them is the AI’s “limited knowledge” of information after 2021. ChatGPT does not know who won the midterm elections, if your team won or lost, or whether you need a coat to walk to the office. It does not know that Elon Musk is CEO of Twitter or that Lula is once again president of Brazil. If its ignorance were consistent, it should not even know that it exists. 

This might not sound like such a big deal; the world did not begin in 2022. Surely most useful information falls outside the immediate. Reviewing the world’s most popular Google searches reveals the weakness of that assumption. The French writer Voltaire said, “judge a man by his questions rather than his answers.” By that measure, we are mostly a hapless species: forgetful, lazy, greedy, and needy. We ask for the URLs of popular websites, what the weather looks like, and where we can find a Starbucks. 

Look at America’s fifteen most popular Google searches by keyword:

How many could ChatGPT answer? Where would it outperform Google? It could not give you location data, weather information, or news. Sure, it could spit out a few website links, and it might have the edge in translation. But that’s about it. ChatGPT shines in solving the complex but doesn’t attempt the simple. 

Another issue for ChatGPT is its reliability. As part of their praxis, the ancient oracles inhaled ethylene, a gas with an anesthetic effect. Perhaps this is why visitors like Croesus received such gnomic advice. 

ChatGPT has a similar issue. It is a rare genius – but one that huffs paint-thinner in the supply closet. As a result, ChatGPT “hallucinates” up to 20% of its answers by some estimates, which means it invents them from whole cloth. It will concoct lists of academic papers or devise theorems about undrainable lakes. 

In some respects, this is the artistically apt end for our era: news devoured by fake news devoured by pure fabrication. But if we are not optimizing for gorgeous, poetic decline, this is an obvious problem. While Google certainly surfaces links with incorrect information, its results are structurally very different. Google presents the user with a menu with an invented dish; ChatGPT delivers a falsehood with the confidence of a dictum.

There are smaller inadequacies, too. ChatGPT does not support image search, for one thing. It does not attempt to link requests to potential actions – we are used to a restaurant recommendation being appended by a link to book a table. It shows no supporting sources, provides no citations.

ChatGPT is also expensive. Talia Goldberg highlighted this point in our exchange. Goldberg noted that while traditional search engines incur costs north of $1 billion per year to index the web, AI players like ChatGPT are “far more expensive. This is part of why the company has had to throttle usage of its product and is seeking further capital.

It’s worth asking, are ChatGPT’s frailties permanent or transitory? Can the product reach parity (or better) than Google in the places it matters? 

Most seem tractable in the short-to-medium term. Given that Google itself is experimenting with a chatbot that incorporates current events, it seems possible that ChatGPT could do the same. That alone would resolve one major discrepancy between the two platforms. Over time, it’s reasonable to expect ChatGPT will reduce its hallucination rate. Adding images and supporting sources also seems extremely doable.

Some differences will take more work to address. ChatGPT is unlikely to build mapping, booking, or shopping features soon. Though it could integrate with other players, doing so may introduce friction for users. Moreover, whether a more restrictive chat interface is optimized for these behaviors is unclear. It’s a fundamentally two-dimensional experience. You can’t click into a new page or easily toggle back and forth between multiple windows. The activities mentioned benefit from a more dimensional experience. When searching for new clothes to buy or a coffee shop within walking distance of a doctor’s appointment, we’re surveying – not looking for an immediate, definitive answer.

In time, AIs may become sufficiently personalized that they can surface a range of tailored options better than the results of human browsing. For the next few years, a more modular interface seems preferable. 

Lastly, ChatGPT is likely to remain more expensive for the foreseeable future. That may limit how widely it is made available.

What happens next?

In 1998, Sergey Brin and Larry Page accepted their first outside capital for Google; that same year, biologist E.O Wilson published Consilience. (Those interested in learning more about Wilson will enjoy our interview with Josh Wolfe). Among the many memorable phrases that Wilson produced, none is more frequently repeated than this passage:

We are drowning in information, while starving for wisdom. The world henceforth will be run by synthesizers, people able to put together the right information at the right time, think critically about it, and make important choices wisely.

It is a remarkably prescient assessment, especially given it predates much of the internet’s proliferation. It also feels particularly apt to the products we are discussing, and the debate this piece represents. Google succeeded, in part, because it was an exceptional synthesizer of the world’s information; ChatGPT takes this sensemaking to another level – even if it does so less reliably. 

It is time for us to demand some synthesis from ourselves. Looking at the market composition, Google’s historical strength, and the thrilling abilities of ChatGPT – where are we headed? What will the next few years look like?

First, we should expect a strong reaction from Google to counter the current narrative. CEO Sundar Pichai must ensure pronouncements of his company’s obsolescence do not become the story. For over a decade, Google has invested aggressively in AI – now is the time to show the fruits of that labor. 

To its credit, Google seems determined not to underreact to ChatGPT’s emergence. Internally, management referred to its release as a “code red” and has moved quickly to kick-start and accelerate related projects. On the company’s most recent earnings call, Pichai emphasized how important AI is to Google and noted that consumers would be able to interface with its cutting-edge LaMDA (Language Model for Dialogue Applications) technology “very soon.” Other initiatives, including a chatbot nicknamed “Apprentice Bard,” are in testing. 

As well as demonstrating Google’s capabilities, launching products like Apprentice Bard could recontextualize how consumers see ChatGPT. Meta’s head of AI, Yann LeCun, has referred to OpenAI’s product as “nothing revolutionary,” and “not particularly innovative.” While the rivalry between organizations could color that assessment, it might also be true.

Beyond showcasing its technology, Google must also ensure it does not forfeit the bottom end of the market to OpenAI. One of Sam Altman’s savviest moves has been to capitalize the sector’s startups aggressively – and provide early access to his company’s models. Not only does this buy good faith, it positions OpenAI favorably: startups use its models first, and it stands to hear about innovations and relevant consumer demands before the rest of the market. 

This may be part of the motivation behind Google’s approximately $400 million investment in Anthropic. The AI business was founded by two former OpenAI employees and is working on a ChatGPT competitor. We should expect more venture activity from the search giant in the coming months. 

We may also see a redux of the “chatbot” craze that dominated tech in 2016 and 2017. The success of ChatGPT has demonstrated the utility of the interface – a demonstration that is now likely to be over-extrapolated. Already, several startups have emerged offering chat-based search engines, including You, Andi Search, and Perplexity. The latter is reported to be raising a $15 million seed round. Neeva, a privacy-based search engine, recently pivoted to focus on AI.

This interest will not be limited to startups. Microsoft is expected to bring ChatGPT’s technology to Bing, and Baidu is working on its own chatbot.

Looking at the products in-market, there is a clear coalescence around design. Namely, a fairly straightforward conversational interface, sometimes as a “tab” in a broader search engine. Though Google’s scale means it’s unlikely to add a “Chat” tab anytime soon – the risks of providing incorrect information are too high at its size – it illustrates what may eventually happen. 

How will these companies monetize? A fundamental premise behind the argument that ChatGPT will make Google obsolete is that an ad-based model doesn’t work in a chat-based world. Talia Goldberg noted this potential, saying, “What could be damaging [to Google] is a potential interface shift that moves the industry away from pay-per-click ads and links.” Paul Buchheit categorized Google’s situation as a classic “innovator’s dilemma” in which responding to a competing threat kills the core business.

Indeed, replicating the sophistication of Google’s advertisements and money-generating links in a new interface could take time. Additionally, the different space constraints of chat might reduce total ad load.

However, a chat interface doesn’t seem incompatible with a sophisticated ad model. After all, there is space to run ads, and it seems possible to introduce a pay-per-click system to a conversational interface. Chatbots may even have some distinct advantages. The direct dialog feels more like a conversation with a real person, which could be higher converting. Moreover, the more complex queries that users turn to chatbots to solve may provide additional context that improves targeting. It’s easy to imagine a slightly pushy, hoggish version of ChatGPT coaxing consumers into specific purchases. Run the OpenAI product through a Black Mirror filter, and you might get something like this: 

For now, ChatGPT has taken the subscription route. It is testing out a $20 per month premium tier that offers unlimited queries and priority support. Though a savvy way to monetize its super users, that price point is not designed to attract a mass-market audience. 

With a few tweaks, however, ChatGPT or similar tools could prove extremely attractive to enterprise audiences – even at much higher price points. Verticalized chatbots for law, medicine, engineering, software development, sales, real estate, and other industries could prove invaluable. While ChatGPT does a good job pitching in across a range of subjects, providing more upfront context and training could take its utility to a new level. Many startups that emerged in 2016 will be tried again – this time with real power underneath the hood.

While chat will be the best interface for some of these needs, the next few years should see further innovation around search’s interface. By leveraging modern AI, we may see new vertical search engines join the likes of YouTube, Pinterest, and Reddit. Voice assistants may move beyond their inability to understand even a basic compound question and become useful. “Ambient” AIs may live in a browser extension or desktop app, following us from one application to another. We might summon them to find a suitable citation in our word processor, the latest financial results in our spreadsheet, or the right words in our email inbox. The art of finding answers is open for the first time since Google’s emergence. 

“Google can bring you back 100,000 answers,” author Neil Gaiman said. “A librarian can bring you back the right one.” 

For the past two decades, Google has won by providing the shortest route between question and answer. It did so by surfacing and ranking the web’s resources, creating an infinite compendium instantly. That product will be difficult to dislodge. 

ChatGPT’s early success suggests it is not impossible, however. OpenAI’s product trades Google’s vastness for directness, dimensionality for simplicity. It is Gaiman’s librarian, leveraging vast sums of human knowledge to provide a definitive answer – even if it sometimes gets it wrong. In E.O. Wilson’s Age of the Synthesizer, ChatGPT’s talent for distillation, generation, coalescence – managed at speed – may prove especially valuable.