Search Generative Experience
The Google I/O 2023 sizzle reel for Search Generative Experience
Rebuild Google Search in 14 weeks
That’s what it felt like, and in a way, that’s exactly what we did. While Google had been at the forefront of AI for years, new entrants motivated the company to apply AI in consumer-facing products.
I designed concepts and investigated how we might use LLMs and generative AI to fundamentally improve the Search experience. I was one of the designers on the Verticals team and responsible for Local results. I worked with other verticals (like Ads and Shopping), our core framework team, executive sponsors, and others through the development, I/O announcement, and launch of Search Generative Experience in Search Labs.
Early prototypes investigating this vision included a Chat-like approach
Embracing LLMs as a UX medium
LLMs offer a unique and powerful way to summarize content. How could this technology be put to use in Google Search, where users are already coming to get answers? Can we do it safely, accurately, and helpfully?
Other strategic and technical problems quickly surfaced:
An LLM for Search relies on training data from the web. How can we summarize this content in a way that supports the ecosystem?
Search latency is measured in milliseconds, but these models need many seconds to return a result. Can we integrate this without degrading the Search experience?
LLMs are probabilistic, and that means the user experience is, too. But we had to incorporate factual info, like opening hours and addresses. Working with content designers and UX writers, we created our own set of “goldens,” or ideal responses. We created a novel templating system to ensure our front end could faithfully present the output to users.
Local responses blend key place facts with LLM-generated summaries
An entirely new way to Search
I worked with designers and writers and PMs on other vertical teams to make sure each of our responses felt cohesive:
Contact me for a full case study walkthrough.