Why traffic estimates disagree between tools, and what to do

Traffic estimates disagree because tools do not observe the same thing. Some sample real browsing, some infer clicks from search rankings, and some only rank sites without counting visits. None of them can see a site's server logs, so each one is an educated guess built on different evidence. Once you understand the inputs, the disagreement becomes predictable rather than confusing.

There is no ground truth outside the site

A site's own analytics or server logs are the only place real traffic exists. Everything else is a model. When you compare outside tools, you are comparing different models of an unobserved quantity.

That is why two respectable tools can differ by a large factor and both still be doing their jobs. The question is not "which one is right?" It is "which one is wrong in the way that matters least for my decision?"

Reason 1: Different data sources

The biggest cause of disagreement is that the tools use different raw material.

• Panel and clickstream tools model total traffic from samples of real browsing, often collected through browser extensions or internet provider partnerships. They can cover all channels.
• SEO platforms such as Ahrefs and Semrush estimate organic search traffic from keyword rankings, search volumes and an assumed click-through rate for each position. Semrush also sells a separate product that models total traffic from clickstream data, so even one vendor can give two different numbers depending on the product you open.
• Free ranking lists such as Tranco, Cloudflare Radar and the Chrome UX Report observe DNS lookups, link counts or opted-in Chrome users. They provide position, not visits.

Compare organic search traffic from an SEO tool with total visits from a panel tool and the second number will nearly always be larger, because it includes direct, social, email and referral traffic. This is not an error. It is a difference in scope.

Reason 2: Sampling and scaling

A panel is only a sample. To turn a sample into an estimate of the whole internet, a vendor must scale it using assumptions about who is in the panel and who is not. If the panel under-represents some regions or device types, the scaled number inherits that bias.

Small sites suffer most. If only a handful of panel members visit a site in a month, the scaled number is noisy. A Screaming Frog test of 25 sites reported that low-traffic estimates were mostly at least 40 percent off. The author acknowledged the sample was small, so treat it as an illustration and not a benchmark.

Reason 3: Click-through assumptions

For keyword-based tools, the click-through curve is the heart of the model. They assume that a given rank position gets a given share of clicks. In practice, the real share depends on:

• Whether the query is a brand search or a generic one.
• Whether the results page shows a map, an answer box, videos or ads.
• Whether the searcher is looking to buy, learn or compare.

Because these factors vary query by query and a model uses averages, errors build up. The Screaming Frog author noted that estimators ignore search intent, brand versus non-brand queries and result page features, and that long-tail, location-specific traffic was consistently underestimated.

Reason 4: Different definitions

The word "traffic" hides many definitions.

Term	What it counts
Visit or session	One period of activity by a browser
Unique visitor	A distinct person or device, however identified
Page view	Each page loaded
Organic visit	A visit that arrived from a search result

If one tool reports unique visitors and another reports visits, a gap is expected. Page views can be several times higher than visits. Always check the label before you compare.

Reason 5: Time windows and freshness

Tools update on different schedules. Some average 30 days. Some use the latest week. Some publish monthly. If a site launched a viral campaign two weeks ago, a weekly source will show it and a 30-day average will only partially reflect it.

The Chrome UX Report, for example, is released on the second Tuesday of the following month, so it always describes a month that has already ended. Cloudflare's bucketed lists cover the last seven days. A comparison between these two is really a comparison between different periods.

Reason 6: Treatment of bots and non-web traffic

Bots and crawlers inflate some measures. DNS-based lists can also count lookups from apps, APIs and background services. Good providers filter these out but with different rules, so the filtered numbers diverge. Cloudflare says it filters automated traffic from its ranking, but no filter is perfect.

Reason 7: Geography and language

Keyword databases are strongest in major languages and markets. A site popular in a smaller market may have fewer tracked keywords, so a keyword model sees less of it. Panels can also lean toward certain regions. The practical effect is that the same site can be well covered by one tool and poorly covered by another, simply because of where its audience lives.

How to interpret conflicting numbers

When two estimates disagree, work through these questions.

1. Are they measuring the same thing? Compare visits to visits and organic to organic.
2. Are they covering the same period? Align the months.
3. Is the site small? If yes, expect larger error and give less weight to both numbers.
4. Do the tools agree on rank order among competitors? Often they do, even when absolute numbers differ.
5. Does one number contradict observable facts? A tool that says a site with no team, no content and no mentions has millions of visits deserves suspicion.

Use disagreement as information

A gap between sources is itself a clue. If the keyword tool is low and the ranking lists are high, the site probably gets most of its traffic outside search, such as direct visits or an app. If the panel tool is high and every ranking list is low, suspect bot traffic or a skewed panel.

This is one reason StartupsIntel blends several independent lists in its traffic checker, instead of relying on one. The idea is that independent errors partly cancel. You can read the full approach on the methodology page, and compare many sites in one go with the bulk checker.

What to do in practice

• Pick one source for tracking. Use the same tool every month to measure change. Do not switch tools mid-series.
• Use a second source for sanity. Another tool, or public signals, confirms the scale.
• Present ranges. A band like "between 300,000 and 600,000" communicates honesty.
• Name the source. Say "per the Chrome-based list, April" and not "the site gets."
• Ask for the truth when it matters. In an acquisition, ask for analytics access, not an outside estimate.

A simple example

Imagine a developer tool that most people reach by typing its name, bookmarking it, or following a link inside a community chat. A keyword model sees only the small slice of visitors who arrived through a search result, so it reports a modest figure. A DNS list sees a large, loyal user base and ranks the domain high. A panel tool lands somewhere between, depending on how many panel members happen to use the tool.

All three tools did their job. They simply answered three different questions, and the apparent conflict disappears once you ask which question you care about. If you are evaluating search potential, the keyword number is relevant. If you are judging overall reach, the others are.

FAQ

Which tool is closest to the truth?

It depends on the site. Panel tools usually do better on large sites, keyword tools on search-driven sites. No tool is reliable for every site, so use more than one and look for agreement.

Why is Ahrefs or Semrush traffic lower than Similarweb for the same site?

SEO platforms mostly estimate organic search traffic, while panel tools try to cover all channels. A lower number from the SEO platform is therefore expected for any site with direct, social or referral traffic.

Do free ranking lists agree with each other?

Roughly, for large sites. They measure different things, so exact positions differ. Agreement across several lists is a good sign, and disagreement tells you something about how the site is used.

Can I trust a traffic number I find in a pitch deck?

Only if it is backed by the owner's analytics. Ask to see them. Without that, treat the number as a claim rather than a measurement.