Findings · 2026-09-23 · By VSNARY | Emmanuel Orta · 0 views
Our homepage says 6,557 sites measured. Six people ran a scan this week.
Both numbers are true. One is the corpus the scanner built to have something to benchmark against; the other is human demand. A week of edge logs shows who actually visits a site about crawlers: mostly crawlers.
The 6,557 figure counts distinct domains with a measured record, most of them scanned by the corpus crawler to build the benchmark population. Human demand is a different number: in one week, eyeball requests to the scan endpoint from addresses that were not ours came to about six, from five countries. The rest of the site's traffic is crawlers, led by Amazonbot and Ahrefs on the pricing page.
The number on the homepage is counted to completion. It is the distinct domains that carry a measured record, and the endpoint that publishes it declares whether the listing it was read from was truncated (it was not). What the number is not is people. Most of those domains were measured by the corpus crawler, a scheduled job that works through a queue of about eleven thousand hosts so that every finding rate on the dataset page has a population under it. A site that says a defect fires on 21% of scans has to have scanned enough sites for the percentage to mean something. That is what the 6,557 is for.
Live, as you read this: the corpus now holds 6,558 domains across 2,310 scans. The figures in this piece were measured on the date above; this line is not.
What a week of edge logs shows #
Cloudflare records every request that reaches the edge, with the path, the user-agent, the country and whether the request was a visit. We read seven days of it, eyeball requests only, and the site that measures crawlers turns out to be measured by them.
| Path | Requests, 7 days | Who |
|---|---|---|
| /pricing | 9,485 | Amazonbot 1,819 · AhrefsSiteAudit 1,627 · ClaudeBot 1,099 · SemrushBot 1,081 · GPTBot 1,066 · PerplexityBot 784 |
| /unlock | 3,282 | the export gate redirect, 49 visits |
| / (homepage) | 2,608 | United States 882 visits, China 407, Singapore, Netherlands, France |
| /demo | 425 | the demo report |
| each blog post | 45–60 | nearly identical counts on every post |
The pricing page is the busiest page on the site by a factor of three and almost no human has seen it. On the two heaviest days it took 2,055 and 3,021 requests; the homepage never passed 700. Every report and every registry page links to it with the domain in the query string, so the crawlers that walk those pages follow thousands of variants of one page. The variants carry a canonical to the bare URL and are served from the edge cache, so the cost is close to zero and the finding is not a defect. It is just not demand.
The blog is the same story in a quieter register. The catalogue of crawlers the site publishes shows the same names in the same order as the log does, which is a small comfort: the site is at least being read by the things it writes about. Forty-five to sixty requests per post, on every post, regardless of age or subject, is the signature of sitemap walkers, not readers. A post people read has a shape: a spike, a tail, and a number that differs from its neighbours.
The scan endpoint #
The one path that measures demand is POST /api/scan, and it answered 557 successful requests in the week. Of those, 468 came from one browser user-agent at one address class that is ours, and 45 came from curl in our own analysis container. The globe on the homepage already drops our own addresses, and it puts the remainder at six scans from five places: Milan, Zürich, Charleston, a city in the Philippines and one in India.
| Scans, 7 days | Count |
|---|---|
| Successful scan requests at the edge | 557 |
| From our own browsers and container | about 545 |
| From anyone else | about 6, five origins |
| Paid | 0 |
That is the honest denominator for anything this site says about itself. It is also, as it happens, the same measurement discipline the scanner applies to everyone else: claimed hits are not visits, and a count that mixes the operator's own traffic with the public's is not a count of the public.
What the humans do #
The people who arrive mostly read. After the homepage, the pages with visit counts that look human are the demo report, the video page, the policy page and the extension's privacy page, the last of those being Chrome Web Store reviewers doing their job. The directory, the crawler catalogue, the glossary and the how-it-works page each draw seventy to ninety visits. The scan box gets used by a handful of people a week. That is the funnel, and it is very short.
Why publish this #
Because the scanner's whole proposition is that a claim should be checkable, and "6,557 sites measured" under a hero headline reads as demand to anyone who does not know what a corpus is. The number stays, because it is true and because the benchmark population is the reason the rates on this site mean anything. It gains a neighbour: scans by visitors in the last thirty days, read from the same log that excludes us. When that number is small it will say so.
This is also the third time this site has caught itself publishing a figure that was correct and misleading at once. The first was a cohort percentage that only means something with a second day under it; the second was a scan counter and a domain counter that had been published under the same label from the same Worker on the same afternoon. The fix each time was the same: name the population, or do not print the number.
Findings behind this #
The per-path figures above are one week of Cloudflare's edge analytics for this zone, eyeball requests only. The scan-origin figures are from the same data that draws the globe, which excludes our own addresses and any browser that has asked to be muted. The corpus figure is live on this page and on the dataset page.
Every figure above came out of this scanner.
Point it at your own domain and see the same measurements, free.
The main product
Found this on your own site? We fix it for $749.
Scan free to see where you stand. The fix is one site, every finding implemented and re-measured, with a sealed before and after.
Questions this post answers
Is the 6,557 figure wrong?
No. It is the count of distinct domains that carry a measured record, counted to completion. Most of them were measured by the corpus crawler that builds the benchmark population, not by visitors.
How many people actually use the scanner?
In the week measured, about six scans came from addresses that were not ours, from five places. The rest of the successful scan requests were the operator's own browsers and analysis container.
Why does the pricing page get so much traffic?
Every report and registry page links to it with the domain in the query string, and crawlers such as Amazonbot, Ahrefs, ClaudeBot, SemrushBot, GPTBot and PerplexityBot follow thousands of variants. The variants carry a canonical and are served from the edge cache, so the cost is close to zero.
How do you tell a bot from a reader on a blog?
Shape. A post people read has a spike and a tail and a count that differs from its neighbours. Forty-five to sixty requests on every post regardless of age is a sitemap walker.
What changes on the homepage?
The sites-measured figure stays and gains a neighbour: scans by visitors in the last thirty days, read from the log that excludes the operator's own traffic.
Related findings
Comments
Comments are read before they appear. Nothing is published automatically, and no account is needed.
Writing about this? Facts, live figures and marks — every number on that page is dated and traceable to a scan.