Alexa Rank was the single most quoted figure about websites for a quarter of a century. It appeared in funding decks, on agency slides, in newspaper articles and in a very large number of peer reviewed papers that needed a proxy for site popularity. It was a position in a list, derived from a panel of people who had installed a browser toolbar, and it closed in 2022: which means a generation of citations now point at a service that answers nothing.
How Alexa Rank produced a traffic number
The mechanism was straightforward. Alexa observed browsing behaviour from its panel, computed for each site an estimate of unique visitors and of page views over a rolling three-month window, combined the two into a single score, and published the resulting ordering. The output was a traffic rank: position one for the most visited site observed, larger numbers for everything below it.
Two consequences of publishing a position rather than a quantity are worth drawing out. First, the figure was extremely legible: "we are ranked 40,000" needs no context to sound meaningful, which is exactly why it spread. Second, the intervals were wildly uneven. The difference in real traffic between rank 100 and rank 200 is enormous; between rank 900,000 and rank 1,000,000 it is close to noise. A traffic rank in the millions was essentially a coin toss, and it was quoted with the same confidence as a rank in the hundreds.
The toolbar, and why it mattered
The Alexa toolbar shipped from 1997 and was, for a period, genuinely widespread: bundled with other software, installed by webmasters wanting to see their own figures, and later reproduced by a long tail of extensions feeding the same panel. That population was the measurement, and it was not the internet.
The bias was consistently in one direction: towards technically engaged desktop users, towards webmasters and marketers, towards the parts of the world where the toolbar was distributed, and away from corporate networks where installation was blocked. Sites about search engine optimisation, web development and hosting therefore scored dramatically better than their real audiences justified, because their visitors were disproportionately likely to be carrying the instrument. This was well documented at the time and almost never mentioned in the citations.
What the rank got wrong
Structural limitations
| Limitation | Effect on the number |
|---|---|
| Self-selected panel | Systematic over-statement for technical audiences; under-statement for mainstream and enterprise ones. |
| Desktop origin | As browsing moved to mobile applications, a growing share of real traffic was structurally invisible. |
| Position, not quantity | Uninterpretable intervals. Movement of tens of thousands of places could mean nothing at all. |
| Subdomain handling | Large platforms and their subdomains were aggregated in ways that flattered or penalised inconsistently. |
| Gameable | A small panel means a small amount of coordinated activity could move a rank materially. An industry existed to do so. |
None of this made Alexa Rank useless. Used as a rough ordering among comparable sites it was informative, and the bias applied to all of them equally. Used as an absolute statement about a single site's audience it was misleading, and that is overwhelmingly how it was used.
Why Alexa Rank closed
Amazon, which had acquired Alexa Internet in 1999, announced in December 2021 that the service would be retired, and Alexa.com closed on 1 May 2022. No detailed rationale was published, so what follows is context rather than explanation.
The method had been under pressure for a decade. Mobile browsing moved a large share of real traffic into applications the panel could not see. Browser extension stores repeatedly purged products that collected browsing history, which thinned the supply. Privacy regulation made the collection and resale of that data substantially harder to justify. And in the professional market, paid competitors with larger and better-characterised datasets had taken the customers who would once have paid for Alexa's reports. When it closed, the free tier that everyone actually cited had long since become a marketing surface for a product few people bought.
Alexa Rank was not alone. Compete closed in 2016, Yahoo Site Explorer in 2011, SiteAdvisor and Web of Trust in the years between. The dated report on closed measurement tools sets out the whole sequence, and the pattern is consistent: free, widely cited, panel-derived, and gone.
What the replacements measure instead
Several commercial vendors publish traffic estimates of comparable ambition, built from clickstream data and search modelling rather than a toolbar panel. Their methods are better documented and their free tiers are small. For research use, the Tranco list, assembled from several underlying rankings specifically to be stable and reproducible, has become the standard replacement in academic work, precisely because it was designed to be citable rather than marketable.
What none of the replacements reproduced is ubiquity, and ubiquity was the actual function Alexa Rank performed. A single flawed number that everybody quoted at least made comparisons commensurable: two figures from the same broken instrument can be compared, whereas two figures from two different vendors cannot. There is now no shared denominator, which is a real loss even though the measurement it replaced was poor.
Reading an old citation
A source that cites an Alexa Rank is not thereby wrong; it is unverifiable, which is a different and more useful thing to say. The figure cannot be re-checked, so it should be read as a claim about relative popularity at a moment in time, from a panel with a known technical skew, with an interval that widens sharply the larger the number is.
The durable alternative is to cite something that does not depend on a vendor staying in business. A whois record and the DNS records for a domain will still answer in ten years, because the registry and the zone are obliged to publish them. That asymmetry, between facts a system must publish and figures a company chooses to sell, is the most durable thing to take from the closure of Alexa Rank.
