How to stop Gemini from accessing internet

Gemini API
,
1

Hi,
I was trying out Batch API with gemini 2.5 pro. I ran a batch with 400 rows out of which 275 of them had citations included.

 'citationMetadata': {'citations': [{'endIndex': 724,
    'startIndex': 514,
    'uri': '********'},
   {'endIndex': 1264,
    'startIndex': 978,
    'uri': '*********},
   {'endIndex': 2609,
    'startIndex': 2379,
    'uri': '*********'},
   {'endIndex': 9326,
    'startIndex': 9048,
    'uri': '*********'}]},

I haven’t provided it with any tools neither explicitly set any parameter for grounding or search. How can I disable this? I don’t want it to be able to access any external knowledge source as I want to only use the model’s reasoning capabilities to solve the problem mentioned in the prompt. The URI’s included were not irrelevant but related to the information given in the prompt.

2

These aren’t search results. It’s citation results from the data Gemini was trained on. They appear automatically if “the model directly quotes, at length, from another source. Can include traditional websites and code repositories.” (see API docs)

1 Like