How to Find Competitor Website Traffic: 10 Sources and How Accurate Each Is

Start with the distinction that decides how much any of this is worth. Analytics platforms count. They sit inside a page, fire on load and record a real session. Traffic estimators infer. They watch a panel of consenting browsers, buy clickstream data, crawl search results, and extrapolate from that sample to the whole internet. Both produce a number labelled “visits” and only one of them counted anything.

That does not make estimators useless. It makes them instruments with a known error, which is a perfectly workable thing to own as long as you know the error. What almost nobody researching this question does is bound the estimate against something that was actually measured, and there are two such datasets available free.

1. 1Decide what the traffic number is for. Sizing a competitor, tracking whether they are growing, or judging whether a channel is worth entering are three different questions. Only the first needs an absolute figure, and it is the one the data supports worst.
2. 2Pick one estimator and stay inside it. Each vendor models traffic differently, so a figure from one and a figure from another are not comparable. Choose the tool you will use for the year, and treat every number that follows as belonging to that tool's scale rather than to reality.
3. 3Bound the estimate with the two measured datasets. Check where the domain sits in the Chrome User Experience Report rank buckets and in the Cloudflare Radar rankings. Neither gives a visit count, both are built from real observed activity, and together they tell you the order of magnitude an estimate has to respect.
4. 4Read the channel split rather than the total. Where the traffic comes from decides what it means. Heavy paid share means growth they are renting; heavy direct means brand; heavy organic means a content position that took years and will take years to displace.
5. 5Cross-check against anything they published themselves. Press releases, investor materials and advertising media kits carry visitor numbers the company chose to stand behind. They are self-reported and often generous, and they still bound the estimate usefully in both directions.
6. 6Convert everything to a ratio before you report it. Express the result as a multiple of your own traffic in the same tool, not as a count of their visits. Systematic bias that would ruin an absolute number largely cancels out of a ratio, which is why ratios survive and counts do not.
7. 7Log the tool, the date and the range. Record which tool produced the figure, when, and the bounds you established. Re-check quarterly, and immediately after a competitor launches a campaign or a major content programme.

There is one substantial public test of this, and it is worth knowing in detail because everything else on the page follows from it. SparkToro, run by Rand Fishkin, collected real Google Analytics data volunteered by site owners and compared it against the estimates four providers published for the same sites and the same months. The analysis, published in November 2022, covered 641 websites and 7,692 monthly comparisons.

Two things in that table matter more than the individual scores. The first is that errors of 100% or more were common: an estimate of 50,000 visits was regularly attached to a site doing something closer to 5,000 or 100,000. The second is subtler and more useful. The provider that ranked first on correlation is not the provider that ranked first on absolute accuracy, because those are different questions. Correlation asks whether the tool gets the ordering and the direction right. The 30% band asks whether the number is close. A tool can be excellent at one and mediocre at the other, and the answer to “which traffic checker is best” therefore depends entirely on which of the two you need.

One honest caveat about the study itself. The underlying analytics data runs from mid-2020 to mid-2021, so the specific correlation figures are dated and the products have all changed since. The structural finding has not: these are panel-based models, their error grows as sites get smaller, and no vendor has since published a comparable independent test against real analytics data to replace it.

Neither of these will give you a visit count, and both are more trustworthy than anything that will. They are built from real observed activity at enormous scale by two companies who see a large slice of the web for reasons unrelated to selling you a subscription, and both publish the result free.

The Chrome User Experience Report

Google publishes a public dataset of real-user experience data collected from consenting Chrome users. Alongside the performance metrics it carries a popularity rank, derived from the total number of navigations to an origin, and expressed in magnitude buckets rather than an exact position: top 1,000, top 5,000, top 10,000, top 50,000, top 100,000, top 500,000 and top million. It is queryable through Google’s public data warehouse, and community-maintained snapshots of the ranked lists are published for download if you would rather not write a query.

Two readings are useful. Which bucket a competitor sits in is a hard floor and ceiling on their traffic, drawn from Chrome navigations rather than from a panel of a few million people. And movement between buckets is a real event: a domain crossing from the top 500,000 into the top 100,000 has genuinely changed size. A competitor absent from the dataset altogether is below the traffic threshold required for inclusion, which is itself informative when a tool is quoting them six figures of monthly visits.

Cloudflare Radar domain rankings

Cloudflare operates one of the largest public DNS resolvers, so it observes a very wide sample of the internet resolving domain names. Radar turns that into a free domain popularity ranking, published as daily top lists and as weekly bucketed datasets running from the top 200 out to the top million. Their ranking metric counts unique client networks rather than raw queries, which stops one large organisation or one noisy resolver distorting a domain’s apparent reach.

Because it derives from a completely different mechanism than Chrome navigations, it is a genuinely independent second opinion rather than a correlated one. When both datasets place two competitors in the same bucket and an estimator insists one is four times the other, the estimator is the one to doubt.

1. Similarweb

Type a domain into the public page and you get an estimated monthly visit total, a trend over recent months, the channel split across organic, paid, direct, referral, social and email, top countries and top referring sites, with no account needed for the basics. The channel split is the part worth your attention. On the accuracy evidence it is the strongest of the estimators in absolute terms and it degrades badly on small sites, so read it with the site’s size in mind.

2. Semrush

Similar coverage with a different model behind it, and the provider that tracked real analytics most closely on the correlation measure in the published comparison. That makes it a good default when what you need is the trend and the relative ordering rather than the level, which is what most competitive questions actually need. Its keyword and paid data is deeper than its traffic data, which is a reasonable reason to standardise on it if you are already using it for search.

3. Ahrefs

Ahrefs estimates organic search traffic by taking the keywords a site ranks for, applying modelled volumes and an assumed click curve. It is measuring a different quantity from the others, so it will always read lower than a total-traffic estimate, and the gap between the two is not an error. Used deliberately that is a feature: comparing an organic-only estimate against a total estimate gives you a rough sense of how dependent a competitor is on search.

4. The Chrome User Experience Report

Look up which popularity bucket the domain falls into and record it. It costs nothing, it is drawn from real Chrome navigations, and it constrains every other number on this page. Track the bucket quarterly rather than the exact rank, since the buckets are what the dataset is designed to express and they are stable enough to mean something when they change.

5. Cloudflare Radar domain rankings

Search the domain on Radar for its current rank and trend, or download the bucketed datasets when you want to place a whole competitive set at once. Because the underlying signal is DNS resolution rather than page navigation, it also picks up traffic that never renders a page in Chrome, which matters for competitors with heavy API or mobile usage.

6. Google Trends

Trends reports relative search interest in a term over time, from Google’s own query data. Comparing brand names gives you a demand signal that is independent of anybody’s traffic model and that frequently moves first, since people search a brand before they visit it. It is relative and unitless by design, which suits the ratio-based approach this page recommends.

7. Their own published claims

Companies quote visitor and user numbers in press releases, funding announcements, conference talks and, for listed companies, in filings where the figure has been through review. These are self-reported and usually chosen flatteringly, and they still bracket the estimate: a company is unlikely to publish a number far below reality, and a regulated filing is unlikely to publish one far above it.

8. Advertising media kits and rate cards

Any competitor who sells advertising, sponsorship or newsletter placements publishes an audience figure in order to price it. Media kits routinely state monthly uniques, page views, subscriber counts and audience demographics. It is a self-reported number with a commercial incentive behind it, and it is also the most detailed audience description that company will ever publish.

9. App store rankings and install estimates

For a competitor whose audience lives mostly in an app, website traffic is close to irrelevant and a web estimator will badly understate them. Category rank, review velocity and third-party install estimates are the equivalent signals. Check whether an app exists before concluding that a rival is smaller than you thought.

10. Your own referral logs and Search Console

Your own data covers the overlap between you and them, which is the part that affects your business. Referral logs show which sites send you visitors and which competitor domains appear in the path. Search Console shows the queries where you both surface, and impression share on those queries moves before any traffic estimate does. That overlap analysis is a natural companion to competitor keyword research, which works the same dataset from the other direction.

The total is the least informative number an estimator produces and the one everybody quotes. Where the traffic comes from tells you what kind of company you are dealing with and what it would take to compete with them.

One caution about the direct bucket. It is where analytics puts anything it cannot attribute, so it silently absorbs traffic from messaging apps, email clients, documents and private browsing. Reading a very large direct share as pure brand strength usually overstates the case. Where a competitor’s audience truly sits is better triangulated against their share of voice in the places their buyers actually gather.

1. 1Bound it with the measured rankings. Check the Chrome popularity bucket and the Cloudflare rank. An estimate implying a scale incompatible with both is wrong, and this check takes two minutes.
2. 2Check the size band. Accuracy collapses on small sites. Below a few thousand monthly users these tools are close to guessing, so for a small competitor report the rank bucket and skip the visit figure entirely.
3. 3Confirm what is being measured. An organic-only figure and a total-traffic figure are different quantities. Half the apparent disagreements between tools are just this, and it is easy to check.
4. 4Look for a self-published number. A media kit, an investor deck or a filing gives you a figure the company was willing to attach its name to. Where it is wildly outside the estimate, one of the two needs explaining before either goes in a report.
5. 5Sanity-check the traffic against the company. Traffic that would imply a business several times larger than their headcount and funding support is usually a modelling artefact, or a domain carrying traffic from something other than the product.

This is the least legally fraught data point in competitive research. Every source here is either published deliberately by its owner, offered as a public service, or a measurement of your own property. Three practical constraints are still worth knowing.

• Do not publish an estimate as though it were their real figure. Presenting a modelled number as a fact in comparative marketing is a misleading-advertising problem in most markets, not merely a research one. If you publish a comparison, say which tool produced the figure and on what date.
• Check the licence before republishing a vendor’s data. Traffic estimates are a commercial product and the platforms’ terms restrict redistribution. Quoting a figure with attribution is normally fine; republishing their charts or exporting their dataset into your own product usually is not.
• Analytics benchmarking reports aggregates, never a named competitor. Some analytics platforms offer industry benchmarking, and it is genuinely useful. It compares you against an anonymised, aggregated peer group by design, and any product claiming to show you a specific rival’s analytics is either estimating or should not have the data.

• Their real visit count. It exists only in their analytics. Proxy: a ratio against your own traffic in one tool, bounded by the two measured popularity rankings.
• Their conversion rate. Traffic without conversion tells you nothing about pipeline, and a competitor with a third of your traffic can be winning more deals. Proxy: how often they appear in your own contested deals, which is the only conversion signal that concerns you.
• Traffic to a specific page. Page-level estimates are far weaker than domain-level ones and should not be relied on. Proxy: the ranking position and keyword volume for that page, which at least tells you whether it can be receiving meaningful traffic at all.
• Traffic for a small or newly launched site. Below the panel threshold there is nothing to extrapolate from and the tools guess. Proxy: absence from the Chrome popularity dataset, plus indexed page count and branded search volume in Trends.
• Who their visitors are as a company or role. Audience demographics from estimators are broad and modelled. Proxy: their published media kit if they sell advertising, and the accounts you can identify by name, which is a separate exercise covered under a competitor’s customers.

Quarterly is the right cadence, and the discipline that matters more than the frequency is consistency: same tool, same market, same period length, same day of the quarter. Switching tools mid-series manufactures a step change that never happened, and it is the most common way a traffic series ends up misleading the people reading it.

Record the estimate, the tool, the date and the two ranking buckets together, so the series carries its own method rather than a bare number. A competitor tracking spreadsheet is sufficient structure for a handful of rivals. What deserves faster attention than the traffic series itself is what moves it: a campaign launch, a content programme, a partnership or a funding round. Flares watches those continuously, which matters because traffic is a lagging indicator and the events that produced this quarter’s line happened last quarter.