Revitalizing Search: The Impact of LLMs on Personalization and Voice Assistants

Some people are better at predicting the future than others, even though no one can do it perfectly.

A study analyzing over 1 million judgments made by thousands of individuals identified a group called "super forecasters" who have a much higher success rate in making accurate predictions.

Super forecasters stand out from lousy forecasters by their ability to gradually adjust their opinions in light of new information.

According to a quote attributed to the individual who holds the world's largest ecommerce empire, the Washington Post, and significant wealth, individuals who consistently make accurate predictions are those who are attentive listeners and are willing to alter their perspectives frequently.

I often get asked what I am reconsidering, and my answer is always the same: voice and personalization.

The most impactful force of change in our current era, even more so than Blockchain, is AI. Through large language models (LLMs), AI has the ability to bring voice assistants and personalization closer to fulfilling the potential they initially promised.

Boost your skills with Growth Memo’s weekly expert insights. Subscribe for free!

Don’t Call It A Comeback

The prediction that "50% of searches will come from voice by 2020" was exaggerated and repeated excessively for its fear factor. The truth is, voice assistant devices are not performing as well as expected. They function more like speakers than actual assistants.

Investors were concerned when Mark Zuckerberg revealed plans to invest $10 billion annually in developing the Multiverse. However, it is surprising that not much attention is given to the fact that Amazon is also losing the same amount of money on Alexa devices, despite selling 500 million of them.

Similarly, Apple and Alphabet find themselves in a similar situation, with nearly 100 million users for Siri and Google Assistant. These virtual assistants come pre-installed on their wide range of hardware devices.

In 2020, 54% of U.S. adults aged 18 and above had used voice commands, with 24% using them daily. However, most people mainly use voice commands for tasks like playing music or checking the weather.

For me, I mostly use my voice assistant to ask Siri to resume the podcast I was listening to when I leave the shower. It's important to note that giving commands and doing searches are two different things.

There is reason to be hopeful. LLMs have the potential to enhance search boxes quickly and make voice assistants more helpful by understanding user intent better and providing more natural responses.

For tech giants such as Alphabet, Amazon, and Meta, any time users are not viewing ads represents a chance to tap into the market.

Americans spend a significant amount of time each day on various activities such as driving, cooking, and cleaning. If just a small portion of that time is dedicated to interacting with AI, it could result in valuable ad or subscription revenue.

Having a reliable voice assistant not only has the potential to increase revenue directly but also create a new app ecosystem and boost sales of hardware. Amazon, with their range of smart home products like home security systems, is in a prime position to capitalize on this opportunity.

Amazon recently had to abandon a $1.7 billion deal to buy iRobot because of regulatory issues. Despite this setback, Amazon is busy enhancing Alexa to improve conversation capabilities.

Voice assistants face challenges with speed and access. Even if they can generate answers quickly, the interaction may not feel natural or conversational.

Having access to a wide range of information allows voice assistants to provide more personalized responses.

In the mobile era, smartphones revolutionized the world by putting a powerful computer in the palm of your hand. But imagine if computing could be available wherever and whenever you need it, always ready to assist you. This concept, known as ambient computing, envisions a seamless integration of devices, services, and AI to provide assistance wherever you are. The technology works in the background when not in use, making you the focal point of the system. This is Google's vision for ambient computing, where help is fluid and always within reach.

My Web Or Your Web?

Web personalization is the Blockchain of the early 2010s: a disappointment.

Ten years ago, we thought that the future of the internet would be all about customized websites and search results. However, instead of that, we ended up with filter bubbles and scandals like Cambridge Analytica. It's not easy to gather and expand personalized data. People enjoy using free services, but they don't like the idea of being constantly monitored.

Google was the first to introduce personalized search in beta form in 2004, and later rolled it out globally in 2009. Nowadays, Google tailors search results for certain queries such as "what to watch," upcoming events, previous searches, and content on Discover.

The level of personalization in Google Search is not very significant.

In 2013, a study discovered that the average level of personalization in Google Search was 11.7%. This percentage varied greatly depending on the specific search query and ranking. It was noted that results appearing in higher positions were more likely to be personalized compared to those in lower positions.

In 2019, a study discovered that Google customizes 2 out of 10 search results for individuals and 4 out of 10 for political parties. This means that there is not a significant level of personalization. When it comes to search, the most personalized feature is Google Suggest. This tool offers query suggestions based on the millions of searches conducted by users similar to me, sometimes providing suggestions that can be quite surprising.

Google now provides personalized fashion recommendations for its search users. Based on their preferences, users receive tailored style suggestions and have the option to save favorite brands for a more customized search experience.

In 2022, Google began personalizing shopping search results according to your past purchases.

When shopping on Google, simply choose your favorite department and brands one time. This way, you will see more of those selections in the future. For example, if you pick the "womens" department and the brand Cuyana, the next time you search for a messenger bag, you will be shown women's messenger bags from Cuyana and other similar brands.

Getting personalized shopping recommendations is way less intimidating than keeping up with political news. Plus, it's more profitable because Google can help you find what you need faster, all while you stay on the search engine.

personalized shopping

Image Credit: Kevin Indig

personalized shopping

Image Credit: Kevin Indig

Personalization is most useful in shopping, news, and local, but less for informational searches with a learning intent.

Google is now using LLMs and machine learning along with the Shopping graph, which contains over 35 billion data points, to personalize shopping searches more effectively. This new approach has been in place for the past few months.

When you search for a product, you will receive a summary of important factors to consider and suggestions for relevant products. The product descriptions will include the latest reviews, ratings, prices, and images. This new shopping experience is powered by Google's Shopping Graph, which contains over 35 billion product listings. This makes it the most extensive dataset of products, sellers, brands, reviews, and inventory in the world. Every hour, more than 1.8 billion listings are updated in the Shopping Graph to ensure users receive fresh and reliable results.

“Hey Google, Where Is My Data?”

Voice and personalization are at a tipping point for success. Expect the unexpected when it comes to the future of these technologies.

If personalization and voice search continue to grow, it becomes even more crucial to be top of mind before a customer expresses their intent.

To retain customers, brands must excel in service, shipping, and returns. Personalized shopping search makes it challenging to change a customer's decision once they have made up their mind.

How would Google personalize results? A combination of first-party data from Google services like Gmail, YouTube, Android & Co., and Topics.

One factor I’m not bullish about is data.

Google has flooded the search results pages with various features like Map Packs, image carousels, and shopping carousels, but they haven't provided much data for marketers to understand their impact. I also worry that Google wouldn't share data about SGE if it is ever implemented, but that's a different issue.

What would make me reconsider is if we could see how queries and search numbers are obtained from existing voice devices, but companies like Alphabet, Apple, and Amazon do not disclose that information.

1 How the Best Forecasters Predict Events Such as Election Outcomes

2 Jeff Bezos

3 Amazon has sold more than 500 million Alexa-enabled devices, drops 4 new Echo products

4 Alexa, why are you losing so much money?

5 Why did Apple’s original HomePod fail? Let’s count the reasons

6 Roughly 1 in 4 U.S. adults now owns a smart speaker, according to new report

7 American Driving Survey: 2022

8 Exploring time spent on cooking, reading, and other activities for National Hobby Month

9 Average hours per day spent in selected household activities

10 Amazon is set to supercharge Alexa with generative AI

11 Measuring Personalization of Web Search

12 What did you see? A study to measure personalization in Google’s search engine

13 9 new features and tools for easier shopping on Google

14 Supercharging Search with generative AI

Featured Image: Lyna ™

Editor's P/S:

Paragraph 1:

The article highlights the importance of adaptability and open-mindedness in making accurate predictions. "Super forecasters" consistently adjust their opinions based on new information, while those who struggle to do so often fall short. The quote attributed to Jeff Bezos emphasizes the crucial role of active listening and willingness to change perspectives.

Paragraph 2:

Despite initial hype, voice assistants and personalized search have faced challenges. However, the emergence of large language models (LLMs) has the potential to enhance these technologies by improving user intent understanding and providing more natural responses. The article discusses the potential financial and convenience benefits of voice assistants and personalized shopping recommendations. However, it also raises concerns about data privacy and the lack of transparency from tech giants regarding voice search data.