Earlier this week, I had the opportunity to witness Gary Illyes from Google in a Firechat at SEODay in Denmark.
I've previously mentioned some of Gary's insights, but this time, there were many more takeaways for SEO marketers on how rankings in Google work and what links are getting featured in Search Generative Experience (SGE).
1. Clicks in SERPs DO impact ranking positions
I think this was one of the most interesting parts.
Gary did not want to be too specific (and as he mentioned there is currently a lot of focus on this due to the ongoing antitrust trial) but made some clear statements.
- Contrary to popular belief, Google does NOT use [Google] analytics insights from websites to affect rankings
- Google does NOT use CTR - but they do look at "clicks"
- The search results are not impacted by the ad results. It's two different compartmentalized teams.
The second point Gary Illyes made is particularly intriguing.
He stated that while Google does not use Click-Through Rate (CTR) in the traditional sense, they do pay attention to "clicks." This suggests that Google is more interested in user behavior on the Search Engine Results Page (SERP) than just the raw CTR data.
This could mean that Google is looking at more nuanced data, such as the relevance of the clicked link to the search query, the user's subsequent actions after clicking, or even the time spent on the resulting page.
Interestingly, this could also involve analyzing if the user is coming back to Google quickly after clicking on a result, a behavior known as "pogo-sticking."
Pogo-sticking occurs when a user performs a search, clicks on a result, quickly comes back to the search results, and clicks on a different result. This behavior indicates that the initial page did not satisfactorily answer the user's query. High pogo-sticking rates can signal to Google that a webpage is not meeting the needs of searchers for a particular query.
This approach would allow Google to gain a more comprehensive understanding of user behavior and satisfaction, which could be used to refine and improve their search algorithms.
In essence, it's not just about getting a high CTR, but about ensuring that the clicks your site receives are meaningful and result in positive user experiences. This insight underscores the importance of not just optimizing for search engines, but more importantly, for the users themselves.
2. Using semantic keywords works
This is more my interpretation of what he said.
There was a question about the SEO Heist that the SEO world has been buzzing about lately, although he wasn't sure if it resulted in a manual penalty. At least he would not say :)
Gary mentioned that it's too easy to generate content these days, and while search engines might like some of the content that is produced users often find it misleading.
Search engines might like, but users will find that it is wrong.Two paragraphs in you think what the hell am I reading
- Gary Illyes
He stated that there are many ideas internally [in Google Search Team] on dealing with this. One of the elements currently in play are models trained by Spambrain to catch some of these low-level, low-quality content pieces.
But here is the quote I think most people should take notice of:
It [some of the low level] doesnt look bad from a computer perspective. Computers don't understand content. They are mapping it to something. There is no real understanding like you and me understand it.
- Gary Illyes
And this leads me to my sub-heading conclusion.
As algorithms understand it from a computer perspective, we cannot just rely on making the best piece of content for user.
Of course, as just mentioned, clicks and user behavior of good content will be something that the algorithms understand and judge the content based on. But the right keywords and keyword semantics are still initially at play along with backlinks etc. That is how computers understand and evaluate content initially.
3. Site Authority as a Ranking Signal
Gary used Forbes.com as an example (although he was not super keen on using an actual site as an example).
If they wrote about something, even something that would normally not be associated with Forbes like “SEOday”, then they would still rank due to their site authority.
In the SEO world we already know this - but also have an understanding that it's not only general site authority but also topical authority that influences ranking.
By the sound of what Gary mentioned, general site authority might be a bigger factor than Google would like compared to topical authority.
4. Search Generative Experience (SGE) results are not based on the top 10 results
So I believe this was quite interesting and maybe the biggest takeaway.
The SGE results are not just based on the current top 10 results according to Gary.
This is also aligned with the findings we earlier covered about SGE.
So to me, this instead sounds like when ChatGPT provides an answer on something based on its training data. And then Gary said, that for each answer they look for content matching that answer and then display that.
It's [the SGE answer] not coming from that site. We will link to what is similar.
- Gary Illyes
I believe this is why it's not the current top 10 that is being displayed as links and the process is like this:
- Google is getting a query from the user
- Google AI is generating an answer based on existing training (so no search yet)
- Google then matches the answer it just makes to all the content in its index and displays links to the element having the highest match (just like an RAG setup does)
What that means for us SEOs is that SGE implementation might be a threat if you have a lot of existing page #1 rankings, and we should focus more on how to optimize for having elements that are similar to the answers, and not just the query itself.
Google will not provide data insight into SGE answers
In a follow up questions, someone from the audience asked how Google might provide data-insight into SGE so that SEOs can use it.
The short answer is they won't.
SGE is just a visual element in the SERP.
- Gary Illyes
Gary said that SGE is to be compared to any other SERP feature, like Featured Snippet or People Also Asked, which Google does not provide data insight for.
Don't publish AI content without a human in the loop
Gary expressed some reservations about AI.
He believes that current language models can't be published without supervision. In his belief, they just create statistically likely words and content (in which he is right), but think they need supervision to avoid producing nonsensical results.
Although I think he underplayed the quality generative AI platforms can create - both text and images - I agree that in most cases you need some sort of human in the loop to get the best results. And even though it requires more effort, you will get a better return in the form of better SEO rankings.
To that note, Gary also mentioned a more personal perspective or a piece of good advice.
That is crucial to attract traffic that one can convert. Otherwise, you're wasting resources in producing this.