LoudPixelBot — who is reading your site, and why
If this name is in your server log, a person asked us to check that site, or an AI assistant cited one of its pages in an answer and we read the page to understand why it was chosen. This page says exactly what we fetch, what we never do, and how to stop us — including exactly what a robots.txt rule does and does not stop.
Last changed 2026-09-13contact@loudpixel.ai
The name we send
Every request from our servers carries this User-Agent, unchanged, in every part of the system. There is no second name and no browser disguise. It carries this page and an address, because a page is what a person reads and an address is what the crawler policies of the large reference sites ask for.
LoudPixelBot/1.0 (+https://loudpixel.ai/bots; contact@loudpixel.ai)Operated by LoudPixel (7288603 Canada Inc.), Montreal. One person answers the address.
When we visit
Three reasons, and nothing else.
Somebody asked us to check that site
The free report, the paid audit and the editor connection all start with a person typing a domain. We read that site the way an AI assistant would, so the report can say what an assistant sees. If the domain is yours and you did not ask, somebody else did — a prospective client, an agency, a competitor running the free report. The report is theirs to read; the site is still yours.
An AI assistant cited one of your pages
When we measure which sources the assistants name for a market's buying questions, we read the pages they cited to understand why those were chosen. That is a single fetch of a page an assistant already quoted publicly, not a crawl of your site.
You publish a specification we keep a dated record of
A small number of published standards pages — payment and agent-commerce specifications — are re-read weekly so our protocol watch can say what they said and when. At most a dozen requests a week across all of them, comparing a fingerprint, copying nothing.
What one visit looks like
Bounded, and never a crawl. A check of a site reads the page that was named, the site's robots.txt, sitemap and the machine-readable files a site publishes for assistants, plus a handful of probes — typically under twenty requests in total, spread over a few seconds. We do not follow every link, we do not fetch images or scripts, and we do not come back on a schedule unless a customer is paying for a monthly re-check of their own site.
We do not run your JavaScript. That is deliberate: the assistants we measure against do not run it either, so a page whose content only exists after scripts execute is reported as an assistant would experience it. Where a rendered copy is needed for a different purpose, we obtain it through a third-party page-rendering service, and that request arrives under that service's own name rather than ours.
The one exception, disclosed
When somebody asks whether AI crawlers can reach their site, we have to find out the only way there is: by asking as those crawlers. So a check includes four requests sent in the names of GPTBot, ChatGPT-User, ClaudeBot and PerplexityBot, one each, against a small number of pages. Every one of those strings ends with our own name and this page's address, so the same log line that shows a GPTBot request shows who really sent it. We never send those four for any other reason, and we never send them without the tail.
How to stop us
- A robots.txt rule addressed to us. A group for `User-agent: LoudPixelBot` is obeyed by every request we send — the page, the machine-readable files and the four probe requests alike. A `Disallow: /` under that name means you will not see us again, and we read robots.txt before we read anything else.
- A rule on the name. Block the string `LoudPixelBot` in your firewall, CDN or server configuration and every request we send is refused, including the four probe requests, which carry the name in their tail.
- An email. Write to the address in the string and the domain is excluded by hand. Say which domain; we do not need to know why.
What a wildcard rule does and does not stop
A wildcard rule (`User-agent: *`) is obeyed for every fetch nobody asked for — a page an assistant cited, a competitor's blog, a classification. It does not stop a check a person requested on your domain: that check exists to report what an assistant would see, and a site that shuts every bot out is a finding the report states rather than a reason to report nothing. If you want no check at all, use the group addressed to our name, and it holds.
How to tell whether it was really us
You cannot, from the name alone, and neither can we — a User-Agent is whatever the sender typed. We do not publish a list of addresses, because our requests do not come from a fixed set of them. What we are adopting instead is a signature: a request that is genuinely ours can carry a Signature-Agent header naming a public key directory on this domain, at the address below, so a firewall that verifies signatures can confirm the sender without trusting the name.
Key directory: https://loudpixel.ai/.well-known/http-message-signatures-directory
The rule to apply is the one we apply to every crawler on our own site: a request that claims this name and carries no valid signature deserves exactly the trust you would give any unverified bot. Treating it as us because it said so is the mistake we measure other sites for. How we count readers on our own site, and why we never trust a name: what our own logs showed.
What we do with what we read
A check produces a report for the person who asked, and that report quotes the lines it grades — a heading, a price, a policy sentence — as its evidence. We do not copy pages, we do not build a search index, and we do not train a model on what we fetch. Pages that assistants cited are read to understand the choice and counted; what is published from that is a count and a page type, never the page.
The method behind a check, and where it stops being reliable, is on the methodology page. What an AI agent needs to see on a site, step by step, is on the developer reference.
Questions site owners ask
I never asked for an audit. Why are you on my site?
Either somebody else asked for one of your domain — the free report is open to anyone, and people run it on competitors — or an AI assistant cited one of your pages and we read that page once. In both cases it is a bounded set of requests, not a crawl, and the name in the string points here so you can find out.
Will blocking you affect my search rankings or my visibility in AI assistants?
No. We are not a search engine and we are not any assistant's crawler. Blocking this name affects one thing only: whether a person who asked for a check of your site gets a report that could read it.
How often will you come back?
Once, for a check somebody asked for. A monthly re-check happens only when a customer is paying for one on their own site. A cited page is read once per measurement.
Do you respect robots.txt?
Yes. A group addressed to LoudPixelBot is obeyed by every request we send, and we read robots.txt before anything else. A wildcard group is obeyed for every fetch nobody asked for; a check a person requested on your domain proceeds past it and reports what it found. Rules are applied in the order you wrote them, first match first — so put your Allow lines before your Disallow lines if you want them to win.
Something calling itself LoudPixelBot is hammering my site. Is that you?
Almost certainly not. We send a bounded number of requests per check and never crawl. A name is free to type, and a crawler wearing ours is exactly the kind of thing we help sites detect. Block it as you would any unverified bot, and if you want us to look at the log, the address is in the string.