Comparing accuracy: SEMrush vs SimilarWeb (2024)

One ofthe most common marketing analysis practices ismarket benchmarking. Let’s suppose your marketing team was 5 percentage points short offulfilling the plan. Ifthe market during the period inquestion shrank by15%, then your marketing team may still have shown apositive result. But ifthe market grew bythe same volume, then your marketing isn’t going sowell.

There are alot oftools for competitor benchmarking. Each ofthem collects traffic data and processes itdifferently, however. And you can’t expect 100% accuracy from them. Given that, which tools are sufficiently accurate soyou can use the results toperform meaningful analysis?

Weconducted some research tofind out how precise two popular traffic analyzers — SEMrush Traffic Analytics and SimilarWeb — are. Asa baseline for comparison, weused data from Google Analytics accounts of787 anonymized websites observed byOWOXBI.

Before weshare our results, let’s consider how traffic analysis services work.

SimilarWeb uses the following data sources:

• Direct data from website owners
• Data from partners and internet providers with millions ofusers
• Public data sources: patented technologies and indexing mechanisms that are constantly scanning public data
• Anonymized panel data from browser extensions onvisited websites

SEMrush Traffic Analytics reports are also based onanonymized data gathered from third-party sources and processed with the help ofartificial intelligence algorithms. This data iscollected and approximated based onanonymized behavioral data onmillions ofinternet users.

Over the last 10 years, SEMrush has launched afew tools that reveal not only data oncompetitors’ search positions but also what competitors are doing intheir paid search, PR, content marketing, and social media along with detailed data ontheir website traffic. SEMrush uses the most relevant data sources for each marketing sphere (SEO, PPC, content, SMM).

Both SEMrush and SimilarWeb can beused for evaluating competitors, leads, and traffic generated bypartners. Aswe have access tothe anonymized Google Analytics data ofOWOX BIusers, weassumed wehad reasonably precise data onvisitors for anumber ofanonymized websites. Wethen compared data provided bySEMrush and SimilarWeb for those same anonymized websites against the Google Analytics data. Our task was todescribe the inaccuracies inparticular business spheres and see how large they are for each service.

The data sample

Our sample consisted ofdata collected from Google Analytics, SEMrush, and SimilarWeb for all sessions inJanuary 2020 oneach of787 websites. These websites are based inAustralia, Canada, the US, the UK, and Germany and were grouped bybusiness sphere:

1. Computers, electronics, and technology
2. Entertainment
3. Finance
4. Health and beauty
5. Jobs and education
6. News and media
7. Professional services
8. Retail
9. Telecom
10. Travel

Comparison approach

Tofind out the accuracy ofcompetitor traffic calculations bySimilarWeb and SEMrush, wemerged the following data into one table:

• Anonymized Google Analytics data from 787 websites with more than 100,000 visitors per month according toOWOX BI
• Data ontraffic from those same websites according toSEMrush and SimilarWeb

Weexcluded websites with low traffic that might becaused byfilters for properties.

Then wecalculated the absolute value ofthe deviations between data provided bySEMrush and SimilarWeb compared toGoogle Analytics.

Asdeviations might bepositive ornegative and wewere interested only inaccuracy, weused the absolute value. This also saved usfrom apotential mistake with calculations inwhich positive and negative values could result inzero.

Then wesegmented our websites bynumber ofsessions per month:

• 100,000 – 500,000
• 500,001 – 1,000,000
• More than 1,000,000

For websites with 500,000 sessions ormore, the numbers were 9 to12 percentage points more precise inSEMrush. For projects with little traffic, SimilarWeb worked slightly better, but both services demonstrated large inaccuracies among this group ofwebsites.

These differences inaccuracy are caused bythe collection and processing algorithms ofSimilarWeb and SEMrush aswell asspecial aspects ofclickstream data. Traffic analysis services use artificial intelligence and machine learning algorithms toapproximate data onall website visitors based onclickstream data, which isdata onsamples ofwebsite traffic. Thus, the smaller the website, the less accurate the approximations based onclickstream data.

What should you doif your website and your competitors’ sites have low traffic and the accuracy ofyour data isextremely low? Inthis case, you should benchmark against bigger competitors inyour market. Ifyou compare some big market players, you’ll see not only their performance but also the general market trends. And bycomparing performance and trends revealed bySimilarWeb orSEMrush with your own achievements, you’ll beable tosee the efficiency ofyour marketing.

The graph below describes the share ofobservations — the percentage ofwebsites within each segment for which SimilarWeb and SEMrush respectively were closer tothe Google Analytics benchmark. For example, inthe 1,000,000+ sessions segment, SEMrush gave more accurate data than SimilarWeb for 57% ofanalyzed websites:

Bycomparing the 100,000 to500,000 sessions segment inthe Standard deviation and Share ofObservations graphs, wediscover aninteresting insight: SEMrush data has ahigher standard deviation, which tells usthat the session approximations are generally less accurate. Atthe same time, SEMrush isstill more precise in53% ofcases. Long story short, SEMrush makes fewer mistakes but the mistakes itdoes make tend tobe massive.

The accuracy ofdata depends ona variety offactors:

• How Google Analytics isset up, which pages have GAtags, and what exactly those tags measure.
• How active the website ison Google. For aredirect website like from anadvertising network ora promo website with massive paid adtraffic, SEMrush will show lower numbers.
• How big the share oforganic traffic isfor the website. SEMrush islikely more precise for sites with alarger share oforganic traffic.
• The region and business sphere. Let’s suppose SEMrush has more projects from the US. Inthis case, itwill have more data from the USand calculations for USsites will bemore accurate.

The next two graphs demonstrate the standard deviation and share ofmore accurate data from both services segmented bybusiness sphere.

Asyou can see, the deviation ofboth SEMrush and SimilarWeb depends onthe business sphere:

For example, inthe Computers sphere, SimilarWeb was more accurate for 58% ofwebsites and SEMrush was more accurate for 42% ofwebsites (the first column inthe graph above).

This dot graph shows the positive and negative deviation values for SEMrush (blue dots) and SimilarWeb (green dots):

Wefound that:

• The accuracy ofSimilarWeb and SEMrush issimilar.
• SEMrush showed better results for medium-sized websites. Mistakes were rare, and ifmistakes were avoided, the data was more accurate.
• Inthe 1,000,000+ sessions segment, SEMrush was more accurate than SimilarWeb.
• SimilarWeb ismore likely tounderestimate traffic volumes.

What’s important toremember isthat neither SimilarWeb nor SEMrush will guarantee you 100% accuracy. Toanalyze your own website data, you have Google Analytics. But SimilarWeb and SEMrush are enough for independently comparing websites and recognizing trends. When using any analytics tool, however, you should understand the origin ofthe collected data and the deviations ofmeasurements.