Five terms are under test here. None of them existed before this station published them, which means every page about them is one this station controls, and any AI that cites one had to find it by searching the live web.

Each term is published twice, identical except for a single variable. Every week the station asks Claude, GPT and Perplexity a plain question about the term, with web search switched on, and records which page came back — or that none did.

Most weeks the answer is that none did. Those are recorded at the same weight as the hits, because a result you only publish when it flatters you isn't a measurement.

Probes run Mondays at 09:00 UTC. Each provider is called with its own search tool enabled — without it a model answers from training data and can't cite anything, so every reading would be a false negative.

Three outcomes are kept separate. Cited means the page appeared in the answer or its sources. Not cited means the probe ran, an answer came back, and the page was absent. Error means the probe never ran, and is treated as missing data rather than as evidence of anything.

Nothing is backfilled and nothing is overwritten. The log only appends. Cells sit empty until a real reading fills them, which is why this page will look mostly blank for its first few weeks.

The marks above are a substitution alphabet, not an obfuscation layer. The page ships plain text in its markup — a font can only change what a person sees, never what a machine reads, and hiding the source from machines would end the measurement.