I spent a chunk of my career as a product manager in search (Shopping.com, GLG, Ask.com, AOL and Google) so I find this particularly interesting. Until now, to succeed in search companies needed two things: 1) an index of the web 2) an algorithm to organise the results. In the early days of the web, companies like AltaVista, Excite, WebCrawler and others competed in becoming the search engine of choice. But two years after Google came to the fore in 1998, Google became the number one search engine and “Googling” has become synonymous with web search. Sure, there is Bing, DuckDuckGo, Ask.com and a few others, but no one comes even near Google in terms of market share or mind share (90.
However, since the introduction of ChatGPT in November 2022, user habits (and perhaps expectations) began to change. It felt like magic to just type a question in natural language and have the answer formatted, generated, and iterated, on the spot.
Why should users sift through ads and 10 blue links for the information they’re looking for if it can just be generated with LLMs and tailored to their specific query/prompt? And if you still want to do a web search, wouldn’t it be better if it was enriched with AI? Think about it for a second, since the introduction of ChatGPT, do you find yourself Googling more or less?
In December 2022, just a month after the introduction of ChatGPT, Google understood the risk as well and launched a ‘code red’ – a company-wide emergency effort to tackle generative AI and come back with an adequate response to ChatGPT and Microsoft Bing. In the six months since, Google worked hard to annouce a suite of new generative AI (namely Bard, Google’s chatbot which started off with a wimper but got much better shortly after) as well as a slew of new products and implement generative AI features across search and Google Workspace (some have been announced but not yet launched at Google Search Labs).
On May10th, at Google I/O, Google took the first steps on its mission to re-invent search and launch a number of generative AI powered products:
With new breakthroughs in generative AI, we’re again reimagining what a search engine can do. With this powerful new technology, we can unlock entirely new types of questions you never thought Search could answer, and transform the way information is organised, to help you sort through and make sense of what’s out there.
The race to intimacy in search, an AI powered experience
There are a number of factors driving the adoption of generative AI in search.
- The increasing demand for personalized search results
- The growing popularity of natural language processing (NLP) and LLMs
- The development of new generative AI models that are more powerful and efficient
The combination of ChatGPT, Bing’s integration of LLMs in search and Google’s embrace of generative AI technologies signalled that we’re entering a new paradigm in information retrieval with LLMs. We’re moving away from the 10 blue links (and sometimes 10 blue ads) and moving into something different.
In a recent All In podcast episode, Brad Gerstner, CEO of Altimeter Capital, quoted Richard Barton, co-founder of both Expedia and Zillow, two of the most popular vertical search engines (travel + real estate) referred to this new era of search as the ‘race to intimacy’. A conversational UI, that changes the user experience from query–> link to personalised extraction of knowledge. where rather than sift through noise to find an answer, users will generate a specific result to match their needs.
Watch this short clip (mixed with the normal banter between the hosts):
The new gen AI search companies
A number of new startups are trying jump on the generative AI search wagon. The strategy deck conveniently mapped the new players in Gen AI search and sized the market at $200 billon.
Most startups in this space is still under the radar, with the exception of Perplexity, which offers citations of sources for its generative AI search, something that to this day ChatGPT did not do (but Bard started to) as well as the ability to personalise the results based on the user’s bio, language and location.
You.com is another one, which offers to democratise search and let users vote on what’s the most helpful link for a query (vs. Google’s ranking which is driven partly by SEO). The company raised a $25M series A in July 2022.
The rest of the competition in the generative AI search space is focused on enterprise search. Much of it today belongs to Algolia. Makes sense, as companies can adopt a SaaS model and not require huge scale to monetise via advertising.
Like any tech advancement at scale, to succeed in this endeavour requires deep pockets: cloud resources and costly GPUs, expensive AI engineers and a working business model, all of which Google already has. But more than anything, it requires a huge amount of data, which is perhaps Google’s biggest competitive moat. Some companies, like Neeva.com, the first gen AI powered search experience, shut down earlier this month. According to the company’s founders, Sridhar Ramaswamy and Vivek Raghunathan:
Neeva’s failure was due to its inability to attract enough users, the rise of generative AI and LLMs and the current economic environment
In addition, companies using LLMs (without developing the underlying foundational models), face disruption from other competitors using the same APIs and risk falling behind in their technology stack.
If all that wasn’t enough, you have the problem of model hallucinations, where the conversational UI offers an answer that is well articulated, but completely wrong. It’s not that wrong information doesn’t appear at Google today, but under Google’s PageRank model, it’s unlikely that a site with wrong information will climb to the top of the search results page.
Finally, there’s competition. I referred to this in my post on incumbents vs. upstarts.
- What does SEO look like in a generative AI powered search world?
- Will Google cannibalise its advertising business in a generative AI search world, or will ads be inserted into generated results?
- How can users trust the results given model hallucinations and the fact that most of the new content will be machine generated? (Will we see Blockchain + Search emerge?)
- What will OpenAI do in search?
- How will regulation affect generative AI search? Will censorship be applied to these models?
- Will LLMs start giving us a warning when they hallucinate by showing us a ‘confidence score’?
Overall, I think competition is good for consumers and offers more choice and incentive for companies to innovate. While web search requires huge resources, I believe that smaller LLMs and open source technology will play a big role in creating specialised vertical search engines for topics like health, legal tech, travel, education, etc and there lies the opportunity for startups.