How confusion arises
Someone opens a report and sees a detailed answer. It could be a week-old record, a fresh browser check or research by the current agent. The texts look similar but answer different questions. History helps continue work; a new answer is needed for repeat measurement; research helps examine a topic. Before interpreting it, establish which action happened now and where the displayed text came from. Otherwise, loading history can easily be passed off as new research.
Reading through MCP
aura_get_observations and aura_get_monitoring_run return retained project data. They do not call a hidden provider API or automatically refresh a previous record. The observation time shows when the answer was obtained, while metadata indicates who saved it and in which mode. Reopening the same record is not a new successful run, even if the report interface was loaded only today. Where needed, show the observation's date rather than the card's viewing date.
A new answer requires an action
The agent takes the exact question from the task and asks it in an interface that is actually accessible. It then saves the unchanged answer, the links used, the actual model or visible mode, and the date. Browser checks use web_ui; the agent's own research uses agent_research with its real name. If the platform is inaccessible, the accurate outcome is a blocked status with the owner's required action. Substituting another model's answer does not complete this task or confirm that the intended service was checked.
Research without substitution
An agent can read public pages, compare service features and propose a plan. This helps prepare content and choose questions but does not prove the company appears in a normal consumer assistant. An educational example: a Codex review includes AURA GEO while the ChatGPT check has not yet run. The first record remains useful research; it does not increase the count of ChatGPT browser answers. The fact that a page was publicly readable does not turn the research into statistics on all its citations.
How AURA GEO helps here
An observation includes executionMode, engine, model, observedAt and a stable runKey. Checks connect the record to the original question and execution status. Work can resume after a break, and evidence saving can be distinguished from report reading. The server itself does not provide access to other people's AI accounts. The next step may therefore be straightforward: read history, perform one new available check or separately prepare a research task with an accurate description of the executor and conditions.
Conditions and limitations
- A project's sample does not represent all AI users' questions and answers.
- AURA checks the consistency of saved metadata but does not authenticate an answer's origin with the AI provider; the evidence remains a report from the connected agent.
Explore the details
A verifiable conclusion and next step tied to an actual question, answer or event, with no guarantee of indexing or citation.
Inspect answer provenance