Separate two tasks. ChatGPT Search may surface public web pages through OpenAI's search crawler. GPT-6 Astra is an API model that a developer can use for research and other workflows; naming that model on a page does not influence crawling or prove that a particular ChatGPT answer used it. The official crawler documentation and model documentation describe these different roles.
Check public discoverability
Inspect the site's robots policy for OAI-SearchBot, then verify that important pages are publicly reachable, return the expected status and expose their main content in usable HTML. OpenAI says that OAI-SearchBot is used for search features and that sites opting out will not be shown in ChatGPT search answers, though navigational links may still appear. A robots change can take time to take effect. Do not edit crawler rules without the site's agreed publishing policy.
Write a source that can be checked
State the organisation, service, audience, location where relevant, scope and limitations clearly. Give a short answer followed by evidence and practical detail. Keep page titles and internal links descriptive. Distinguish a real case study from an illustrative scenario and avoid invented prices or testimonials. These choices make the page useful to people and to any system asked to inspect it; they do not guarantee a citation.
Make model-based workflows testable
If a team uses GPT-6 Astra through an application, it can provide selected pages or documents as inputs and evaluate answers on representative questions. Specify what the application may read, what constitutes a correct answer and when a person must review it. This is an implementation decision by that application, separate from public ChatGPT Search indexing. A model name in metadata or an LLM-oriented file is not an access mechanism.
Measure only what can be observed
Review server logs when available, inspect referral traffic with its attribution limits, and manually test a stable set of questions. Record whether the answer names the organisation accurately and links to the correct page. Search exposure, visits and enquiries are different signals. Do not report a citation count as a precise platform metric unless the platform actually provides it.
Practical next step
Choose three public service or guide pages. Confirm their crawl policy and human readability, then test whether a visitor can verify every important claim from the page itself. The structured data guide helps keep machine-readable descriptions aligned with visible facts.
Verification workshop: evidence before a decision
A proposed exercise on a chosen scope, using fictional data where needed. Record what you observe; this scenario describes no test already performed on your website.
- Action to try
Navigate a page by keyboard, locate its headings and open its main action. Try enlarged text and a narrow screen with the same content.
- Record to retain
Record focus order, understood labels, reached sections and observed obstacle. Identify the page, environment and reproducible steps.
- What this evidence does not establish
Success on this journey does not certify site accessibility. Consistent heading levels do not guarantee usable forms.
Further documentation : W3C WAI — Headings