In the emerging power struggle of the search engines, it must be taken into account that Google is exposed to the so-called innovator’s dilemma. The search engine is the basic innovation and the driving force of Google’s advertising business. With digital transformations, it is always so easy to say “destroy yourself before others do”. Of course, this is not so easy when you have to practically question your own original business model, with which you generate more than 80 % of your turnover. In addition, Google has always pursued the performance promise of using intelligent algorithms and indexing to make the most reliable and credible pages possible. Even if the strong advertising penetration seems somewhat annoying in some cases, Google fulfils this performance promise very reliably.
In contrast, the strength of language models is not necessarily to provide reliable sources, but to generate answers in natural language in the respective context, but not with the guarantee that these sources are really quality-assured – even ChatGPT has no source information at all. This quality uncertainty inherent in the language models contradicts the Google principle EAT (Expertise – Authoritativeness – Trustworthiness). This may also have been the reason why Google was hesitant to make a corresponding conversational AI solution available as a dialogue system for the mass market, despite immense AI expertise and the powerful – and in comparison to ChatGPT larger – language model LaMDA (Language Model for Dialogue Applications).
Certainly, chat question slots will complement the classic search slot. And of course it’s more convenient to be able to ask your question linguistically, of course, and then get a well-formulated answer in one piece that makes navigating through Google hit lists obsolete.
Ultimately, however, the user has to rely on the answer given without being shown sufficient quality indicators for this. If the quality and resilience of a request is particularly important, language models alone will not suffice. Navigating through a Google solution space can provide additional security here.
On the other hand, the respective “intent” (user intention) is decisive: In the case of transaction intents, consumers are looking for products and services; in the case of navigation intents, the aim is to find out, for example, where the product can be bought at a particularly favourable price. Here, the classic search is clearly superior. Information intents are more about providing and explaining information.
Here, models like ChatGPT have an advantage in that they provide explanations and answers to the user without the user having to laboriously click through any lists of results and practically construct the answer himself. Considering that informational search accounts for about 80% of all queries, it quickly becomes clear that there will be a shift in the search market.
As a result, a hybrid approach combining both classical search and natural language conversation will prevail. There will also be a levelling of both approaches in terms of advertising. The commercialisation pressure on the language model-based approaches will lead to the introduction of advertising messages here as well. The first ChatGPT extensions from Bing already show examples of advertising.
The question of the scalability and economic viability of foundational models as a replacement or supplement to classic search will be exciting. Question-answer dialogues must take place in real time. Given the immense computing effort of language models, the question arises as to how this can be mapped economically. Moreover, there is still no clear concept of what the hybrid model should ideally look like. How can the best of both worlds – conversational AI and search engines – be systematically combined?
The increasing competition in this market is ultimately good news from the consumer’s point of view. In contrast to the monopolistic search engine market of the past, there will be three to four major language models on the market in the future that will enable intelligent conversations in addition to traditional search. In addition to Google and OpenAI/Microsoft, Meta as well as some Chinese players and possibly also European start-ups can be expected.
