Why ranking signals do not correlate
How do ranking signals work in a modern search engine and why do we not always find the correlations that we would expect?
Background: How does a search engine work?
There are many misunderstandings about how a search engine works. Many people think that a search engine looks at words at any level and that the person with the most relevant words gets a high position in the right places. Although that sometimes quite matches the SERPS, that is not the case. A modern search engine looks at the context within an article and tries to guess whether the content is suitable to answer the search query. A search engine searches for the best content for a search query. Technically we are looking for the article with the shortest distance to the ideal article. The distance is initially determined by the distance between the contents and the ideal content (the worldview) and then adjusted by dependent ranking signals such as backlinks. In the influence of ranking signals such as backlinks, therefore, it is not always the same and smaller for good content than for poorer content.
In a simple formula you could write this as pos = d (p, q) + (1-sum (d (p, q)) * factors / weight
Background: how do we substantiate our ideas?
Some time ago we decided to build a Content Tool. Not just a content tool but a content tool that analyzes the SERPS and can extract the ideal content from it. How do you do that? You start by building your own search engine based on the latest AI techniques. Our search engine understands language, can answer questions about any topic and gives results that can be compared with a 'real' search engine (we only take a little longer, but that doesn't matter). That is the basis for our Content Ingelligence. We noticed a number of things while building the search engine. The most important conclusion was that there are 3 types of ranking signals and they work differently than you might expect.
Type of ranking signals
We think there are 3 types of ranking signals. Independent ranking signals, dependent ranking signals and worldview ranking signals.
#1 - Independent ranking signals
Independent ranking signals are ranking signals that can directly determine a position in the search results. In fact, independent ranking signals consist of 100% content such as the text on the page, headings and the title.
#1 Good content
#2 Bad content
#3 Mediocre content
#2 - Dependent ranking signals
Dependent ranking signals are ranking signals that depend on the 'independent ranking signals' such as backlinks, page speed and website quality. The quality of the content determines the maximum influence (the playing field) that this ranking can receive signals.
#1 Good content, low influence other ranking signals
#2 Slechte content, high influence other ranking signal
#3 Mediocre content, moderate influence other ranking signals
In this example you can see that the possible influence of dependent ranking factors such as # 2 backlinks can be twice as large as those of #1
#3 - Worldview ranking signals
Worldview ranking signals are signals that adjust the worldview of a search engine such as pogo sticking, personalization and search history. Worldview signals determine what a search engine thinks a user would like to see.
Perhaps that is not the most extensive and technically the best article, but rather the article that summarizes all the facts in a short, concise, somewhat funny and readable way.
#1 Good content, low influence other ranking signals
#2 Bad content, high impact on other ranking signals
#3 Mediocre content, low impact of other ranking signals
In this example you can see that because the distance to the world image of # 1 is smaller, the possible influence of dependent ranking factors such as backlinks of # 2 can be 2.2x as large as that of # 1.
What about correlation?
Hopefully you have already seen it, the influence of possible ranking factors, in our model, depend on the quality of the content and the distance to the world view. Do you write great content that also perfectly matches what the user wants to read? Then you have almost no other ranking factors more, while they have been.
Take a look at our example. # 1 and # 2 secretly have just as many independent ranking factors, but the influence of these factors varies dramatically.
Then you consider that there are certainly 100 ranking factors that will work indirectly, all of which depend on each other (if you pay a lot of attention to good content, then the chance is higher that you also pay attention to link building) and the influence of this per search, may vary per item and per person. Then you may understand why ranking factors, which are secretly an important factor to rank, will never rise when you look at the SERPS.
How do you find correlations?
In our model we find a .7 correlation with content, worldview and the SERP positions. Note: that does not mean that content and worldview explain 70% of your ranking. We have only found one trend, no explanation! But you can say that the quality of the content and the distance to the world view explain much more than all the separate ranking signals combined. If you then try to explain the ranking differences by normalizing the 'playing field' you will see clear and recurring correlations between the different ranking signals.
Above the fold
Black hat seo
Advanced Search operators
Ranking Signal correlation
Time to first byte
First Contentful Paint
Largest Contentful Paint
Smart WebFont loading for better performance
Icon fonts lazy loading
Improve page rendering with content-visibility
Analytics without Core Web Vitals delay
Self host Google fonts tutorial