How accurate is Similarweb? What studies found

Someone quotes you a traffic number for a competitor, or for a directory or launch site you're thinking of paying to list on. It came from Similarweb, Ahrefs or Semrush. Before you act on it, know what kind of number it is.

We read how each tool builds its figures, then every independent comparison we could open in full.

Short answer: Similarweb estimates traffic; it doesn't count it. Its numbers are good enough to rank big sites against each other and not good enough to measure a small one. Two tests that scored tools against Search Console found typical errors near 50%, and the two studies that split results by size found small sites fared worst. Similarweb says it is more confident above 100,000 monthly visits, and its platform shows no number below 5,000. Treat any competitor figure as a range, and trust a site owner's own analytics over every estimate.

Where Similarweb gets its data

Similarweb's Data Methodology page lists four sources (read 7 October 2026).

• Direct measurement. Sites and apps that share their own analytics, such as Google Analytics. Owners can connect it privately, to compare themselves with the market, or publicly, so anyone who looks the site up sees the owner's numbers.
• Contributory network. Consumer products that collect anonymous behavior data from devices, which Similarweb's 2024 update notes call global panel data from millions of users. The page doesn't name the products.
• Partnerships. Organizations that collect "digital signals" across the internet, including internet service providers, measurement companies and ad-buying platforms.
• Public data extraction. Automated crawling of public pages and apps each month, plus census data such as population by country.

Those inputs are blended and fed to machine-learning models that scale a sample up to a whole-site estimate. None of the Similarweb pages we read gives an exact panel size, an error rate or a list of the consumer products.

Its accuracy page is more candid: "we don't expect our estimations to align exactly" with your own analytics, and it expects "trend alignment" instead.

One more fact matters. Similarweb's July 2024 data version added a new method for smaller sites, better bot filtering, a model adapted to GA4 and a five-year re-run of core metrics.

Why small sites are the least accurate

Similarweb works from a sample. Its own guidance says that for sites with a small number of visits the estimates are less statistically reliable, and that it is more confident above 100,000 monthly visits. Below 5,000 it stops showing numbers: if a site's latest month is under 5,000 visits, it won't display any data, and the platform shows "Not enough data" or N/A. An older help article says its API returns a value anyway.

Ranking-based tools have a small-site problem of their own, in our reading. A small site's traffic hangs on a handful of keywords, so one missing query or one wrong click-rate assumption moves the total a lot. And a new site's early visits often come from a launch, a newsletter or a social post, which a tool that only models Google clicks can't see.

What independent comparisons found

We list only studies we could open in full on 7 October 2026, with sample size, date and who ran each.

• Jansen, Jung and Salminen, PLOS ONE, peer-reviewed, 27 May 2022. 86 sites with Google Analytics connected to Similarweb, Sept 2019 to Aug 2020. Similarweb's visits were 19.4% below Google Analytics on average: lower on 66 sites, higher on 15, within about 5% on 5. The two ranked the sites very similarly (.954). Keep in mind: the sites were picked by working down the Majestic Million list (median 7.8M visits a month), five outliers were dropped, and it measured an older data version.
• Omniconvert, a conversion-optimization vendor; page updated 5 June 2026. 1,787 eCommerce sites, from 4,000+ that granted Google Analytics access; the period isn't stated. Similarweb's sessions were about 94% above Google Analytics, closest above 100,000 sessions a month and least reliable below 10,000. Keep in mind: eCommerce only, and the page doesn't say which months or data version it covers.
• Ahrefs (Tim Soulo), 3 May 2022. 1,635 random sites, US Search Console clicks against US organic estimates. Ahrefs' median deviation was 49.52% and Semrush's 68.36%; correlations with Search Console were .76 and .74. Some sites were under 5% off, some over 1,000%. Keep in mind: Ahrefs ran it on its own tool, it covers Google search traffic only, and Similarweb wasn't tested.
• Promodo (agency), 30 Oct 2024, with per-tool figures in Collaborator's write-up, 5 Dec 2024. 184 sites, first half of 2024, Search Console as the reference. Average error: Ahrefs 48.63%, Similarweb 56.95%, Semrush 61.58%. Semrush ran high on 112 of 184 sites. Keep in mind: the write-ups don't say which metric from each tool was compared.
• Screaming Frog (Patrick Langridge), 13 June 2016. 25 sites, UK organic visits in Google Analytics, Feb to Apr 2016. In total Similarweb was +17%, Ahrefs -17% and Semrush -30%; signed per-site averages were +1%, -36% and -42%. For the three sites under 10,000 visits, nearly every estimate was 40% or more off. Keep in mind: it's ten years old and small; the author calls it "by no means a thorough mathematical or scientific experiment."

What the five have in common:

• The error runs both ways. Similarweb read low on the large sites in the PLOS ONE sample and high in Omniconvert's, Promodo's and Screaming Frog's. Sample, period and data version all differ, so there's no single "Similarweb runs high" rule to carry around.
• Size matters. Omniconvert and Screaming Frog found the worst numbers on small sites, which matches what Similarweb says. The exception: Collaborator's write-up of the Promodo test says its anomalous misses didn't depend on a site's size or niche.
• Rankings hold up better than totals. In the PLOS ONE sample the two sources ranked the 86 sites very similarly (.954), and Ahrefs found .76 between its estimates and Search Console. Ask "which of these is bigger?" before "how big is it?"
• Typical error is large. Ahrefs found a median deviation near 50% and Promodo an average error near 50%. With a median, half the sites were worse.
• Models change. The 2016 and 2019 to 2020 results describe an older Similarweb, and independent tests since its 2024 update are thin. It upgraded its channel data again in April 2026, recalculating 37 months.

Two studies aren't in the list. Similarweb's accuracy page says a SparkToro study found it the most accurate of the tools tested against Google Analytics, strongest for sites with 5K to 100K users a month. SparkToro's site blocked our automated requests, so we couldn't read it or quote its numbers; a critique on the 99signals blog questions its design. Ahrefs cites an Authority Hacker test of six tools on 50 sites, but that page now redirects to the homepage. Treat both as secondhand, not our findings.

Are Ahrefs and Semrush organic traffic estimates accurate?

About as accurate as Similarweb's, for different reasons. Ahrefs' own study found a median deviation of 49.52%, and Promodo's test put Ahrefs' average error at 48.63%, the lowest of the three tools. Their organic numbers aren't panel-based; they start from rankings.

Ahrefs describes the method in three steps: find the keywords a site ranks for, estimate the clicks from each one using its position, the keyword's monthly search volume and an estimated click-through rate for that position, then add them up. The result estimates clicks from Google. Semrush's keyword-based tool, Domain Analytics, uses the same ingredients: search volume times average click-through rate. Semrush calls the output how much traffic a site "COULD get" from its positions.

Semrush also sells a second, panel-based estimate. Its Traffic & Market data comes from a panel of over 200 million anonymized users and hundreds of clickstream providers, combined with its own backlink and ranking databases. Its knowledge base says the two tools "will almost always report different estimations of traffic".

What a ranking-based estimate misses, in Ahrefs' own words:

• Keywords it doesn't track. Ahrefs tracks the most popular ones; it calls tracking all of them technically impossible.
• Search volume and click-through guesses. Its volumes blend Google Keyword Planner and clickstream data, which isn't exact. "Only Google knows the real numbers of clicks," Ahrefs says, so click rates are modeled too.
• Everything that isn't a Google click. Direct visits, referrals, social, email, paid and other search engines. Google Analytics counts those; a ranking model can't.
• Timing. Ahrefs' monthly figure averages daily estimates from each day's rankings, so it moves when rankings move.

Ahrefs' help page says these estimates can't be exact but "work incredibly well for comparison".

Why Similarweb and Ahrefs disagree on organic traffic

If you've put the same domain into both and got different organic numbers, that's expected:

• Different yardsticks. Similarweb's Organic Search channel counts unpaid clicks from search engines as one slice of modeled total visits. Ahrefs estimates Google clicks from rankings.
• Different timing. Similarweb publishes monthly data by the 10th of the following month. Ahrefs updates as rankings change.
• Different failure modes. Similarweb struggles when its panel sees little. Ahrefs struggles when keywords are missing or its click-rate assumptions are off.
• They overlap at keyword level. Similarweb's search data also uses a click-through model, which it says draws on its panel, Google Ads and Search Console data.

For the products side by side, see Similarweb vs Ahrefs and Semrush vs Similarweb.

Which number to trust for what

How to sanity-check any estimate

1. Calibrate on your own site first. Pull 90 days from Search Console and your analytics, then look up your own domain in the tool. The gap is the tool's error for a site like yours. Ahrefs suggests applying the ratio between your own Search Console traffic and its estimate to competitors in your niche. It's rough; keep it to a range.
2. Use two tools and a range. If they disagree by more than about 2x, treat the number as unknown. That's our rule of thumb, not a published threshold. In Ahrefs' test the median site was about 50% off, so an estimate of 10,000 could mean roughly 5,000 to 20,000, and half the sites were further off.
3. Look for measured numbers. Similarweb shows Google Analytics sessions for sites that have connected publicly. Some founders publish live analytics; Plausible, for example, lets owners make stats public. Otherwise ask for a dated screenshot showing the date range and filters.
4. Match the metric. Visits, sessions, users, pageviews and Google clicks aren't the same number. Search Console counts clicks from Google Search only; analytics counts visits from every source. When you compare your own GA4 with Similarweb, remember its mobile data includes tablets and it counts one domain at a time. Semrush itself says to use Google Analytics, not Semrush, for your own site.
5. Cross-check popularity. Tranco is a free research list that averages several popularity rankings over 30 days (see our glossary entry). By our rule of thumb, a site in the top 100,000 is almost certainly real and visited, and one missing from the top million is small. It's one input, not a traffic count.
6. Ask before you pay. If a number is driving a sponsorship, directory fee or newsletter ad, ask for analytics first. A seller who can't share a screenshot is telling you something.

We apply this on our own pages: our directory pages show a Tranco rank band rather than a visit estimate, and when a site has no rank they say we can't tell you how much traffic it gets. If you'd rather have directories chosen for you, that's our directory submission service.

A quick free look, with the same caveat

Our free website traffic checker estimates a domain's monthly Google organic traffic from its rankings, by country, with the keywords that bring the most. The data comes from DataForSEO. It's a model, in the same family as the Ahrefs and Semrush organic numbers above, so use it for a first range and check it the way this post describes. Our traffic estimates glossary entry is the short version. Weighing Similarweb itself? See LaunchRanked vs Similarweb.