Since Google introduced the Search Generative Experience in mid May pretty much every SEO and AI enthusiast has been all over the new feature trying to pry it of its secrets and learn about its inner workings.
Want early access to Search Generative Experience? We have created this short guide for you
It's a black box approach where we try to learn as much as possible without opening the box. Well, in this blog post, we go to the source, Google, to learn more about SGE.
Read more: You can dig into the 19-page PDF from Google right here
Google published a PDF about search Generative Experience, and we will dissect it and put it under the microscope to provide you with an in-depth look at the new feature based not on assumptions and anecdotal knowledge but on what we know from Google.
Enjoy!
- Google is revolutionizing search functionality by integrating generative AI technology into Google Search, referred to as Search Generative Experience (SGE). This development aims to make Search more interactive, conversational, and intuitive.
- SGE incorporates the advantages of Large Language Models (LLMs) such as MUM and PaLM2, to provide enhanced conversational interactions, catering to the specific needs of users' information journeys. Many users have already engaged with these LLMs through experimental applications like Bard.
- SGE has been designed to carry out tasks specific to Search, such as identifying high-quality web results that corroborate the information presented in the output. These models work in synergy with Google's core ranking systems to deliver helpful and reliable results.
- Google is deploying SGE thoughtfully and responsibly, following its AI Principles, and leaning on protections and approaches honed over years of experience in Search. This includes the use of independent Search Quality Raters, focused analysis, red-teaming, and human input and evaluation during training and fine-tuning.
- SGE also features a new color-coded interface to help users understand the new way of interacting with search. The color container of the AI-powered snapshot dynamically changes according to specific journey types and the query intent itself.
- SGE is designed to produce responses that are more factual than conversational, striking a balance between fluidity and information quality. It is intended to provide objective, neutral responses corroborated with web results, rather than reflecting a persona.
- Google has implemented various quality systems and robust processes for SGE to ensure reliable, helpful, and high-quality information. This includes a high standard for generating responses about certain critical topics, like finance, health, or civic information, referred to as “Your Money or Your Life” (YMYL) topics.
- SGE is designed not to generate snapshots for explicit or dangerous topics, or for queries that indicate a vulnerable situation. Additionally, it adheres to Google's policies to prevent policy-violating content from appearing in SGE.
- Despite several safeguards, there are some known limitations of LLMs and SGE. These include instances of misinterpretation during corroboration, hallucination (misrepresentation of facts), reflecting biases from high-ranking search results, and occasionally generating content that may seem opinionated or contradictory with existing Search features.
- SGE is currently an experimental feature available only to users in the US through Search Labs, as Google continues to improve its functionality and effectiveness, with feedback from users playing a crucial role in its evolution.
So how might this impact SEOs?
There is no doubt this will lead to one of the bigger changes within SEO practices.
I've tried to summarise some of them below, but you can get the full analysis here: How Search Generative Experience Will Impact SEO: 5 Insights
.svg)
