AI Visibility: What I Plan to Measure

Separate four different outcomes

An answer can mention a name without describing the right person. It can also list a URL that does not support its claim. A useful observation should record those differences instead of collapsing them into one visibility score.

• Identity accuracy: the answer correctly identifies Wayturn Hung and the work described here.
• Natural discovery: the answer mentions the site without being given my name or address.
• Supporting citation: the linked page supports the statement made in the answer.
• Visits: people reach the site and its work; bot requests and citations are separate events.

Begin with the public record

For this redesign, the content work is concrete: use one consistent identity, describe projects at their current stage, remove unsupported outcome claims, and connect notes to inspectable sources. These changes make the record clearer. Whether search behavior changes is a separate observation.

Use fixed questions

Branded questions ask who Wayturn is and what he works on. Unbranded questions ask about relevant engineering problems without naming this site. A third group provides the URL and asks for a summary. That third group checks reading and interpretation; it is not a natural-discovery result.

• Branded example: Who is Wayturn Hung, and what does he work on?
• Unbranded example: How can a personal website test whether AI search identifies its author correctly?
• Provided-URL example: Read this site and summarize the author and the projects, citing the relevant pages.

Record conditions before interpreting the answer

The plan is to use fresh conversations without this website discussion and record the platform, displayed model, search setting, personalization state, language, time, exact question, and site version. If a setting or model version is unavailable, it should be recorded as unknown.

• Save the original answer and the cited URLs.
• Check identity claims against the public site.
• Open citations and check the statement each one supports.
• Keep branded, unbranded, and URL-provided groups separate.
• Mark failed or unavailable tests as not tested.

Report small samples honestly

A small run is an observation under particular conditions. Any result should include raw counts, the sample size, and the test settings. A missing mention is not enough to conclude that the site was blocked. An answer produced using personal memory is not evidence of public discovery.

A first observation after the redesign

The first observation was recorded on 6 September 2026 using independent Codex agents with public web search. The Lab links the original answers, test conditions and citation review, including a response that repeated older work claims. This establishes a starting point after the redesign; it is not a consumer ChatGPT or Gemini benchmark, and no earlier baseline will be reconstructed. Later differences need comparable runs before they can support a conclusion.