Ten to one: AI assistants read my pages for people ten times more often than search sends me a reader
Lesson from:
30 days, three sites, one new counter: 1,030 pages fetched because someone asked their AI, against 104 visitors from every search engine combined
Earlier this month I counted who reads this site and found that machines read it more than people do. This time I wanted a smaller, sharper number: when someone actually needs something from my pages, how does it reach them?
So I built a little readership counter over my own web logs and let it run over the last 30 days, 18 October to 17 November 2026, across my three public sites: this site, LSN News and my personal blog.
The answer fits on a sticky note.
Contents · 6 sections
TL;DR #
- 1,030 times, an AI assistant fetched one of my pages because a person had just asked it something.
- 104 readers arrived from a search engine. All of them together: Google, DuckDuckGo, Bing, Brave, Yahoo and Yandex.
- That is roughly ten to one.
- Some of my most-used threat-intelligence dossiers are read almost only through AI. One had 75 assistant fetches and 3 human visitors.
The sticky note #
| Last 30 days | Count |
|---|---|
| Pages an AI assistant fetched for someone’s question | 1,030 |
| Readers who came from a search engine | 104 |
| Readers who came from Hacker News | 150 |
| Readers who came from LinkedIn (web and app) | 80 |
| AI crawler fetches (training and AI search) | 71,262 |
| Search engine crawler fetches | 9,197 |
The assistants, by name: ChatGPT 874 page fetches, Claude 121, DuckDuckGo’s assistant 26, Perplexity 7, Mistral 2.
On the search side, Google sent 60 readers in a month. DuckDuckGo sent 35, Bing 4.
What “an assistant fetch” means #
There are two very different kinds of AI traffic, and the difference is the point of this post.
- A crawler (GPTBot, ClaudeBot, PerplexityBot and friends) reads pages in bulk, to train a model or to build an index. Nobody is waiting for the answer.
- An assistant fetch (ChatGPT-User, Claude-User, DuckAssistBot, Perplexity-User) happens in the moment: a person types a question, the assistant decides my page holds the answer, and reads it right then.
The second kind is the closest thing to a reader that an AI has. Someone had a question; my page answered it. They just never saw my site.
One honest caveat: a fetch is not a person. One question can lead to several fetches, and one person can ask several questions. I count fetches, and I say so. The readers on the other side are counted the way my analytics always counted them: real browsers that opened a page, once per site, with bots, scrapers and cloud-hosted addresses left out. The two numbers are not the same unit. They do not have to be, to show which way the wind blows.
Read through AI, almost never visited #
The pages assistants fetch most are my threat-intelligence dossiers, the signed investigations built from my own honeypots and logs:
| Dossier | Assistant fetches | Human visitors |
|---|---|---|
| The Lithuanian Connection | 75 | 3 |
| The Marketplace Economy | 73 | 0 |
| The Baltic Bridge | 66 | 1 |
| The DMZHOST Trinity | 44 | 19 |
If I judged these pages by visitors, I would call them failures. Judged by use, they are among the most useful things I publish. People ask about IP brokers, bulletproof hosting and the networks behind attacks, and their assistant comes to read the evidence.
The lesson #
Ten years ago, “being findable” meant ranking on Google. On my sites today, Google sends about two readers a day. AI assistants read my pages for someone about 34 times a day, and AI crawlers read the whole shelf thousands of times a day.
So I do three things, and they are cheap:
- I let assistants in. My robots.txt welcomes crawlers, and an llms.txt tells AI systems who I am and how to cite the work.
- I write pages that answer one question well: a clear title, the claim, the evidence, the date. That is what an assistant can quote.
- I count the right thing. A page with three visitors and seventy-five assistant fetches is not dead. It is being read somewhere else.
If you publish anything worth knowing, look at your logs for ChatGPT-User, Claude-User and DuckAssistBot. Those are your invisible readers. Mine outnumber my search visitors ten to one.
How I counted #
Every request to my sites passes one reverse proxy and is logged. The readership counter reads those logs and sorts each address by the user agent of its first request: named AI agents, search engines, link previews, scripts, scrapers and known probes are separated out. A reader is a real browser that loaded at least one page, counted once per site. Assistant fetches count only real pages, not images, scripts or robots.txt. The window is 18 October to 17 November 2026.