# GeoGenius real-estate AI citation observations — v1

Release prepared October 20, 2026 from existing saved exports. Source report generated May 14, 2026 at 16:36 UTC. That is a report-generation timestamp, not the collection interval. Per-observation timestamps were not present in the exported CSVs, so exact collection dates are unknown in this release.

## Files and reproducibility

- probes.csv: one attempted (city, state, engine, query) observation; 720 rows, 622 completed and 98 failed. Query is the recorded prompt text. model_used is the export value and may be blank on failed requests. citation_count is the recorded count. status replaces raw provider error text.
- citation-domains.csv: counts grouped by city, state, density tier, region, engine, question category and the exported citation_domain. Sum citation_count to recover 4,455 citations; count distinct citation_domain to recover 1,002 domains. Repeated citations are not deduplicated into unique visitors or recommendations.
- summary.json: attempted/completed counts by engine, recorded model versions, headline totals, source generation timestamp and source-file hashes.

The included fields are aggregate/public-market observations. Account errors, billing messages, latency, raw answers, named-agent extraction, full citation URLs and titles are intentionally omitted. No private customer outcome or product-lift claim is represented. Domain fields retain the source export values; this release does not silently reclassify them.

Run `python3 scripts/build-public-research.py --check` against the repository’s saved exports to validate the release. Independently reproduce the public totals by counting probe status and summing the domain table. The reduced table cannot reproduce named-agent extraction or editorial source-category classifications; those are separate analyses on the report, not independently reproducible from this release.

## Interpretation

A citation is a returned source reference, not proof that an agent was recommended. A mention names an entity; a recommendation presents an entity as a possible choice. These tables release citation observations, not a complete verified recommendation dataset. Failed requests remain in the attempted denominator and must not be treated as completed answers with no citation. Engine comparisons have unequal completed sample sizes.

The observations use APIs and specific model configurations. They do not represent every consumer product, user, location, date or personalized answer. A domain’s prevalence is descriptive; it does not establish a causal ranking lever or the effect of GeoGenius. No sampling-based confidence interval or market-wide representativeness is claimed.

## Citation and reuse

Suggested citation: GeoGenius, Real-estate AI citation observations, ai-citations-2026-v1, source report generated May 14, 2026; public release prepared October 20, 2026. https://geogenius.ai/ai-report#methodology

These GeoGenius observation tables may be reused with attribution to the version and methods. This permission does not grant rights to third-party content, brands or the underlying cited pages. Retain the limitations when reporting results.