Understand Google SGE in Under 10 Minutes

A concise rundown of Google Search Generative Experience (SGE) & generative AI search in general so you can get your head around it & take action right away.

Friday, February 16, 2024
Updated:
generative AI in search

Generative search exploded in 2023 and shows no signs of slowing in 2024. This blog provides a digestible recap at this point in time to catch you up. If you’re looking for a deeper dive, check out the resources at the bottom of the page.

This blog focuses on Google’s Search Generative Experience (SGE) as it’s the most widely known and talked about generative search platform, but Bing CoPilot, Perplexity.ai and You.com provide a similar experience: a chat interface that uses AI to generate a summary answer, with links to website sources. There are probably others as well. 

The information here does not apply to Large Language Models (LLMs), which do not link to websites by default (example: ChatGPT). Google's Gemini, formerly 'Bard', is sort of in between generative search (SGE) and a straight LLM because it includes a “Google button” at the end of each response so you can verify Gemini’s answer (the verification adds citations).

What Google has said about SGE

1. Using AI in Search results is nothing new:

  • AI has been enhancing Google Search “under the hood” for decades
  • Google’s Multitask Unified Model (MUM) is an LLM that has already been deployed in “dozens of features across Search”
  • SGE is powered by multiple LLMs, including MUM and PaLM2
  • SGE LLMs are used along with Google’s core ranking systems to generate results

2. Generative AI in search helps people get more out of a single search because users can:

  • Ask more complex questions
  • Quickly get an overview of a topic
  • Use links to relevant results to explore further
  • Write drafts or generate imagery right in SGE
  • Decide what to purchase more quickly and easily
  • Compare and explore options in local searches
  • Ask conversational follow-ups
  • Try suggested next steps

3. Google has emphasized that SGE is a “jumping off point” for people to use before exploring web content:

  • “we’re committed to continue sending valuable traffic to sites across the web”
  • SGE provides links to resources supporting the information in the summary
  • Follow-up questions generate new overviews with additional links
  • Ads will continue to appear in dedicated slots
  • Ads will be distinguishable from organic search results
  • The user interface for links will be recognizable
  • SGE has been trained to identify high-quality web results corroborating the AI output

4. Safety and responsible use of AI are priorities:

  • SGE is designed not to show potentially harmful, hateful or explicit content
  • Quality analysis is especially focused on “classes of queries that may be at higher risk for safety or inclusion issues”
  • Adversarial testing is carried out by “red teams” to identify bias issues, safety concerns and other areas that can improve
  • SGE includes disclaimers where appropriate, for example health-related queries
  • SGE will not generate a response when Google’s systems have lower confidence due to information gaps
  • SGE will not generate a response for queries indicating vulnerability, e.g. self-harm queries
  • Policy-violating content will not appear in SGE
  • SGE does not reflect a persona and is designed to be less conversational on purpose, because they found that more fluid and conversational responses were less accurate, and also less likely to be flagged as inaccurate by human evaluators

Even so, SGE has known limitations and drawbacks:

  • Misinterpretation of language
  • Hallucinations (lies and mistakes)
  • Bias (because it was trained on web data which can have bias in it)
  • Reflecting the opinions someone else published - giving the impression of a persona, even though they designed it to be objective and neutral
  • Duplicating Google Search results
  • Contradicting Google Search results

5. “We are rolling out SGE thoughtfully as an experiment”:

  • Protections and approaches refined by years of use in Search are used in SGE
  • “We will continue to evolve and improve our approach over time”
  • They use human input and other quality checks to evaluate and refine SGE
  • In the first few months, Google saw the highest SGE satisfaction scores from users aged 18-24
  • In January 2024, Google said “we’ll continue to offer SGE in Labs as a testbed for bold new ideas”

6. Things from SGE will migrate over into regular Google Search results:

  • “as we continue to experiment and uncover which applications of gen AI are most helpful, we’ll introduce them into Search more broadly”
  • “we’ve only just scratched the surface of what’s possible”

Facts about SGE

SGE timeline:

  • SGE in Search Labs debuted in May 2023, only for users in the U.S.
  • Google initially had a target launch date of December 2023, which was later quietly removed
  • In August 2023 Search Labs became available in India and Japan
  • In November 2023 Search Labs became available in 120 more countries and territories, including 4 new languages

Helpful to know:

  • You can’t stop SGE from scraping your website content unless you block Googlebot completely (Google-Extended only blocks access for Gemini & Vertex AI)
  • We don’t have any data from SGE. Currently there’s no way to tell if your site is already getting traffic from SGE or how much, and Google has been mute about when/if or what type of data might become available.
  • SGE is not a default Search experience: users have to opt in
  • Only personal Google accounts can opt in to Search Labs (where the SGE experiment is live)
  • Workspace Google accounts cannot access SGE
  • SGE is limited to Google Chrome on desktop (and Google App users on Android/iOS mobile)

Using SGE

This section is brief because the best way to know what SGE is like is to try it out. It’s free. If you don’t want to use your real everyday personal Google account, just create a new one to use for SGE.

On desktop:

Log in to a personal Google account (or create one) and go to Search Labs

On a smartphone:

Download the Google app and log in with a personal Google account.

Here are the basics of using SGE (at the time of publishing). After you enter a query, SGE may or may not provide generative AI results:

  • Most often you get a generative AI snapshot right away
  • Sometimes you get regular search results with a prompt asking “Get an AI-powered overview for this search?”
  • Sometimes you get a message saying “An AI-powered overview is not available for this search”
Google SGE results asking if you would like to generate an AI-powered overview for this searche
If SGE doesn't automatically provide a snapshot for your query, you may be able to request one

Google SGE result saying that an AI-powered overview is not available for this search
Generative AI is probably denied because of the word 'blood' triggering the health safeguards (it did generate a snapshot for the same search when 'oil' or 'dirt' were used instead)

When your query results in a snapshot, at the end you’ll find:

  • AI-generated follow-up queries you can search with one click
  • A field to type in a follow-up questions
  • Thumbs-up and thumbs-down buttons for feedback

Alongside the snapshot, informational queries get a carousel of web results with an image, heading, date and brand favicon/name. Within the snapshot, little “down arrows” alongside content reveal web sources when opened.

A typical Google SGE snapshot - you can scroll right in the web results carousel to see more

Things change so quickly you’re better off using SGE than reading about what the results are like! Which brings us to the next topic: should you be worried that SGE is going to take over Search and stop traffic from going to your website?

How is SGE affecting organic website sessions?

We don't know, but we can guess.

Is your website going to lose traffic because of SGE?

It depends on:

  1. What each user is searching for
  2. What kind of results will fulfill their query

If someone just wants a quick answer and they’re satisfied with what they find in the SGE snapshot, they probably weren’t going to click through to your site (featured snippets serve pretty much the same purpose on regular SERPs) anyway. 

And if it’s a purely informational query, and you are losing a click, what you’re really losing is an opportunity to briefly put your brand in front of another set of eyeballs (or a screen reader), not necessarily a sale.

Intent is hugely important when deciding whether or not to worry about traffic erosion from SGE.

What you can do to understand more is:

  1. Assess the likely impact of SGE on your organic traffic (the brilliant Aleyda Solis provided a Google Sheet you can copy for this purpose)
  2. Put yourself in a potential customer’s shoes, think about what they’d search, and see what turns up in SGE and regular SERPs
  3. Put yourself in a current customer’s shoes and search something about your brand in SGE
  4. Analyze the results you see
  5. Start using SGE for your own searches and think about when it’s most useful to you, and why

Generative AI is impressive, but it doesn’t replace critical thinking. If you’re worried about how SGE will affect your website traffic and online presence in general, start using it so you can get a sense of what it’s most and least helpful for.

Reasons not to panic

Generative search like SGE and other AI tools were trained on content written by human beings.

If SGE doesn’t give publishers a reason to keep publishing quality content, they’ll eventually run out of fresh training material to keep developing AI tools.

Should SGE and other generative AI search tools fail to produce helpful, up-to-date results, people will stop using them. Then ad revenue will decline. Google doesn’t want that to happen.

Google is highly incentivized to encourage publishers to keep feeding the machine by producing more high-quality content that helps people get the answers they need. There are new questions being asked every day, and not all of them can be answered with existing content.

So SGE will need to link to “first generation” content to motivate companies to keep participating in their ecosystem by creating content by and for humans.

Yes, it will take time and effort to adapt content strategy and SEO to the new realities of search. But that’s why this work is awesome, because it’s never boring!

<div class="post-note-cute">If you think I got something wrong or left out something vital, I’d appreciate hearing about it! If you have questions about SGE’s potential effect on SEO and want help navigating these new waters, we’re here for that, too. Whatever your feedback or question, don’t hesitate to reach out: <a href="mailto:info@momenticmarketing.com">info@momenticmarketing.com</a></div>

Resources

We used the following resources to create this article:

  1. An overview of SGE [Google PDF]
  2. Supercharging Search with generative AI [Google blog]
  3. Bringing generative AI in Search to more people around the world [Google blog]
  4. Generative AI in Search expands to more than 120 new countries and territories [Google blog]
  5. New ways to search in 2024 [Google blog]
  6. Chat-Search Optimization, your first step is monitoring
  7. SGE Impact Assessment V.1 [by @Aleyda]

Here are some additional resources for a deeper dive into SGE & AI:

  1. Google SGE: Google Search Generative Experience Explained [Semrush blog]
  2. The 6 Best AI Search Engines To Try Right Now [Search Engine Journal]
  3. Get AI-powered overviews & ask follow ups on SGE in Search [Google documentation]
  4. Learn about generative AI [Google documentation]
  5. Making AI Helpful for Everyone [Google AI]
Bar chart showing increase over time with Momentic logo

About the author(s)