# Review an AI survey summary

Life Linked Lab, 2026-11-09. Free example and review worksheet.

The accompanying twelve rows are author-designed fiction. They do not represent
customers, real respondents, market demand, or a statistically representative sample.

## Reproduce the arithmetic locally

Requires Python 3.10+; standard library only. Put the script and JSON in one folder.

    python survey_summary_check.py --input fictional-responses.json --out reviewed-counts

The script writes reference-counts.json and claim-checks.json. It does not upload
anything, call a model, read a Google account, or analyze prose automatically.
The six claim checks are specific to this twelve-row fixture.

## A hand check before reading the AI narrative

- Total rows: 12. Contexts: Solo 6, Team 6.
- Multi-select question: 11 nonblank responses, one blank (R06).
- Clear steps: R01, R02, R04, R05, R07, R09, R10, R12 = 8 selections.
- Reusable files: R01, R03, R05, R07, R08, R09, R11 = 7 selections.
- Video walkthrough: R04, R08, R11 = 3 selections.
- Total selections: 18, because a person may select more than one option.
- Open comment: 11 nonblank, one blank (R12).
- Clear steps is 8/11 = 72.7% of valid answers, or 8/12 = 66.7% of all rows.
- Solo valid multi-select answers: 5, not 6. Clear steps = 4/5 = 80.0%.
- Team valid multi-select answers: 6. Clear steps = 4/6 = 66.7%.

## Blank review record for an actual permitted survey

Source/export snapshot and date:
Number of submissions:
Question under review:
Question type (single choice / multiple choice / text / other):
Nonblank answers to this question:
Excluded rows and reason:
One quoted AI claim:
Underlying count and supporting row IDs:
Denominator used and why:
Independently calculated percentage:
Does a subgroup have a different denominator?
Which parts are observed counts, inferred themes, or proposed actions?
What would need confirmation before acting?
Reviewer and decision:

## Use in Google Forms

The official November 5 announcement and the English Help page were read on
November 9, 2026. Check current eligibility, rollout and account permissions.
At observation time, Help specified at least 12 responses and 3 questions,
quiz mode off, and access to the linked response Sheet. Our UI required linking
responses to Sheets before generating Actionable Insights.

Do not import these fictional rows into a real customer-results dataset. Use a
separate test form; the Lab's test form was closed to further submissions after
the run. Never treat this fixture's preferences as a business decision.

## File meanings

- fictional-responses.json: fixed source rows.
- reference-counts.json: deterministic arithmetic, not model output.
- observed-insights.txt: narrative displayed by Forms in one actual run.
- survey_summary_check.py: local arithmetic and structured-claim checks.
- claim-checks.json: two valid structured claims and four deliberately invalid controls.
- review-kit.md: this worksheet and instructions.

Sources:
https://workspaceupdates.googleblog.com/2026/11/get-actionable-insights-with-gemini-in-Google-Forms.html
https://support.google.com/docs/answer/16231981?hl=en

One run is not a benchmark of reliability, speed, price, all languages or plans.
The model version and per-generation cost were not exposed in the inspected UI.