Der Schachtel-Shop – Effects of User Signals, RankBrain and more

This box shop is a small but nice shop in Munich, which specializes in the sale of special gift packaging. All products can be ordered also over online Shop, which is accessible under .  With this analysis, we would like to look at how the shop is doing with regards to its positioning in search engines, what effect the measurable user behaviour has on the Google ranking, what influences RankBrain shows and what opportunities result from it.

Note: As the source of the data used in this analysis, we make use of and Google Analytics – unless otherwise noted.

Let’s first take a look at the Sistrix visibility gradient. Here it becomes clear that especially the start page of the shop (main directory) can generate many rankings: 


In addition, we find rankings of documents located in subfolder 2016 and blog. However, this differentiation is not sufficient for us at this point. We want to know more about it in order to better understand what is going on here. First we want to know how many rankings there are for the documents in which subfolders (first level).

As already seen with Sistrix, especially the folders 2016, root and blog achieve many rankings. We now want to check how they are distributed in the top 100 on Google, which is why we now take a look at the following streamline diagram, which spans a line from the first to the last of the four measuring points for each ranking:

Open example in the Lab

As we can see, the first 10 places (blue) are primarily generated by the start page. Starting from the third search result page we find many rankings of the blog articles from the year 2016 (green). The following figure filters out exactly these ones:

Open example in the Lab

We should ask ourselves why so many rankings are subject to such a sharp and categorical devaluation. What is the principle behind it? What commonality is responsible for this? If the common ground could be uncovered, then an overall increase in these rankings might be conceivable

Visit duration and page views

So, we first want to check whether the documents differ in terms of user behavior. Therefore we create two paralles coordinates axes (red) with the parameters page views and average time on the page (seconds) at the left side of the rankings in the diagram. In this way, we can see which ranking is characterized by which parameter value; e.g.: whether good rakings can only be achieved with a higher number of page views. However, after a detailed examination we can state: there is no clear pattern which could explain the bad rankings in connection with the described parameters

Open example in the Lab

We now want to show the rankings of the start page:

Open example in the Lab

As you can see here, the homepage is accessed much more frequently than the blog articles from 2016. The question, however, is whether the number of page views is really such a crucial factor for the ranking. If we take a look at the visit duration of the pages, the start page achieves much better rankings, sometimes with much shorter visit durations. If we were to assume that identical rating rules are applied below the domain, we could conclude that the visit duration is not a significant ranking factor.

Open example in the Lab

Overall, we can only state that content that is consumed between a few seconds and several minutes achieves good rankings. We couldn’t find any good rankings for overlength visits. 

Loading time

We would like to look at another parameter: the loading time – already communicated years ago as an official ranking factor by Google. To do this, we add the corresponding parameter axis to the left of the rankings. In the following diagram we look at the loading characteristics of the documents that achieve top 10 rankings (152 items):

Open example in the Lab

None of the top rankings are generated from documents that are delivered slower than about 2.5 seconds. Now let’s consider the opposite case:

Open example in the Lab

Documents that take longer than 2.5 seconds to be delivered do not lead to a top ranking in our analysis. It can therefore be said that speedy delivery can be seen as a basic prerequisite for good rankings. However, good loading times alone do not guarantee good rankings, as can be seen again in the following example of the 2016 blog articles:

Open example in the Lab

Here another aspect can be observed: the additional axis size (bytes) shows how much the data transfer depends on the file size. In practice, however, the greatest delays seldom occur during transmission, but when the server prepares the data. It is interesting to note that smaller files often take longer to load than larger files.

Distance to start page: Depth

In practice, it can often be seen that documents that can be reached from the start page with fewer clicks receive better rankings than documents that are further away. In the following figure, we again add the structural depth information to the left of the diagram:

Open example in the Lab

Here it can be seen very well that more “distant” documents indeed achieve lower rankings. The blog articles from the year 2016 can be reached with at least 2 clicks – the shop that can be reached with one click has significantly more top rankings. However, since we also determine bad rankings for nearer pages, we have to state again: the distance between the documents cannot be deduced as the exclusive or main cause.


The bounce rate is often assumed to be an important indicator of visitor acceptance. If the visitor leaves the page immediately, it is assumed that the page could not sufficiently meet the needs of the visitor. We therefore have to look at how the bounce rate and the visit duration relate to the ranking development:

Open example in the Lab

All ranked blog articles from 2016 have a high bounce rate (up to 85%), which is certainly not unusual for blog articles. The detected duration times also do not seem to be particularly noticeable. Again, the question arises as to whether and if so, to what extent these values affect the ranking.


Let’s now check whether the number of page entries (direct access via external link or search engine) affects the ranking performance. To do this, we only look at the top 10 rankings again.

Open example in the Lab

As can be seen well, such documents can also achieve good rankings, which – at least in comparison to some others – are accessed much less often directly.

Open example in the Lab

Many blog articles from the year 2016 are accessed more often directly (again green) than others that achieve better rankings. The number of entries does not seem to be a decisive factor for the ranking.

Text length

A very common question in connection with the optimization of content is the question of the ideal text length. How much content should be offered if good rankings are to be achieved? Let’s take a look at the following figure. More than half of all rankings are generated by documents containing between 1,300 and 1,600 words:

Open example in the Lab

However, this does not mean that one could not achieve good rankings with more or less words.

Open example in the Lab

At the Schachtel-Shop we find documents containing between 600 and about 3,400 words per document, all of which achieve top 10 rankings. So there is no recommendation for the ideal number of words. It depends largely on which words are written there.


Of course the incoming links from external sources still have a big influence on the ranking. Therefore we want to take a look at the situation at the Schachtel-Shop:

Open example in the Lab

That was not to be expected differently: the homepage has the most independently referring domains. But we should also take a look at the internal linking – an often underestimated ranking factor:

Open example in the Lab

The figure above shows that our poorly ranked blog articles have very few incoming links. But: even documents with fewer links are able to generate top rankings.

Open example in the Lab

This also applies to the number of internal links: Documents with few incoming, internal links are also able to reach top positions. Thus the links also seem to be an influencing factor that we have to keep in mind, but they cannot explain the sharply defined ranking structure that we observe with blog articles.  

IN vs. OUT

What is the situation with the documents that entered the top 100 in the observation period and those that left the top 100? Can we find here any perhaps illuminating characteristics that could better explain the standard of evaluation? Why do certain documents ascend and descend? The following figure shows all rankings that leave the top 100 in light blue. Newcomers appear in red.

Open example in the Lab

First of all we can state that within the observation period more rankings are eliminated than new ones are added. However, with regard to the parameters, no coherent pattern can be identified that would qualify or downgrade a document. This is in line with the observations already described.

Semantics and RankBrain

However, perhaps the effects of the semantic classification (see also Restriced Area) by RankBrain can be observed in the sharp ranking demarcation. To check this, let’s look at the area above the rankings (slot 1), which is drawn up above the blog articles from 2016, and the area below (slot 2) and compare the content. The terms marked in red do not appear in the comparative/other list:

It is interesting to note that the context “geschenkbox” (gift box) achieves very good rankings. For all search phrases that stand in the context “verpacken” (pack) and “basteln” (tinker) (i.e. the do-it-yourself aspect), not a single good ranking exists. These terms appear particularly frequently in search results from rank 18. In the following we check this again with a text filter:

Open example in the Lab

The figure shows that in the context of “verpacken” (packaging) not only the blog articles do not achieve good rankings, but also others, such as the homepage (blue). So it seems that Google does not consider the context “verpacken” to be adequately appropriate – an effect that can be related to the effects of RankBrain.

Final evaluation

In the analysis we were able to uncover or exclude some important correlations. The interaction data of the users could not quite explain why a large part of important search phrases could not make it to the top ranks. They rather show a contradiction, because the articles are found, accessed and read via the search engine despite bad rankings. In addition, we found a few structural peculiarities that could support the observed ranking structure. However, the most important explanatory variable is RankBrain, which associates the context “verpacken” (packaging) for the Schachtel-Shop only as a second choice.


How can we now further increase the rankings of the blog? In any case, we should look at the structural aspects revealed by this analysis:

  • Ensure speedy delivery of all documents.
  • Important documents should be quickly and easily accessible.
  • The internal linking should be revised or intensified.  
  • Promotions with external links are never wrong.

If these measures are not able to adequately convey the meaning of the blog content, it may be necessary to consider outsourcing the content, e.g. to a (sub)domain. If you look at the current top results at Google on this topic, we will only find offers that sell the gift along at the same time. Or there are pure tinkering guidebooks, which make appropriate guidances available. A mixture of packing and instruction supplier, as is the Schachtel-Shop, just does not exist on the top ranks with Google in the context “verpacken” (Packaging).

Analyze yourself

Would you like to check all the parameters discussed here again in their entirety for correlations? Then open our lab with all already prepared analysis parameters.

Open Lab with all parameters