Who actually visits a new website? I asked Cloudflare

aaronorelup.com went up on 2 August. In the 77 days since, I haven’t promoted it, installed Google Analytics, or looked at a single traffic number. Cloudflare, which hosts it, has been counting the whole time. This week I connected Claude to my Cloudflare account and asked the question anyone with a new site wants answered: who actually visits?

In the last 30 days the site served 11,637 pages. About 9,000 went to machines, about 1,600 went to me, and between 290 and 980 went to other people: 10 to 33 a day.

If you’ve never looked at web analytics, this is also a starting point: the few words you need, what a brand-new site really gets, and what I’m doing about mine.

The numbers stop on 17 October. A version refreshed every day is on the stats page.

Six words you need

• Request: anything a browser or program asks the site for. One page is often dozens of requests: the page, then its pictures, video and scripts.
• Page view: a request for a page itself. The closest thing to “someone opened a post”.
• Unique visitor: a distinct internet address in one day. Not a person. A bot is one, my laptop is one, my phone is another.
• Bot: a program fetching pages. Search engines, AI companies, SEO tools, and scanners looking for weak spots.
• Analytics script: a tiny piece of code that runs in a real browser and reports the visit. Most bots don’t run it and ad blockers stop it, so it undercounts, but it’s the best sign a visit was a person.
• Referrer: the page a visitor clicked from, when the browser passes it on. It’s how you’d know whether someone came from Google, X or an email.

Claude got the numbers through Cloudflare’s analytics connection (an MCP server: a plug-in that lets an AI agent call a service). I asked in plain English, and Claude wrote the queries.

What the dashboard says

Read the way a dashboard invites you to, this is a site taking off: about 145 page views a day in August and September, over 500 in October, and “unique visitors” up from about 115 a day to 412.

But nothing on that chart is a count of people. Every line counts anything that asked: bots, scanners and me. Nearly half of October’s jump is me alone. From 26 September the Claude app’s built-in browser started checking each post while my agents wrote it, and requests from my own connection went from about 14 a day to 615.

Who’s actually there

To get to people, Claude took one group away at a time:

For every page a person read, machines took at least nine.

The surprise is who the machines are. AI crawlers, AI assistants and AI search made 6,671 requests in those 30 days, twice the 3,313 from search engines. One of them, ChatGPT-User, fetches a page when someone’s question in ChatGPT needs it; it came 558 times. On a site this young, more people may be meeting my posts inside an AI’s answer than in a search result, and no dashboard calls them visitors.

What people read

The homepage, by a long way, then one post: Jefrie in 3D.

The analytics script gives a cleaner view of where people arrive from outside the site:

Both lists agree. Most people come in the front door and look around, and one post pulls readers in on its own.

Where they are: 470 of the 570 visits came from the US, then about 20 each from Japan, China and South Korea. The raw request count shows far more Germany, the Netherlands and Singapore, which is where data centres are, so that’s scrapers, not readers.

Where they came from: mostly unknowable. 550 of 570 visits arrived with no referrer, which is what a click from X, Discord, email and most apps looks like: the app doesn’t pass it on. Google sent about 20. So I can’t yet tell which post on X brought anyone in.

You, in this data

Your visit to this page is in the same tables, and Cloudflare will show you your own row:

If you start a site like this

What I’d tell anyone starting out, from these 77 days:

• Most of your traffic will be machines. Here, about nine pages in ten.
• AI companies will find you before people do. Within weeks, and twice as often as search engines.
• Scanners will try doors you don’t have. A quarter of all requests here were “not found” errors, mostly programs looking for WordPress logins and leaked password files. On a simple static site like this one, that’s harmless background noise.
• You will be one of your biggest visitors. Me and my agents made a quarter of all requests. Take yourself out before you read any number.
• The visitor count will flatter you. “412 unique visitors a day” was 10 to 33 people.
• You won’t know where people came from unless you mark your own links (below).
• Early numbers are small, and that’s normal. This is 55 posts with no promotion at all.

What I’m doing about it

Already done, with this post:

1. A favicon.ico. Browsers ask for it by name, and the site only had an SVG icon: 512 “not found” errors in 30 days.
2. A sitemap, and a robots.txt that points to it, so crawlers stop guessing at pages.
3. Logging switched on for the site’s Worker (the small program that serves its video), so errors get kept.
4. A reader counter of my own. It counts a browser that stays on this post for a few seconds, once a day, and keeps only the post, the day and the country:

Next:

1. Tag every link I share. ?ref=x, ?ref=bluesky, ?ref=devto on the end of each link, and teach the counter to keep the tag. Then “which post brought people in?” has an answer.
2. Take myself out first, every time. Any number I read starts with my own connection removed.
3. Promote something, and watch. This is the baseline: 10 to 33 people a day. The next post I share, tagged, shows whether that number moves.
4. Decide whether I need time on page. Cloudflare counts page loads, not reading. Knowing whether people scroll or leave needs a small script of my own, or a privacy-friendly tool like Plausible or Umami. Not yet.

httpRequests1dGroups(limit: 100, filter: {date_geq: "2026-08-01"}, orderBy: [date_ASC]) {
  dimensions { date }
  sum { requests pageViews bytes }
  uniq { uniques }
}

httpRequestsAdaptiveGroups(limit: 20, filter: {datetime_geq: "2026-09-18T00:00:00Z"}, orderBy: [count_DESC]) {
  count
  dimensions { verifiedBotCategory }
}

httpRequestsAdaptiveGroups(limit: 12, filter: {…, clientIP: "<my address, from /cdn-cgi/trace>"}, orderBy: [count_DESC]) {
  count
  dimensions { userAgentBrowser }
}

httpRequestsAdaptiveGroups(limit: 50, filter: {…, edgeResponseStatus: 404}, orderBy: [count_DESC]) {
  count
  dimensions { clientRequestPath }
}

httpRequestsAdaptiveGroups(… dimensions { clientRefererHost })
rumPageloadEventsAdaptiveGroups(… dimensions { refererHost })

The dashboard said 412 visitors a day. Every useful number in this post came from asking “and who was that?” one more time, which is tedious by hand and takes an agent a minute. I built the site for people, and most of what reads it is machines and me. The people are a small number, and now one I can watch.