A test of GigaChat across sport, finance, B2B and marketing queries found source links under 73% of answers in the chat interface and under 4% of answers returned through the GigaChat API. The sample ran to about 430 responses, split roughly evenly between the two channels.
Two channels, two behaviours
The chat interface assembles a sources block beneath the answer itself. The API returns the answer as text, so the citations do not travel with it. That does not mean the model works without external data — it means an API response gives no view of where the information came from. Anyone checking a site's visibility inside GigaChat should look at the chat, not the API.
Even in the chat the room is tight: an answer carries two or three links on average. Repeating the same question changes part of that set, so placement is not fixed and the same query can surface different sources on different runs. A link that appears once in five attempts is a coincidence, not a presence.
What it means for affiliate work
Generative search sends traffic only to whoever the model picks as a source, and in GigaChat that is a slot of two or three lines. The competition is hard, and the instability makes it harder to measure: visibility has to be tracked over a series of repeats rather than a single snapshot. For media buyers the question is attribution as well — a click out of a GigaChat answer is difficult to separate from ordinary search without tags in place.
The practical test is a list of queries in your own vertical, each put to the chat several times, recording whether your site is cited, where, and whether the source set moves. Against the API the same queries will mostly return no links at all, which is expected.
