Update 20.02.2016: The sad fact about Similarweb is not that sad anymore. Happy to announce, that since mid February Similarweb has introduced two major changes to their product:

1. Now free Similarweb includes both desktop and mobile web data in their visits and engagement metrics.

2. Additional information including breakdown by desktop / mobile web and daily visits by platform is available in Pro.

Congrats, Similarweb – very useful update. And pretty good timing :)


A few months ago I blogged about Similarweb doing great job in web competitive space. I still use it on a daily basis and recommend for web marketers. But there is one thing that casts a shadow on this awesome tool and gets more and more annoying as web becomes mobile-first.

Mobile Web Not Included

This one thing is the fine print that can easily lead you to draw the wrong conclusions about other websites. See below:

Similarweb - Desktop tickbox

By default in Similarweb mobile web is not included.

When I compared Brainly websites to competitors I was surprised that Similarweb tend to heavily underestimate traffic of our platforms. Typically it assessed our traffic to be around 50-80% of its real size, in some cases going extremely down to 20%. C’mon Similarweb, we are better than this!

Well, my bad that I took for granted that Similarweb estimates total traffic of the website in its free version. Information box stating that it shows “Estimated number of internet users who visited a particular website over the last six months” even increases the confusion. But if you read their blog or pay attention to small tickbox above the graph it is clear that what Similarweb is measuring in its free version is only desktop traffic. For mobile web you need an Enterprise account.

That’s kind of disappointing taking into consideration the global web trends. According to We are social report published in January 2016 there is already around 39% of global web traffic coming from mobile devices (users browsing Internet on their mobile devices). Having a piece of information about remaining 61% is still nice, but you can easily miss a big picture.

In case of Brainly, globally around 45% of traffic is mobile web with some markets like Indonesia with over 70% visits coming from mobile devices. So if you want to grasp the real scale of any given platform you must keep in mind, that it can be two or more times bigger than what you get from basic Similarweb.

Study Setup


To quantify how big the difference can be I decided to run a similar research to the one conducted by Rand Fishkin in mid 2015. This time I took eleven platforms of Brainly Group and compared how close you can get to their real traffic with basic Similarweb.

Data Overview

The websites were relatively big ranging from 300k visits to 32M visits monthly. Countries covered by these platforms include: USA, Brazil, Portugal, Russia, Ukraine, Kazachstan, Turkey, India, Indonesia, Philippines, Romania, Poland and France. One of the website that I intended to include was not yet available in Similarweb due to renaming and domain change that happened in late September.  Time period included 4 months: September – December 2015. As a result I got 10 websites x 4 months = 40 data points with three dimensions: Similarweb estimation, Real total traffic, Real desktop traffic.


I’ve calculated four metrics:

1. Avg of Similarweb Estimation / Actual traffic (aka Accuracy)
Representing how close you are to the real value using Similarweb estimations.

2. % of data within 70-130% of Actual traffic (aka Precision)
To find out how often you are close to the real values.

3. Spearman’s Correlation with Actual traffic
To see how well the estimated data follow the real traffic changes.

4. Data coverage
Showing for how many of the websites I wanted to check the data was available in Similarweb

Study Results

Below you can find the results of this brief study:

Accuracy of Similarweb

Using Similarweb to estimate Brainly websites’ total size you would end up on average at 58% +/- 17% of its real size (where 100% is a real size). Estimations never got even to 85% of the real total traffic. Not good enough.

In case of estimating desktop web size – Similarweb was much more accurate. On average it got 118% +/- 28% of the real size. Also the distribution of the results was pretty healthy with slight bias towards higher numbers.

Concluding, Similarweb did the job pretty well in case of desktop traffic. It is much much better than 406% in Rand’s sample.

Similarweb Precision

Following Rand’s methodology, I calculated in how many cases Similarweb got into 70-130% of the real range. In case of total traffic it was only 27.5% but when it comes to Desktop only – 60% of estimations got that close to the real number. Nice! In both cases it is better than 22% in former Rand’s research.

Correlation Similarweb

Time for correlation. In this case the average correlation coefficient was high for both: total and desktop traffic, however surprisingly it was higher for total traffic (0.87) than desktop (0.81).

When I looked into details I realized that in fact Similarweb was better at tracking changes in traffic for desktop (which makes sense) however due to two outliers, the average number was significantly decreased. The two outliers were: Brainly.com (US) with correlation coefficient at 0.05 and Znanija.com (RU) – 0.21. Others were between 0.95 – .99. In the table below you can find all the coefficients.

Website Total Desktop
Brainly.com 0.62 0.05
Zadane.pl 0.91 0.95
Znanija.com 0.43 0.21
Eodev.com 0.98 0.99
Brainly.com.br 0.96 0.97
Brainly.in 0.89 0.97
Brainly.co.id 1.00 0.98
Brainly.ro 0.97 0.98
Brainly.ph 0.99 0.99
Nosdevoirs.fr 1.00 0.98
Average 0.87 0.81

Let’s discuss the data coverage:

Similarweb Data CoverageAs mentioned previously out of 11 websites I intended to include in the sample, only one missed the full data from the period. It translates to 91% data coverage. I would consider this result good. It was also slightly better than in the previous research. 

To sum up, pay attention to the fine print. Basic Similarweb is here only to assess the desktop traffic and to some extent – to analyse general traffic changes of a given platform. When it comes to the total traffic of a website – expect the real numbers to be even few times higher. 

Any ideas about how to estimate the missing mobile web traffic?