# AI visibility baseline kit

Use the [full public method](https://www.jacgrowthmarketing.com/guides/ai-visibility-baseline-kit/). Files contain no client information. The completed JAC file is an actual dated starting record, not a result forecast.

## Work in this order

1. In buyer-questions.csv, replace the blank YOUR-Q01 fields with one question, its buyer decision, and its source. Mark generated questions as hypotheses. Choose a permitted surface and repeat plan before collection.
2. In answer-observations.csv, keep one row per attempted check. Save complete answers and displayed source links in an approved evidence location. A new chat alone does not establish neutral memory or personalization. Record only visible settings; mark hidden settings unknown.
3. Use completed, failed, or missing as status. For failed and missing responses, leave outcome fields blank and record the reason. A rerun is a new attempt, not an overwrite. Put the earlier observation ID in repeat_of_observation_id. Keep the initial attempt blank in that field.
4. Review mention, owned_citation, and recommendation as yes, no, or unknown. Mark a recommendation unknown until a passage establishes that role. Use accuracy_status match, mismatch, or unassessed; retain the fact source and passage.
5. In page-checks.csv, connect one controlled change to the observation. Public fetch, native index record, live availability, and generative inclusion are separate checks. Keep unavailable account checks unknown. Status can be proposed, in_progress, completed, or blocked. Completion means the stated change passed its checks, not that an outside answer changed.
6. Keep platform reports outside answer-level rates. Save date window, property, filter, aggregation, and unavailable displayed values before export. Preserve missing values even when an export writes zero.
7. Repeat exact questions under recorded conditions after a dated intervention. Compare only matching surfaces, panel versions, settings, and assessment coverage. Report assessed outcome counts and assessed denominators, plus failed and unreviewed attempts. No composite authority or market-share score.

The question sheet includes both JAC spot-check questions, and the observation sheet includes the filled JAC-O01 example, alongside blank working rows. JAC-O02 appears in the completed JAC record. Keep your own IDs distinct. Public evidence references summarize the reviewed answer; retain your complete captured answer privately so a reviewer can inspect it. JAC-Q01 and JAC-Q02 are distinct spot-check questions, not the operational repeated portfolio.

## Collection brief

Investigate the exact recorded question on the permitted surface. Save the completed answer and all displayed citation links. Record the actual time and visible settings. Do not add product facts to an unbranded question, repair the answer, or fill missing evidence. Review material claims against the supplied current fact record. Identify one inspectable issue and the smallest controlled repair. Keep outside answer changes separate from implementation completion.

## Completed JAC record

The 24 Google generative AI impressions are a property total for September 9 to October 6, 2026, observed October 9. The ChatGPT description is one consumer observation, not an API result or visibility rate. Its retired service labels conflict with current JAC pages; the quoted public prices matched the current offers. The Google live availability test does not establish indexing. The [public experiment log](https://www.jacgrowthmarketing.com/blog/jac-ai-search-experiment/) preserves the context. One separate unbranded answer did not name JAC; this is one hypothesis-question observation, not universal absence. The checked Search generative AI control was Include and remained unchanged.

Keep customer names, contact details, private transcripts, credentials, and confidential product facts out of public copies.