TL;DR: 2023 Google I/O - No Bard Chatbot Search Debut - Twenty-Five Billion Dollars At Risk?

Google announced dozens of new software, development, and hardware products and features at its annual Google I/O developer conference today.

However, it is more remarkable what Google did not announce today: namely, the release of a new, Bard-powered Google Search that could compete with Microsoft’s chatbot-powered search, GPT Bing. While Google showed numerous examples of nifty features of a test version of Bard Search, no information was forthcoming on when and to whom exactly the product will be rolled out. And while the Bard Search demos were impressive, as anyone in the tech industry knows, a demo is a demo is a demo – not the real thing.

This tells us that Google feels it is not yet time to release Bard Search into the wild. Being the engineer-y company that Google is, it may be erring on the side of functionality and safety rather than that of making a PR splash. Bard Search is not quite ready yet from a technological point of view, and neither is GPT Bing. However, Microsoft chose to release it anyway to gain mindshare and perhaps try to drive some search traffic from Google to Bing.

Google I/O attendees were invited to use Bard Search, dubbed “an early experiment,” via Google Labs, only to be greeted with a waitlist signup – clearly, the product needs more time in the oven.

Alphabet/Google CEO Sundar Pichai must also be concerned about how the release of Bard Search will change users’ search behavior and what that might do to its bread-and-butter search advertising business. There are three types of searches: navigational searches (“Yahoo Finance”), transactional searches (“I want to order a pizza”), and informational searches, i.e., complex research sessions (“What is the best car insurance?”).

It is this latter category that a chatbot-based search will change in fundamental ways. Here is how informational searches work today: you conduct several searches about the given subject. In each case, you follow several blue links to destination sites. On each site, you search for the sought-after information. Finally, you manually synthesize the information. A chatbot-based search could do all that for the user with just one inquiry. This would reduce search traffic and directly affect search ad sales.

The question is, by how much will Bard Search reduce search traffic? There are no publicly available statistics on how inquiries on Google Search break down into the three types of searches. But we can perhaps safely assume that most searches are of the “navigational” and “transactional” types – but what is the share of informational searches that could be affected by a chatbot? 15%? 25%? 35%? Whatever the share, it would be prone to be wiped out. If we assume that just 15% of Google searches are taken up by complex research sessions, then nearly $25 billion of Google’s search advertising sales would be at risk based on 2022 sales numbers.

The next question is, would advertising in chatbot-based search results be able to make up for those search ad losses? Probably only for a small portion. Traffic would be reduced, so the real estate available for advertising would also decrease. But perhaps cost-per-click rates would go up because users would be more ready to purchase since presumably, chatbot search results would put them right on the brink of a purchase decision.

The bottom line is, nobody knows for sure how Bard Search might affect search advertising sales, including Google – so one can’t blame Mr. Pichai for moving slowly and deliberately.

However, as long as Google does not release Bard Search, the company risks losing search traffic and ad sales to Microsoft’s GPT Search. Fortunately for Google, at least so far, that does not seem to be the case. Search engine market shares have not changed after GPT Bing’s release in early February. In fact, March and April stats may have convinced Google that it could take some more time to perfect Bard Search.

The ultimate outcome of the Microsoft-Google duel will depend on whether the latter can find a sweet spot between enhancing its search product with chatbot functionality without ruining its core search ad business (which, of course, is also needed to finance all that expensive AI research). If Google can, it will win the contest because Bing cannot nearly match Google Search’s distribution.

While Google did not announce Bard Search release details, it did announce the general availability of its ChatGPT rival, the Bard chatbot, today. Dubbed “an experiment by Google,” Bard will be available in 180 countries and 40 languages. In some initial testing, it seemed to be faster than GPT and also appeared to present more complete, better-structured responses—when it worked. It didn’t always, though. For example, when I asked Bard to summarize several texts in a row, it did fine on the first one, but for every text after that, it just told me, “I’m a language model and don’t have the capacity to help with that.” In another test, I asked it to copy-edit a text—only to get a summary of that text rather than a cleaned-up version of it.

There were a few more notable announcements at Google I/O. Most notably, the fact that all of Google’s AI products are now based on the latest version of its large language model, PaLM 2. Google announced it would infuse all of its products with PaLM 2 wisdom, as did Microsoft for its products. Stand by for new versions of Google Mail, the company’s productivity apps—Docs, Sheets, and Slides—Workspace, Google Cloud apps, and developer tools. (No release dates were announced, however.) Also, the next incarnation of a Google LLM is already in the works (again, with no release date announced): “Gemini” will be to Google what GPT-5 will be to OpenAI.

Google also announced a partnership with Adobe to use the company’s Firefly generative-AI graphics tool. This was somewhat surprising, given that Google also talked about “multimodal” versions of PaLM 2 and Gemini—models that would be able to process and generate graphics, music, and videos. Perhaps this indicates where Google is with some of its multimodal research.

Personally, I felt “Project Tailwind” was the most interesting announcement. This is a platform that enables anyone to train their own LLMs. Wouldn’t it be cool if a company could train its own little local LLM on all its data to make them more useful? (Enterprise search was also mentioned and is, of course, another exciting application of chatbot search engines.) Or, even cooler, if any person could train their own LLM on all the files they collected over the years, and it would give you the smart answers that were always on the tip of your tongue but never left it.

The market seemed to be not concerned about Google's future - Alphabet's stock price rose by 4% today.