# Stage 4 public revision decisions — November 3, 2026

This review changes presentation and adds the question-oriented Learning Hub. It does not collect new answers, alter the historical selection data or transform an observational audit into a causal test. Helen Bird is the consistent public author attribution.

| Recommendation | Decision and implementation | Evidence boundary |
| --- | --- | --- |
| Replace “Discoverable” with “Recommended” | Accepted. The reader title describes observed recommendations. | The experiment records C/M/R; it does not directly test indexing or general searchability. |
| Prioritize Q1/Q2/Q3 groups | Accepted. Per-question n=3 results precede pooled n=9 summaries. | Pooled results combine different tasks and are not population visibility probabilities. |
| Distinguish personal names and associated content brands | Accepted. Explicit person mention/recommendation is separate from verified associated-asset observations. | Recommending The GEO Lab does not imply naming Artur Ferreira. |
| Translate observed features into testable practices | Accepted. The source audit connects observed organization to candidate hypotheses and future tests. | No page feature is established as the cause of an AI recommendation. |
| Make API tool limits visible beside comparisons | Accepted. All cross-environment figures retain the fixed-tool-cap qualification. | Three of nine API answers reached eight calls; source counts cannot isolate a model preference. |
| Deploy the complete technical appendix | Accepted. The public release includes its full expanded answer/source/coding/recovery record and downloadable data. | Private originals and account evidence remain excluded. |
| Build a question-oriented learning center | Accepted as a content-organization practice. A separate Learning Hub connects the four existing reports, datasets and appendices. | Stage 1 keeps its root URL. No improvement in AI discovery or recommendations has been tested. |
| Attribute content consistently and connect related formats | Accepted. Hub entries use Helen Bird and separate experiment, first-publication and revision dates. Only verified related distribution links are included. | Consistent authorship is a hypothesis to test, not a demonstrated person-recognition effect. |
| Put concise findings before detailed evidence | Accepted without changing the existing default-expanded requirement. The concise reader links a fully expanded appendix. | Readers retain access to methods, data and limitations; evidence is not hidden by default. |
| Preserve the Field Manual evidence issue | Accepted. The correction note and remaining denominator/significance inconsistency remain together. | Recommendation is not research verification; underlying source data were not independently replicated. |

Detailed report-specific decisions are also preserved with [the Stage 4 public data](../stages/04-creator-benchmark/data/review-response.md). [Version metadata](../report-versions.json) and [the changelog](../CHANGELOG.md) distinguish this editorial release from experiment-date observations.

> The public release keeps OpenAI Responses API + native web search distinct from consumer ChatGPT Search, retains short-term and small-sample qualifications, and makes no claim about long-term visibility, conversion or GEO optimization effectiveness. No new AI responses were collected for these revisions.
