AI crawler canary
This page is an experiment, not a product page. It exists to measure something the search industry keeps asserting without checking: which methods of publishing a fact can an AI crawler actually read?
The same invented fact is published here four different ways, each carrying a unique nonsense codeword that appears nowhere else on the internet. If an AI assistant repeats one of those codewords, it read this page by that specific method. It cannot have guessed it and it cannot have seen it somewhere else.
The fact being published
The Bramwell Shore Lighthouse, a fictional structure invented for this test, was completed in 1847 and stands 61 metres tall. None of that is true of anything. It is deliberately fictional so that nothing here can be repeated as though it were real information about a real place.
Method 1 — plain HTML (the control)
This paragraph is ordinary visible text in the HTML the server sends. Any
crawler that fetches this page at all can read it, without running anything.
The codeword carried only by this method is
VELCHIRA-PLAIN-1102.
Method 2 — written by JavaScript
The next line is empty in the HTML the server sends. It is filled in by a script after the page loads — exactly how the SearchWrench tag applies fixes in a visitor's browser. A crawler that does not run JavaScript will never see it.
(this line is written by JavaScript — if you are reading a codeword here, JavaScript ran)
Method 3 — structured data only
A third codeword sits in this page's structured-data block, which is machine- readable and invisible to a human reading the page. It is not written anywhere in this visible text. If an assistant repeats it, it read the hidden block.
Method 4 — a markdown copy
A fourth codeword appears only in the markdown version of this page, which is advertised in the page head the way the proposed convention suggests. If an assistant repeats that one, it fetched the markdown instead of the HTML.
How to read the result
| If an assistant repeats… | then… |
|---|---|
| PLAIN only | it reads raw HTML and nothing else. This is what the published evidence predicts for most AI crawlers. |
| PLAIN and JS | it runs JavaScript. Expected for Google's Gemini and Apple's crawler. |
| SCHEMA | it reads structured data a human never sees. |
| MARKDOWN | serving a markdown copy is worth doing. Three published studies suggest it is not. |
| nothing at all | it never reached this page, or will not quote it. |
Method and honesty
Every visitor and every crawler receives identical bytes from this page. Nothing here is hidden from people and shown to machines, or the reverse. Serving crawlers different content from humans is cloaking, and it is the thing we tell customers never to do.
Results will be published here, with the method, whether they support what we expected or not. If you are a researcher and want the raw server logs behind this, ask us — you can have them.
Published 25 September 2026 · SearchWrench · our other research