A source in an AI search response is not automatically a citation.
Define a reproducible question
Choose a real question that a visitor might ask and record its exact wording, language, target domain and model. Do not change all four between observations. This workflow uses the configured Perplexity Agent service with web search; it does not test every assistant or claim to reproduce a logged-in user’s personal chat.
Keep retrieved sources separate
The service can return search-results items and a generated answer. A page in the results is evidence that the provider retrieved that source, not necessarily that the answer cited it. The tool associates explicit bracketed reference IDs in the answer with the corresponding source records before calling a target cited.
Use unknown when evidence is missing
If there are no returned sources or no usable citation markers, the result is unknown. It is not a zero ranking. A missing target in one response does not prove that no AI system has ever cited it. Citation order also does not map to a Google organic position. Read the actual answer and the linked page rather than trusting a badge.
Save a limited observation
The signed-in history stores the target, question and returned result only when you run and save a measurement. These checks are manual. Compare the same target and question later, account for provider and source changes, and use the observations to investigate content coverage. Do not turn one response into a guaranteed visibility claim.