Does the API calculate prompt volatility?

No. Compare repeated answers and source sets in your own application. The API supplies collected provider outputs; this analysis belongs in your application.

How should failed or empty collections be handled?

Inspect per-target failedRuns and each successful run identity. Empty sources or null fields are valid output; a missing failed run is not an empty successful answer. A completed job can contain failed collections.

Customer-built analysis

Gemini app: prompt volatility

Compare repeated answers and source sets in your own application.

Collection target: gemini. This is customer-built analysis, not a computed API response field.

Build this analysis in your application

Repeated collections can help measure changes in text, brand presence and source overlap. For example, Jaccard similarity compares the intersection and union of two source sets; decide how to handle two empty sets before aggregating.

Customer-built analysis

The API returns individual successful runs and failedRuns, not volatility scores or recommended sample sizes. Failed collection is not evidence that an answer changed.

The gemini target returns collected answer text, markdown, pageUrl and sources with each successful run identity and capture time. Fields can be null and sources can be empty. Gemini can present a login wall that prevents collection. Eligible failures are retried, and unsuccessful runs are recorded in failedRuns instead of fabricated responses.

Collected input for your analysis

These are illustrative successful run excerpts, not full result files or calculated metrics. Submit POST /v1/crawl-jobs, track completion, then download the authenticated result. Outputs depend on the target; a chatbot has no news articles or Google position fields.

{
  "runNumber": 1,
  "runTaskId": "9a7536e7-b610-420a-a34c-5f829956f4d3",
  "status": "completed",
  "capturedAt": "2026-10-09T10:01:00.000Z",
  "text": "Illustrative captured answer text.",
  "markdown": null,
  "pageUrl": "https://gemini.google.com/",
  "sources": []
}

Illustrative provider output. Empty lists and null fields are valid; inspect failedRuns in the result file.

Full result format and failed-run handling →

Prompt volatility FAQ

Does the API calculate prompt volatility?: No. Compare repeated answers and source sets in your own application. The API supplies collected provider outputs; this analysis belongs in your application.
How should failed or empty collections be handled?: Inspect per-target failedRuns and each successful run identity. Empty sources or null fields are valid output; a missing failed run is not an empty successful answer. A completed job can contain failed collections.

Use collected data to build your own reporting and analysis. Our team arranges API access and follows up with next steps.

Request API access Read the docs