How AI in search engines will affect online stores

The Internet was full of ideas on how to use the new technology - both for business and very casual purposes. Teachers indicated that students use this software to prepare essays. Marketing specialists praised the solution for its support in brainstorming and researching new ideas for campaigns.

The spectacular popularity of ChatGPT also prompted the announcement of "code red" in Google. Corporation decided that a chatbot with such capabilities is a threat to the business model and the functioning of the entire organization. For most online stores the search engine from Google is one of the most important channels for acquiring users - therefore ecomms should also be interested in how Chat GPT impacts the future of those businesses.

What are Large Language Models (LLM)

Behind the functioning of ChatGPT and similar solutions operating on the market is the development of the so-called Large Language Models. These are machine learning-based algorithms that can read, process, and produce word-based content. Preparing such models is possible by training them on large sets of texts - in the case of ChatGPT it was the entire Internet.

The model prepares its answers based on query analysis based on the meaning of words and the relationships between these meanings and then searches its resources (i.e. the entire Internet on which it was trained) in search of matching answers.

Due to its accessibility for users, ChatGPT is currently the most popular solution of this type. However, most of the tech giants, have prepared their own models. Google is about to launch Bard, Nvidia works on NeMo LLM and Meta is developing XLM-RoBERTa.

Advantages of searching the Internet using LLMs such as ChatGPT or Bard

"Code Red" on Google was announced for a reason. Searching the Internet using solutions based on large language models has many advantages. Among them are:

• A more natural, conversational form of obtaining information - ChatGPT and similar solutions communicate with the user through conversation and remember the context and information obtained during the conversation. This is the closest to the natural form of interaction. At the moment, interaction with the search engine is carried out using a specific para-language, i.e. properly structured queries to the search engine.
• The ability to synthesize information or provide studies - language models can provide studies and summaries at the user's request. This can be compared to snippets in Google, which display the most interesting part of the page for the user, but the answers prepared by language models are based on many websites and are a mixture of knowledge contained therein.
• More effective reading of the user's search intention - the language model can get more information from the user about the context. In the case of overly complex queries, search engines get lost - it is even easier for the language model, because, unlike the currently existing search engines, it is able to process the obtained information.

Disadvantages of search engines using LLM

Large language models that allow you to automatically search for ready information from the Internet can be a powerful support for users, especially if their query is more complex, or they are unable to include their need in a few accurate words that fit in the search box.

This does not mean, however, that natural language processing models are without flaws. Among them are:

• Providing a summary instead of links - when using a standard search engine, the user is provided with links along with short summaries. It is up to you to make the final decision on which of the links you want to click. Providing a summary and a result instead of links strengthens the effect that can be described as "the winner takes it all" - at the moment, pages with the top 3 search results take up to 70 percent of traffic. When the links disappear from the results, this effect will be even more powerful.
• Mistakes (or thoughtlessness) of models - language models seem to be omniscient, and they provide linguistically correct and coherent answers to the questions asked. However, such models are not infallible. What's more, they do not see their lack of knowledge and do not admit it, which can mislead the user.
• Unpredictability - The final challenge is the unpredictability of these models. When encountering unusual situations, which are easy with millions of users, they may start to behave strangely. Currently, one of the most important problems faced by the creators of this type of solution is the so-called "Hallucinations". The model creates plausible facts, fabricates evidence, and presents as the result of a fair trial and factual information. An example of sometimes incorrectly functioning LLM is the chatbot provided by Microsoft in the Bing search engine, which describes the magazine "The Verge".

What Search Engines Are Using LLM Right Now?

Despite these adversities, several solutions are already available that use large language models as the basis for their operation. Currently the most popular are:

• you.com - one of the first search engines that set itself the goal of searching the web using AI. The company was founded by former salesforce employees who felt there was a need to refocus on the user experience when using search engines.
• neeva.ai - a company founded by former Google employees who set themselves the goal of providing users with a solution that better cares about their privacy and data. From the very beginning, the search engine used solutions based on large language models.
• Bard - Google developed its own language model with similar capabilities as a response to ChatGPT. However, due to the importance of the search engine product, the company is very cautious about experimenting.
• Bing - Unlike Google, Microsoft doesn't have much to lose in the search engine market, which allows it to experiment boldly. The Bing search engine has been integrated into a model based on the same operating principles as ChatGPT.

How LLM search will affect online stores

At the moment, it is difficult to talk about the clear impact of new technology on the daily operation of online stores. However, in the coming years, answers provided by large language models will increasingly be part of the everyday use of search engines. This will have the following effects:

• People will trust search results more - in a situation where the answer is not a dry list of links but an essay on the topic provided by the user, the results will seem more credible, and people will believe in them the more
• Stores will need to focus on producing more accurate, more useful content - language models provide their own answers, but they need to learn them from online content. The better this content and the more precise the answer to the user's query, the more likely the model will redirect to this particular page.
• There will still be an algorithm behind the operation of LLM and the search based on it, the operation of which will be studied by SEO specialists and specialized agencies - although it will be much more sophisticated and complicated than the current algorithm that controls the search engine, it will still be cracked by specialists.
• Interaction with the search engine through conversation will further deepen the popularity of services such as Alexa, Google Assistant, Siri, or Cortana - in a situation where the result of searching the Internet will be in the form of natural speech, the obvious direction is to shift towards conversation as a model of searching the Internet.

Summary

Browsing the web using conversational models is the future, which will become a reality in the coming years. Still, the only answer that online stores can prepare is focusing on the user and knowing their needs.