Tracker
What changes at ChatGPT, dated.
A fixed set of 26 German questions runs against ChatGPT every morning — always the same wording, otherwise the comparison would be worthless. The tracker compares network and response object field by field with the day before. A change only counts once it holds for two measuring days.
Today's status changes · 4 of 12
Results, domains, citations, source footnotes, widget rate — per day and question category.
How one prompt becomes many queries: count, modes, freshness, language and domain filters.
Products per run, offers per product, carousel rate — plus the instant-checkout watch.
Reasoning seconds, processing times, and the parameters that currently hold, each dated.
All events →·What currently holds →·Metrics →·Measuring days →
Knowledge base
Every interface, every field — explained.
When the tracker reports a change, next to it stands what the object is: structure from the measurement data, effect on visibility, change history, dated external sources.
The search pipeline
9 topicsFrom your prompt to the answer: how web.run decides, fans out, filters and reads.
Interfaces & objects
13 topicsThe conversation object, the endpoints behind chatgpt.com, the SSE stream, the send payload — field by field.
Metrics, explained
22 pagesEvery number the tracker publishes has a page: how it is measured, what moves it, its history.
Source register
26 sourcesExternal findings, each dated and checked against our own measurements — because a finding can be outdated without ever having been wrong.