This blog includes our findings on Google's Search Generative Experiment as of April 2, 2024, providing a snapshot of the ongoing developments of the experiment.

Google’s Search Generative Experience experiment proved to be their biggest and boldest implementation of generative AI in search so far. 

Because of its potential to be one of the most disruptive SERP features in the history of Google, I’ve been following it closely since it was introduced as a Google Labs experiment in May of 2023.

 For now, it’s still unclear if Google has any immediate intention of releasing it into public search (more on that below). But SGE still has much to tell us about the impact generative AI technology may have on search in the near future. 

In this post, I’ll review what SGE is, its developments since we entered 2024, the takeaways we garnered from it, and what the future may hold for generative AI in search.

Note: On May 14, Google integrated AI Overviews (AIO) into public search, previously tested in SGE. Explore the impact of AI overviews on SEO, ways to measure it, and additional insights.

Table of Contents:

 

What is Search Generative Experience (SGE)?

If you’re already familiar with SGE, skip to the next section.

Search Generaritive Experience was Google’s first major response to the overnight revolution which was the introduction of ChatGPT in late 2022. 

Bing’s unveiling of its own generative AI-based search feature in March 2023 probably intensified the pressure on Google to come up with something to compete.

But when SGE was finally unveiled in May 2023, it was restricted to a new Google opt-in segment called Google Labs. To use SGE, one had to sign up for Google Labs using a personal (non-business) email address. 

 

SGE and Generative AI

As mentioned, SGE makes use of generative AI, the same technology that powers ChatGPT. 

In simple terms, generative AI is a computer algorithm trained on a large set of content (Large Language Model). Using machine learning, the generative AI algorithm detects and maps the statistical relationships between words and groups of words. 

This gives the program the ability to analyze a user prompt and then predict the words most likely to occur in response to that prompt. The user can interact with further prompts or queries. The end result looks similar to a human conversation.

 

SGE in Google Labs Search

If a user is signed in to Google Labs while using Google Search, they will often be offered
“an AI-powered overview” for the search. Sometimes it appears as just the offer with a “Generate” button that must be clicked to reveal the SGE response.

Other times the SGE response will be partially revealed, with a button to reveal the full response. 

 

The layout of a typical SGE response has changed somewhat over time. Currently, it often appears like this:

 

Previously, there was a carousel of suggested sites to the right of the main response. But now, the suggested sites for the first section of the response typically open by default. (Otherwise, to see the suggested or source sites for each section, you have to click the down arrow at the end of the section.)

These sites seem to be the cross-check sources, used to verify and correct the initial SGE response, as described in this Cornell University paper that several Google scientists contributed to.

Each section of the response below typically has a down arrow that reveals the confirming sources for that section.

Following the text response, there is a follow-up section where the user can either select one of the suggested questions or enter their own, leading to a new SGE response.

 

SEO Concerns About SGE

Many SEOs were very concerned about the potential impact of SGE on organic search traffic for the following reasons:

  • When fully expanded, SGE results take up a lot of SERP real estate, pushing the regular results far down the page and reducing their visibility and potential for clicks.
  • If the SGE result satisfies the user’s need, they will be less likely to click on any of the recommended pages in the result or any of the organic search results beneath it.
  • The recommended pages in SGE results had very little crossover with the top-ranking content in organic search, making it difficult to know how to “rank” in an SGE result.

But perhaps, for now, they don’t have much to worry about.

 

What Happened to SGE in 2024?

When SGE was introduced, we were told the experiment was scheduled to end for US users in December 2023. Many in the SEO community assumed it would be rolled out into public search soon after that.

But December came and went and SGE was still running in Google Labs only.

On January 17, 2024, Google published a blog post to unveil two new ways to interact with Search (Circle to Search on Android devices and multi-search which allows you to take a photo with your phone camera and ask questions about it).

At the end of the post, Google also revealed the current status of SGE: For at least the foreseeable future, it will be kept in Google Labs “as a testbed for bold new ideas.”

It appears that, for now, they are going to concentrate on smaller, more limited applications of AI in regular search. That being said, there have been some recent hints that SGE may be released in a very limited version in the near future (see below).

So why wasn’t SGE made a public part of Google Search after nearly a year of development and testing?

 

Why Google Search Generative Experience Is Not Ready for Prime Time

Of course, only Google knows why they chose not to make SGE public, But after extensively observing and testing it for almost nine months, I think I can offer some possible reasons.

  1. Search intent Mismatch. It’s fairly common that SGE results seem to respond to a different intent than regular search. 

For example, if you enter the search query “booking hotels,” Google’s first page is entirely results from the site booking.com:

Why is that the case? Over time, Google observed that most people who searched for “booking hotels” clicked on a result for booking.com. In other words, the actual intent of this search is what we might call “branded navigational.”

But here’s the result for the same query in SGE:

SGE interprets the query as someone merely wanting to know how to book hotels and provides several resources for accomplishing that task.

As we saw above, Google already knows that’s not what searchers are looking for with this query. But SGE seems unaware of that valuable information.

  1. Lack of Authority. SGE also doesn’t seem to make use of the authority signals Google has refined to determine the best pages to show for a given query.

While at least one or two of the sources shown in SGE typically correspond to pages ranking in the top ten on regular Google, the majority of them don’t rank so well. 

Here’s an SGE result for “how to check a used car before you buy it”:

The first source is a United Arab Emirates site similar to Craigslist, and the second is information about buying cars in Nova Scotia, Canada. 

There are very good reasons why neither of those are top-ranked results in Google for US car buyers.

This lack of authority vetting for sources raises a concern about the reliability and value of SGE results. 

  1. Not as Useful. SGE results often lack the sophisticated filters and rich result information that we now see in transactional and other results in regular Google.

Here’s what we see in regular Google Search for the query “sunglasses”:

These days, transactional queries for products are shown with a sophisticated array of filters, selectors, and image carousels to create the kind of online shopping experience people have come to expect from major retailers.

But look at the SGE result:

We get a pretty useless block of text about what sunglasses are, a carousel of clickbait “X best…” pages, and a random assortment of category tiles. While this is better than what SGE shopping results used to be, why would anyone choose this experience over what regular Google already offers?

Aside from the quality issues noted above, there are at least two other reasons why Google appears to be dragging its feet on releasing SGE into public search:

  1. Generating these results is expensive. It takes exponentially more processing power to create a generative AI result as compared to traditional results.
  2. If shown for queries that show ads, the SGE response may push down or draw attention away from the ads, impacting Google’s revenue.

 

Is SGE Useful for Anything?

There is only one category in which we’ve observed SGE results being potentially more useful than regular Google results: informational queries.

Not surprisingly, generative AI is pretty good at summarizing existing information and organizing it for the user.

Take for example a search for “how to buy a bicycle”:

SGE creates a very nice buying guide with lots of useful information organized to help the user find exactly what they need.

And just announced as I’m writing this, SGE in Google Labs can now build you a complete travel itinerary, with places to see, stay, and eat, mapped out day by day.

Here’s my result for “plan me a three-day trip to Philadelphia that’s all about history.”

If you don’t like anything about the itinerary, you can request changes with a follow-up question. For example, I asked it to make the restaurant recommendations more budget-friendly and it republished the itinerary with more economical places to eat.

When you like the itinerary, you can export it to Gmail, Google Docs, or Google Maps. This is what it looks like in Maps:

 

What’s Next for Google Search Generative Experience?

Given the Google blog post of 17 January 2024 and the difficulties I outlined above, it was looking like SGE might remain a Google Labs-only feature indefinitely.

Until two things happened.

First, in mid-March, Google announced Elizabeth Reid as its new Head of Search. Apropos of our subject, it turns out Reid worked closely with the SGE team.

Then in her LinkedIn post announcing her new position, there was this:

Reid devoted about a fifth of her inaugural announcement to SGE, ending with “more coming soon!” We have to give that some significance.

Then came this announcement on 22 March 2024:

Google said they were beginning to show some SGE results to a limited number of users who had not opted into Google Labs. This certainly indicates that they have not lost interest in SGE and are continuing to test how it might be a useful addition to Search.

But remember how I showed above that it is really only useful and (fairly) reliable for informational queries?

Notice in the pull quote from this article that they will be starting only with queries where they think SGE “can be especially helpful,” and those queries “involve questions where it may be useful to get information from a range of web pages…”

I think that is the most significant aspect of this announcement and the most overlooked. I’ll explain why that’s important in the next section.

NOTE: SEO Eli Schwartz claims to have inside knowledge that some form of SGE will be released into public search at or shortly after the Google I/O event on May 14, 2024. We have no way of confirming that, but if they do intend to make a release this year, that would certainly be the time and place to announce it.

 

What Will Be the Impact of Google Search Generative Experience and What Should You Do?

Our extensive testing and observation of SGE has led me to believe that it will be limited to informational queries for a long time to come. 

After almost a year of development, Google has not made significant strides in improving SGE results for transactional and local searches. Why would they introduce something into Search that makes Search less useful?

For these reasons, I do not see generative AI in search as a threat to organic search for the searches that actually make businesses money, at least not for some time.

Recommended Reading: Generative AI's Impact on SEO

 

What About Those SGE Impact Studies?

You may have seen a number of studies in recent months claiming that SGE will impact anywhere from 20-40% of organic search traffic. But you need to keep in mind that those studies are all based on SGE as it appears in Google Labs. 

Google Labs is a testing ground. It makes sense for Google to use it to throw all the spaghetti against the wall, so to speak, and see what sticks.

I don’t think we’ll ever see SGE in public search to the extent we see it in Labs.

For that reason, I do not put much weight on the predictions in those studies.

It may, however, be more of a challenge to get traffic for your top-of-funnel content – the kind of content where you try to provide helpful resources on topics related to your business. 

As such, it may be useful to review that content and make sure it conforms to practices that make it more likely to appear in AI-generated results such as:

  • Make sure your content is clear, on topic, and structured in a way that makes it obvious what each section is about (i.e., good use of headlines and sub-headlines.)
  • Use schema where appropriate to enhance the clarity of the contents of your pages. (TIP: use our free tools at schema.dev to make implementing and validating structured data a lot easier!)
  • For many more tips, see this excellent post by Garrett Sussman.

 

A Final Note: Don’t Panic!

Given how cautious Google has been about beginning to unveil SGE, I feel confident in saying that its impact will be much smaller than some others have been predicting. 

Of course, that doesn’t mean it won’t have any impact if it does start appearing in some form in regular search.

If informational, top-of-funnel content is an important part of your organic search strategy, then consider implementing the recommendations above.

If you’re an seoClarity client, we are fully prepared to measure and report the impact of SGE. If and when it begins appearing more broadly. You’ll be able to assess its effect on your efforts (if any) and we’ll continue to supply you with updates and recommendations as this feature develops.