For Gemini grounding with Google Search, use the API contract that matches the current SDK and inspect the returned citation evidence. The current documentation uses the Interactions API with a google_search tool. Its steps and citation annotations should not be mixed with legacy GenerateContent grounding_metadata fields.
Use the current request boundary
import os
from google import genai
client = genai.Client(api_key=os.environ["GEMINI_API_KEY"])
interaction = client.interactions.create(
model=os.environ["GEMINI_MODEL"],
input="Answer this dated research question with sources.",
tools=[{"type": "google_search"}],
)
for step in interaction.steps or []:
print(step)
client.close()
Choose a currently supported grounding model from the official documentation and configure GEMINI_MODEL explicitly. The revision tested the request shape with installed google-genai 2.29.0 and a local HTTP mock. No paid Gemini call, real search or answer-quality measurement was performed in that test.
Inspect what the interaction actually returned
| Step or content | Check |
|---|---|
| google_search_call | Which searches were actually requested |
| google_search_result | Returned search information and suggestion data |
| model_output text | The text being presented to the user |
| url_citation annotations | Which URL and text span the annotation identifies |
| No search step or no citation | Do not assume a grounded answer was produced |
Check support before treating citations as evidence
Preserve annotation URLs and text spans with the answer. Inspect the cited source to determine whether it supports the relevant claim and date. A citation marker is not a guarantee of factual correctness.
An external SearchCans search-and-read workflow supplies evidence through its own application contract; it is not the same operation as Gemini’s native Google Search grounding. Compare the two from the task and returned evidence, not from obsolete Gemini 1.5 context-window claims or guessed prices.
Reference documentation: ai.google.dev — /gemini-api/docs/google-search.