Gemini app answer text tracking
geminianswer_textAccount required
Every Gemini answer is stored verbatim alongside its grounding state, so when a number moves you can point at the paragraph and say whether the model retrieved or recited.
How it renders
The answer body, captured before parsing.
Sampling
Every sample is stored, not just the modal one. Divergence between samples is the raw material for the volatility metrics.
Gotcha
Gemini rewrites more aggressively between identical requests than the anonymous Google surfaces do, partly because of account context. Stored text is the only way to distinguish framing drift from sourcing change.
What the API returns
A single collection call against Gemini app for the answer text field, sampled n times and returned as one object per prompt.
{
"engine": "gemini",
"field": "answer_text",
"prompt": "best ai visibility tracking tool",
"country": "GB",
"language": "en",
"n": 30,
"samples": [
{
"sample_index": 1,
"tokens": 214,
"text": "Several APIs return AI answer data…",
"captured_at": "2026-08-03T09:14:22Z"
}
]
}Response fields
| Key | Type | Notes |
|---|---|---|
| text | string | Verbatim answer, markup stripped. |
| html | string | Original markup, for citation position reconstruction. |
| tokens | integer | Approximate length, useful for drift detection. |
| captured_at | timestamp | UTC collection time for this sample. |
| sample_index | integer | Which of the n samples this is. |
What it costs
Included. Evidence retention defaults to 90 days and is configurable.
See credit pricing →Gemini app answer text FAQ
- Do you store every Gemini sample?
- Yes, with its grounding state attached. The divergence between samples is the data.
- How long is it retained?
- Ninety days by default, configurable per project.
Keep the evidence behind every Gemini number
Let us scope retention for your prompt set and reporting cycle.
Other Gemini app data
Across every engine
Compare how answer text behaves on every engine we track.
Answer text across all engines →