How accurate is Similarweb? What the studies show

How accurate is Similarweb? Accurate enough to tell you which of two sites is bigger and whether a site is growing. Not accurate enough to tell you how many visits a site actually gets. Published comparisons against real analytics put the typical miss for Similarweb, Semrush and Ahrefs at somewhere between a fifth and a half of the true figure on any single site, and sometimes far more, while their ordering of sites by size is much more reliable.

So: compare with them, trend with them, and never quote them as visits. Below is how each tool makes its numbers, taken from the vendors' own methodology pages, and what the studies you can open yourself actually found. Everything here was read on 4 October 2026.

Three ways to estimate someone else's traffic

Nobody outside a company can see its analytics. Every traffic tool infers visits from something else, and that something else decides where the numbers break.

1. Panels and clickstream

A panel is a large group of people whose browsing is recorded, with consent, through apps and browser extensions. The tool counts how many panellists visited a site, then scales that up to the whole internet population.

Semrush's Traffic Analytics reports, part of its Traffic & Market toolkit, work this way. Semrush says they draw on "petabytes of clickstream data" from "over 200 million real internet users and over a hundred different apps and browser extensions", processed with its own machine learning. In December 2023 Semrush also took a majority stake in the clickstream provider Datos, according to its own press release.

The strength is coverage. A panel sees direct, referral, social and email visits, not only search. The weakness is who is in the panel. A small site gets few panellist visits, so scaling multiplies the noise, and people who never install consumer apps or extensions, such as staff on locked-down work laptops, are likely to be under-represented. That is one reason B2B sites are hard.

2. Rankings multiplied by click-through rate

The second method estimates organic search traffic only. Take every keyword a site ranks for, multiply each keyword's monthly search volume by the share of clicks a result in that position usually gets, and add it all up.

Ahrefs describes its Organic traffic metric in exactly these terms, with click-through rates estimated from third-party data. Semrush's Domain Analytics figures are "estimations based solely on keyword positions and their estimated traffic (search volume X average CTR)", which is why Semrush warns that its two traffic numbers "will almost always report different estimations".

Every input is itself an estimate. Search volumes are modelled. Click-through rates swing with whatever else is on the results page: ads, a map pack, an AI Overview. And the tool only counts keywords in its own database, so long-tail searches go missing.

3. Blending many sources, including ISPs

Similarweb lists four sources: direct measurement from "millions of websites and apps" that share first-party analytics such as Google Analytics; a contributory network of its own consumer products that collect anonymous device data; partnerships that include "internet operators (ISP's), measurement companies, and demand-side platforms"; and public data it indexes itself. Machine learning models then blend and calibrate the lot.

The sites that share their analytics are the ground truth. They give Similarweb real numbers to calibrate against, so you would expect estimates to be better for sites that resemble those contributors than for unusual ones.

What the accuracy studies found

Five comparisons you can open and check. None is perfect. The samples are sites whose owners shared their analytics, which may not look like the sites you care about, and one study was run by a vendor about its own product.

The peer-reviewed one. Jansen, Jung and Salminen compared 86 sites from 26 countries. Similarweb reported 19.4% fewer visits and 38.7% fewer unique visitors than Google Analytics on average, with bounce rate 25.2% higher. Yet when the 86 sites were put in order of size, the two agreed closely (a correlation of 0.954 for visits). Order right, level low.

SparkToro. In Rand Fishkin's study, 641 sites shared a year of Google Analytics data. Semrush correlated most closely with it (0.790), then Datos (0.720), Similarweb (0.659) and Ahrefs (0.504). On landing within 30% of the real figure, "SimilarWeb is the clear winner with one exception": sites under 5,000 monthly visitors, where it was worst. Ahrefs was "almost always underestimating"; Semrush was "much more often over than under". Fishkin's conclusion: "no 3rd-party estimate today is consistently accurate enough to place high confidence in their numbers." Datos became a Semrush company the following year.

Ahrefs on Ahrefs. Ahrefs' own study compared US organic estimates for 1,635 random sites with Search Console: a median deviation of 49.52% for Ahrefs against 68.36% for Semrush. Read it as a vendor marking its own homework, but note the honest line: "For some websites, we are off by less than 5%. For some others, we can be off by more than 1,000%."

Promodo. The original study reports an average error of about 50% across all three tools; the per-tool figures above come from Collaborator's write-up of the same research. Semrush overestimated on 112 of the 184 sites, and Similarweb improved after a system update in late July 2024.

How accurate is Similarweb, then?

Pulling the studies together:

• Ranking sites by size: good. A 0.954 correlation in the peer-reviewed study.
• Visits to one site: often out by a fifth to more than a half, sometimes far more. The PLOS ONE average was 19.4% low; the 2024 organic comparison averaged 56.95% error.
• Small sites: nothing, or noise. Similarweb does not display data when a site's latest month is under 5,000 visits.
• Trend over level. Similarweb's own accuracy page says it expects "trend alignment" with your analytics, not matching numbers. That page also quotes the SparkToro study in its favour, which is fair as far as it goes; the same study found Semrush correlated more closely overall.

Semrush and Ahrefs sit in the same zone with opposite habits: Semrush tends to overestimate, Ahrefs to underestimate.

Where estimates break

Small sites. A few panellists or a few keywords decide the whole number. Under roughly 5,000 visits a month, treat any figure as a guess.

Brand-heavy sites. Most visits come from people typing the brand or the address. A ranking model applies an average click-through rate to a brand search, while the real share of clicks on your own name can be far higher, and direct visits never enter a ranking model at all.

B2B and niche sites. Many keyword volumes are too small for the tools to report, panels are thin among office workers, and buyers return through email and bookmarks the models cannot see.

Sites that live off other channels. Newsletters, apps and logged-in products are invisible to ranking models and patchy in panels.

Sudden changes and tool updates. Models smooth, so a launch or a penalty shows up late. And when a vendor changes its model, every site moves at once. If all your competitors drop in the same month, it was the tool.

How to use traffic estimates safely

1. Calibrate on a site you can see. Look up your own domain and compare the tool with Google Analytics (for total visits) or Search Console (for organic clicks) over the same months. If it reads you at 60% of reality, it probably reads similar sites low too. Repeat for two or three sites you have access to, such as clients'.
2. One tool, one metric, one country. Never set Similarweb's number for one rival against Semrush's for another.
3. Use ratios and direction. "Roughly three times our organic traffic, rising for six months" survives a 50% error. "41,200 visits" does not.
4. Match like with like. Organic estimates against Search Console clicks; total visits against analytics sessions.
5. Label it in every report. Tool, country, month, and the word "estimated". The benchmarking rules explain why mixing estimates and observed facts misleads.
6. Prefer what you can observe. New pages, live ads and review counts are facts. Traffic is a model.

For the step by step of looking a rival up, see how to check competitor website traffic. For the tools side by side, see the website traffic checkers roundup and Semrush vs Similarweb.

Where Figo's traffic numbers come from

Ours, so judge accordingly. Figo's estimated traffic uses the second method above: keyword positions and the search volume behind them, for the country set on your workspace. It is organic search only, it is an estimate rather than visits, and it shares the same weak spots, especially on very small sites where a handful of keywords swings the figure.

That is why Figo labels it an estimate, to be read as a size and a direction, next to things it reads directly, such as new pages, ads and review counts. The data methodology page marks which numbers are measured and which are estimated. If you need total traffic, including direct, referral and app visits, a panel-based tool such as Similarweb is the right one, and Figo does not try to replace it.