Google’s Liz Reid explained on the Bloomberg Odd Lots podcast how AI Mode and AI Insights enable detailed, needs-based query models that create new challenges for Google. This indicates a consequent change in search behavior that has a direct impact on the way we approach SEO.
Keyword Fragmentation in AI Search
Liz Reid explained that users always wanted to express longer queries in natural language, but were forced to narrow them down to keywords like “best restaurants in New York,” even though what they really wanted might have been more specific, like a restaurant with vegan options and open for a party of five.
Since I work in SEO and have been in the industry for almost 30 years, keyword research is the foundation of digital marketing. You choose the keywords you want to rank for, then create content in a way that is optimized for that keyword. The problem with optimizing for a short keyword is that there are hidden meanings in that keyword and that has always been the case.
The way Google has used the issue of latent meanings in keywords is to use things like clicks to better understand what users meant when they typed ambiguous keyword phrases like “restaurants in New York.” Some SEOs believe that clicks were used to rank websites, but another use of clicks is to understand what people mean when they type ambiguous phrases. What Google has been doing for a while now is ranking the most popular meaning of the keyphrase first and, regardless of how many links a page receives, if the content matched a less popular meaning, the page would not rank.
Liz Reid said people who use AI-powered search use longer queries that articulate the problem or need for information, making it easier for Google to retrieve the information they’re looking for. This change goes to the heart of the organic search problem that AI search is solving and the implications for SEO are profound.
Liz Reid begins:
“We’ve seen with AI insights much longer queries. We’re seeing more natural language queries, but it’s not even something as basic as that.”
It can also be like searching for restaurants. We used to laugh about that stuff before working on search, I was working on maps and local information, some intersections with search, and people would just say “New York restaurants.”
And you ask yourself, what do you want me to do with this request? Like, okay, the best restaurants in New York are going to take three months and 99.9% of the population can’t afford to go there.
Okay, but do you choose 10 randomly, etc.? ?
But one of the reasons people would do that is because they had a much more complex environment: I want a restaurant in this location for five people. It can’t be too expensive. I have a vegan member. I also have children. That was the question on their minds.
And in the old world of keywords, this information would be spread all over the web. And so you wouldn’t be sure if you could just ask the question.
And now with AI Previews and AI Mode, you can start doing that, and you see people doing that, they tell you the real problem, right?
They don’t take their needs into account and translate them into what the computer understands. They try to give the computer what it actually needs and expect us to do the translation.
The big ideas to unpack are:
- A typical complex question asked in AI Search may not be answered by a single web page.
- Complex questions can be one-off and rarely, if ever, repeated, which in many cases can reduce the value of optimizing these sentences because the time spent writing them could be more cost-effectively spent on something else.
- Since a site will likely share AI Overviews (AIO) space with another site, this increases the need to optimize other factors such as brand icons that stand out in a positive way, the use of relevant images, and even the use of videos to claim as much AIO space as possible.
- And yet, perhaps the most important takeaway is that it’s not all long tail, because Google breaks long tail phrases down into smaller, very specific keyword phrases that reflect some of the information need, query distribution, and returns them to classic search. Google’s AI then selects from the top three for each query and uses it to synthesize a response.
So it’s not really that SEOs should be optimizing for long-tail queries, because query breakdown uses classic search, boiling it all down to the specific queries that web pages are relevant and optimized for.
Meet real needs
Reid didn’t go into detail on this point, but it’s still interesting because she said that the process of splitting a complex natural language query into smaller queries becomes a quality issue. One of the problems with AI Search is that people don’t search with the same keyword phrases, which means Google can’t cache similar queries the same way it can with organic search.
She explained:
“I think that means you have to do it, it’s a harder job on quality, right?
You need to answer this question, it has many parts, and you need to figure out how to break it down. And you have to work to think about things like latency, because you can’t, you know, if everyone is using the same keyword and it’s not personalized, then you can cache everything. If all of a sudden the queries become much more diverse, you know, that has consequences.
But I think we just see that it gives a lot of power to people, right? This simplifies the search.
A few years ago they asked: What more can you do with Google Search? But if you actually ask them, okay, when was the last time you spent 20 minutes searching when you would have rather spent 2? It’s actually not that difficult for me. …And so it was kind of exciting to just…make people’s lives easier by helping them meet their real needs.
At first glance, the idea of meeting real user needs seems like one of those useless slogans like “be awesome” or “content is king.” But it’s actually a way for every SEO to audit web pages. Rather than limiting their scope to keywords, titles, and technical issues, look at how they fill a need.
Today someone asked me to check out their website which was having trouble getting indexed. They suspected it might have been a technical problem. My answer is that yes, everyone hopes it is a technical problem, but in many cases, especially the one I was looking at, the problem becomes obvious when looked at from the perspective of the question: “what need does this page fill??” as well as asking: “How is this not only different from another page, but different and better?»
Watch Liz Reid’s interview here:
Google’s Liz Reid on Who Will Own Search in an AI World
Featured image by Shutterstock/TierneyMJ
