Description:
Search grounding recently stopped working correctly. This was not the case in previous days/weeks — the tool was returning proper web results before. Tested on both Google AI Studio and the Gemini API (google-genai SDK), with models gemini-3.1-pro-preview and gemini-3-flash-preview.
Observed behavior
Queries containing a place name return local time for that location instead of search results
All other queries return empty results
The model itself diagnoses the failure clearly (reasoning trace):
"the search function isn't working as expected. It seems broken and incapable of returning anything other than time information when searching anything other than locations"
"queries with 'London' are handled as expected, returning London time. Without a location, the search yields an empty result"
Impact
The model falls back to training data recall, defeating the purpose of grounding:
"I'm pivoting to internal knowledge, as the search isn't yielding results"
Description:
Search grounding recently stopped working correctly. This was not the case in previous days/weeks — the tool was returning proper web results before. Tested on both Google AI Studio and the Gemini API (google-genai SDK), with models gemini-3.1-pro-preview and gemini-3-flash-preview.
Observed behavior
Queries containing a place name return local time for that location instead of search results
All other queries return empty results
The model itself diagnoses the failure clearly (reasoning trace):
"the search function isn't working as expected. It seems broken and incapable of returning anything other than time information when searching anything other than locations"
"queries with 'London' are handled as expected, returning London time. Without a location, the search yields an empty result"
Impact
The model falls back to training data recall, defeating the purpose of grounding:
"I'm pivoting to internal knowledge, as the search isn't yielding results"