Checking a website's traffic means two different things, and they come from two worlds that never touch. On a site you own, the numbers are measured: a tracking code counts visits one by one. On a site that is not yours, the numbers are estimated: an algorithm reads the rankings, multiplies them by the search volume of the keywords and by an assumed click-through rate, and returns a figure. The first is a measurement, the second a statistical model.
Both arrive with the same face, the same chart and the same unit of measurement. That is where much of what companies get wrong when comparing their traffic with a competitor's comes from. Below you will find the tools for both questions, how wrong the estimates are according to those who have measured them, and Search Console data from two sites we manage that shows where the model breaks.
On your own site: the tools that count visits one by one
Nothing needs estimating here. The site is yours, you can put whatever code you like on it and read the data at source. The only things that change are who hosts that data and what it costs you.
| Tool | What it measures | Where the data lives | Cost |
|---|---|---|---|
| Google Analytics 4 | Sessions, users, events, conversions, traffic sources | Google servers | Free up to 10 million events a month |
| Google Search Console | Clicks, impressions, average position and queries from Google Search | Google servers | Free |
| Matomo | The same metrics as GA4, with the option of installing it on your own server | Cloud or your own server | Open source, paid cloud |
| Plausible | Visits, sources, pages, without cookies | EU servers | Paid |
| Microsoft Clarity | Session recordings and heatmaps, not traffic counts | Microsoft servers | Free |
| Server logs or Cloudflare Web Analytics | Actual requests to the server, bots included | Hosting or Cloudflare | Included with hosting |
In Europe this choice also has a regulatory history. In 2022 several EU data protection authorities, Italy's Garante among them with a decision announced on 23 June 2022, found the version of Google Analytics then in use unlawful because it transferred IP addresses and other identifiers to the United States without adequate safeguards. That is why many European sites moved their analytics to Matomo or Plausible. In the UK the question has been settled differently: since 12 October 2023 UK businesses can transfer personal data to US organisations certified under the UK Extension to the Data Privacy Framework. Analytics cookies still fall under PECR and the ICO's cookie guidance, so the "where the data lives" column remains worth reading, just for different reasons.
Search Console and Analytics never match, and it is not your fault
The question always comes up. Google addresses it in the Search Central documentation: the closest metrics are Search Console clicks and Analytics sessions, but they are calculated differently and will never line up. Google lists eight causes of discrepancy, including the time zone Search Console uses, cookie consent refused by the user, the fact that Search Console only reports canonical URLs and that Analytics filters bots while Search Console does not. Small differences are expected; large gaps should be investigated.
The practical difference is another one. Search Console tells you what happens before the click, Analytics what happens after. If traffic drops, the first place to look is the ratio between impressions and clicks: if impressions hold and clicks fall, the problem is in the SERP and not on the site. We have a dedicated article on website traffic drops.
On someone else's site: what you see is a model, not a counter
No external tool can read the analytics of a domain it does not control. When Similarweb, Semrush, Ahrefs, SE Ranking, SpyFu, Ubersuggest, Sitechecker or Serpstat show you "48,000 visits a month", that number is calculated. The vendors say so themselves.
SpyFu describes its method as calculating all the search results a site appears in multiplied by the rate at which people click those results. SE Ranking, in the FAQ for its traffic checker, explains that it starts from positions, search volumes and CTR adjusted for SERP features, and admits that third-party data cannot match the accuracy of Search Console or GA4. Similarweb works differently, with a network of panels, partners and connectivity providers described on its methodology page, but it is still a sample scaled up to the whole population with machine learning models.
A detail that says a lot about the state of guides on this subject: in Italy, page one still includes articles recommending Alexa and Quantcast. Amazon announced in December 2021 that it was closing Alexa.com, which happened on 1 May 2022. Four years on, that advice is still there.
How wrong the estimates are, measured
The most serious study on this point is by SparkToro. Rand Fishkin compared the estimates of four vendors with the real Google Analytics data of 641 sites, for a total of 7,692 monthly comparisons over the period June 2020 to June 2021. The correlation with the real figure, on a scale from 0 to 1, is as follows.
| Estimate provider | Correlation with real Google Analytics users |
|---|---|
| Semrush | 0.790 |
| Datos | 0.720 |
| Similarweb | 0.659 |
| Ahrefs | 0.504 |
On Similarweb, the vendor that does best on large sites in the study, Fishkin writes that "almost 2/3rds of the time (~63%) they get traffic numbers right to within +/-30%". Put plainly: two times out of three the number is right to within a third. One time in three, it is not. And on sites under 5,000 monthly visits, Similarweb's estimates were the worst of the group.
A second study, published by Collaborator on 31 March 2026 across more than 7,500 sites in Europe, the United States and Asia, measures correlations of 0.67 between Ahrefs and Search Console, 0.59 between Ahrefs and Analytics, and 0.42 between Search Console and Analytics. The surprising result is the last one: Google's two tools, both installed on the same site, are the least aligned pair of the three.
The part almost nobody quotes, though, is what Similarweb says about itself. On its Similarweb vs. Direct Measurement page the company states that even when comparing the same site with two direct measurement tools the variance can reach 30%, and that it has more confidence in its estimates for sites above 100,000 monthly visits. Turned around, that sentence says confidence is lower below 100,000 monthly visits. That is exactly the band where almost all small and mid-sized company websites sit.
Where the model breaks: the click-through rate curve
All ranking-based estimates rest on the same pillar: a curve that assigns an average click-through rate to each ranking position. If that curve is wrong, everything built on it is wrong in proportion.
We took the Search Console data of two Italian properties we manage and calculated the real click-through rate by position band. The first is an outdoor equipment e-commerce site, the second a site in the dental sector. Client data that is not public stays anonymous; the numbers are from Search Console, July 2026.
| Average query position | Impressions | Clicks | Real CTR |
|---|---|---|---|
| 1 | 1,526 | 423 | 27.7% |
| 2 | 1,468 | 74 | 5.0% |
| 3 | 3,037 | 59 | 1.9% |
| 4 to 10 | 4,466 | 48 | 1.1% |
The jump between first and second position is more than fivefold, and it is the main reason to prioritise work by how much it actually moves clicks. From third position down the click-through rate is below 2%, an order of magnitude away from the theoretical curves found in tool vendors' marketing material, where second position is typically worth between 10% and 15%.
A comparison with the US market helps show the direction of travel. GrowthSRC Media analysed the Search Console data of more than thirty clients, across 200,000 tracked keywords and around 74,000 ranking in the top ten, comparing the same period in 2024 and 2025: first position went from a 28% CTR to 19%, second from 20.83% to 12.60%. These are US figures, collected in e-commerce, SaaS, B2B and EdTech, and should be read as a trend, not as values that apply to a specific SERP in another market. The full research is here. The point is that the curve has dropped everywhere, while the estimates you read today are still built on yesterday's curves.
On the same e-commerce site, the monthly totals show how little sense it makes to read an estimate as if it were a stable figure.
| Site | Month | Impressions | Clicks | CTR | Average position |
|---|---|---|---|---|---|
| Outdoor e-commerce | June 2026 | 46,799 | 1,023 | 2.19% | 6.9 |
| Outdoor e-commerce | July 2026 | 54,549 | 1,493 | 2.74% | 6.3 |
| Dental sector site | July 2026 | 705 | 9 | 1.28% | 25.4 |
Between June and July organic clicks rise by 46% while the average position improves by half a point. An estimation tool that only sees positions would have recorded a much smaller change. The second site is the opposite case and the most common one among small businesses: 705 impressions and 9 clicks in a month are numbers no statistical model can reconstruct from the outside, because below that threshold the noise outweighs the signal. If you are interested in how to read the relationship between position and clicks, we cover it in the articles on click-through rate and on the SERP.
Why the problem is bigger for smaller companies
The threshold Similarweb states, 100,000 monthly visits, leaves out almost the entire business population. According to the Department for Business and Trade's Business population estimates 2025, the UK had 5.7 million private sector businesses at the start of 2025, 99.85% of them SMEs, and 258,520 with between 10 and 249 employees. Most of those that have a website sit well below the level at which the estimates become reliable.
The pool of users, on the other hand, is not the issue. The UK also has an industry-standard measurement of online audiences, Ipsos iris, endorsed by UKOM, which combines a panel of users with data from the sites that take part: the same logic as Similarweb's panels, applied to a defined set of participating publishers, and of no help for a manufacturer's website that is not in it.
The traffic no estimate can see yet
Traffic estimation tools start from positions in the search results. Visits that come from a conversational assistant do not pass through there, so they are left out of the calculation by design. If someone lands on your site after reading an answer from ChatGPT or Perplexity, in GA4 you will find them in the acquisition report as a referral source, with the assistant's domain. In an external estimation tool you will not find them at all.
On the Google side something has moved: in June 2026 Search Console introduced performance reports for generative search, so the AI part of Google's results is starting to get its own line in first-party data. If you want to go deeper on that side, start with AI Overviews ranking factors and the piece on zero-click searches.
Five steps to estimate a competitor's traffic without being misled
- Use one tool and stick with it. Estimates from two different vendors cannot be compared with each other, because they start from different panels and click curves. A comparison only makes sense between domains measured by the same model in the same month.
- Read the number as an index, not a measurement. "The competitor is estimated at three times our traffic" is useful information. "The competitor gets 48,000 visits" is not.
- Look at the keywords, not the total. The list of queries a domain ranks for is much more solid data than the traffic figure derived from it, because the first can be checked by searching on Google and the second cannot. It is the starting point of any serious keyword research.
- Check the size. If the domain you are analysing is below 100,000 estimated monthly visits, you are in the band where the vendors themselves state lower confidence.
- Check the real traffic when you can get it. In an acquisition, a due diligence or a supplier relationship, you can ask for read access to Search Console and Analytics. Almost nobody does, and they carry on debating estimated numbers when they could have the real ones.
How these numbers turn into decisions is better covered in two articles: the one on SEO KPIs, which explains which metrics are worth putting in a report, and the one on ROI. If instead you want to see what a long project looks like on real data, in the SEO for manufacturing case we published Macropix's rankings before and after five years of work, including a keyword that went from position 88 to position 2. The rest of the projects are on the case studies page, and if the point is to go from reading the data to a plan, that is the work we do in our SEO service.
Frequently asked questions
No. No external tool reads the analytics of a domain it does not control. All the numbers you see for someone else's site are estimates calculated from rankings, search volumes and assumed click-through rates, or reconstructed from user panels. The only way to get the real figure is to obtain read access to the site's Google Analytics or Search Console.
In the SparkToro study of 641 sites and 7,692 monthly comparisons, the correlation between estimates and real Google Analytics data ranged from 0.790 for Semrush to 0.504 for Ahrefs. Similarweb, the most accurate vendor on large sites, got the number within 30% about two times out of three. The free versions of these tools use the same models with less historical data, so they are not more accurate: they just limit the number of lookups.
Because they measure different things. Search Console counts clicks on Google search results, Analytics counts sessions on the site. Google lists eight causes of discrepancy in its documentation: time zone, refused cookie consent, attribution models, canonical URLs, bot filtering present in Analytics and absent in Search Console, handling of non-HTML files such as PDFs, and the split between web, image and video search.
Similarweb states that it has greater confidence in its estimates for sites above 100,000 monthly visits and, in the SparkToro study, its estimates for sites under 5,000 monthly visits were the least accurate of the group. Almost all small and mid-sized company websites sit in that lower band, where measured traffic in a month may be a few hundred clicks and no statistical model can reconstruct it from the outside.
Ranking-based estimates do not, because they start from classic search results. Visits from a conversational assistant show up in GA4 as referral traffic with the assistant's domain. For the generative part of Google Search, since June 2026 Search Console has offered dedicated performance reports for generative search.
Something the industry rarely says: traffic estimation was born as a sales tool, not a measurement tool. Its job is to show a prospective client a big number next to a competitor's name, and it does that very well. When someone presents a tool's estimate for your domain as if it were a snapshot of your traffic, they are using a number that the vendor itself, in its own documentation, describes as less reliable in exactly the size band you are in. Your real figure is two clicks away, inside Search Console, and it costs nothing.
