How Accurate Are Website Traffic Estimates? (The Real Error Range)

Every free website traffic estimate is a model, not a measurement, and on a great many sites the model is out by a factor of two in one direction or the other. Across the 11,242 listings in the Revised directory as of 16 September 2026 we publish no traffic figure at all, because we cannot measure one honestly and neither can anyone who lacks the site owner's analytics. The largest public comparison checked four providers against Google Analytics on 641 real sites and found the gap was often 100% or more. Three input families produce these numbers: clickstream panels, search-volume modelling, and the site's own public data. Use an estimate to order two sites. Do not quote it as a figure.

From SparkToro's comparison of third-party traffic estimates against Google Analytics, published 22 November 2022 and still the largest public test of its kind:

For some ranges of traffic, some providers are pretty good. But, no 3rd-party estimate today is consistently accurate enough to place high confidence in their numbers.

How far off are website traffic estimates?

SparkToro asked site owners to volunteer their Google Analytics data, then bought the same sites' figures from four providers. After cleaning, 641 websites remained, giving 7,692 site-months.

The correlations against Google Analytics' reported Users metric, on a scale where 0 is no relationship and 1.0 is perfect:

Ahrefs sits last for a reason that is not accuracy. Its figure only tries to estimate organic search traffic, and the study notes it correlated at roughly 0.75 against Search Console's own numbers.

Correlation says how well the estimates track reality across a dataset. It does not say how wrong one number can be, which is the figure worth carrying around:

As you can see, that answer is often +/-100% or more, meaning a 3rd-party might say that website XYZ received 50,000 visits in June, but it actually got 5,000 or 100,000.

One caveat. The study is from November 2022 and every provider has changed its model since. Nobody has published a comparable independent replication, so read the table as the shape of the problem rather than as this month's accuracy.

Where does a traffic estimate come from?

Three input families, and knowing which one you are looking at tells you its failure mode.

Clickstream panels. A provider buys or operates a panel of browsers and devices, records the pages those users visit, then scales the sample up to the whole web. Semrush's own documentation puts the boundary plainly: "The central difference between our service and Google Analytics is that Google Analytics is an internal service for studying your own website, while Semrush is an external service for studying another company's website."

Search-volume modelling. No panel here at all. The provider keeps an index of search results, works out which keywords a domain ranks for, and multiplies each keyword's search volume by an expected click rate for that position. Ahrefs describes its own version that way, then tells readers how to use the output: "Remember to treat the organic traffic estimations in Ahrefs as precisely that: estimations." They "work incredibly well for comparison".

The site's own public data. A few providers fold in what the site publishes itself. This is the only family with measurement in it and it covers almost no sites.

Our website traffic checker is in the second family, and the tool page names the provider and shows no figure at all when the index does not answer.

Which sites are traffic estimates worst at?

Small sites are the first problem. A panel that sees a few thousand visits a month to a site is working from a handful of panellists, and a handful of panellists is noise.

Non-US audiences are the second, because most keyword indexes behind these tools are US English, so a non-US or non-English site reads low by a margin nobody publishes. Sites whose traffic is not search are the third and catch people out most: direct, email, social, referral and app traffic are absent from a search-volume model, so a site with a mailing list and no rankings can read close to zero and still be busy. Logged-in and paywalled sites are the fourth, because a crawler sees only the wall.

Why do two tools disagree by multiples on the same site?

Because they are often not measuring the same thing. One reports users, another reports visits, another reports organic clicks only, and comparing those three is a category error before accuracy comes into it.

Panels and indexes differ after that. Two providers with different panels, countries and device mixes scale up to different totals from the same reality, and a domain ranking for rare terms is invisible to a smaller keyword index and visible to a larger one.

What can you measure rather than estimate?

Three things, all of which need access you either have or do not.

Search Console reports clicks and impressions for a property you have verified, and its definitions are counts rather than models: a click is "How often someone clicked a link from Google to your site" and an impression is "How often someone saw a link to your site on Google". Our guide to what Search Console shows about your links covers the rest of that report. Server logs count requests that reached your server, bots included, and first-party analytics counts the sessions its script managed to record.

None of the three works on somebody else's site, and none works on a domain nobody owns. That is the whole reason the estimate industry exists.

How does this change on an expired domain?

A search-volume estimate for a dropped domain reads a site that has already gone. The keywords behind it are the ones the index last saw the old site ranking for, and those rankings do not survive the pages under them. Google's crawling documentation is explicit about a domain that stops resolving:

For Google Search, the lack of content means that Google can't index the crawled URLs, and already indexed URLs that are unreachable will be removed from Google's index within days.

So the estimate is a lagging record of what used to rank, not a forecast of what you will get. It is worth reading for one narrow purpose: a dropped domain with a wide keyword footprint had a real site on it about a real subject, which tells you something about the link profile you are buying. Our guides to finding expired domains with traffic and what happens when a domain expires cover the free ways to read that footprint, and the index test covers what to check once you own the name.

Using the Revised directory for this

1. Run any domain through the website traffic checker and read the keyword count before the traffic figure. It moves slowly and describes the breadth of a site's search footprint, which makes it the more stable of the two.
2. Compare two domains in the same tool rather than reading one absolute number. Both are scored by the same model, so the ordering is worth more than either figure.
3. Open the directory and sort on the referring-domain band instead. As of 16 September 2026 it holds 11,242 listings, the median sits in the 10 to 50 band, and 47 have 250 or more. Referring domains are counted, not modelled.
4. Start from aged domains when a long site history is what you care about. That page holds 409 listings, each at least five years old, with a median band of 100 to 250 referring domains and 325 of the 409 passing a low spam screen.
5. Use the Open filter to work with named domains for nothing. 3,373 of the 11,242 listings, about 30%, publish the name and the exact counts with no account. The free plan reveals 10 masked names a month.
6. Verify a shortlist with the backlink checker, which lists each referring domain with its source page, anchor text and link type, then check status and archived site history in the expired domain checker.

Every listing is spam-screened before it goes up, and 9,541 of the 11,242 live listings pass a low spam screen. We grade a listing on referring domains, site history and link-graph standing. Traffic is not in that grade, because it left with the site.

FAQ

How accurate are website traffic estimates? On a large share of sites, not accurate enough to quote. The biggest public test found the gap against Google Analytics was often 100% or more, so a stated 50,000 visits could be 5,000 or 100,000. Accuracy is better on large sites.

Which traffic checker is the most accurate? Semrush correlated best with Google Analytics in the 2022 SparkToro comparison at 0.790, ahead of Datos, Similarweb and Ahrefs. That was one study on one year of data, and no provider is accurate enough to treat its number as a measurement.

Why does Similarweb show different traffic from Semrush? Different panels, different countries and often different units. One may report users where another reports visits, and a search-volume model reports only clicks from search.

Can I see another site's real traffic? Not unless they show you. Real counts come from Search Console, server logs or first-party analytics, and all three need access to the site.

Do expired domains have traffic figures I can trust? No. The estimate is computed from keywords the index last saw the old site ranking for, and Google removes unreachable URLs within days. Read the referring domains and the archived site history instead.

Is a traffic estimate useless then? No, it is useful for ordering. Comparing two sites inside one tool, or tracking one site's direction in that tool over months, is what these models do well. The failure is quoting a single number as a fact.