What to record before a run
Record the exact questions and target domain, selected platform, language and requested market. Separately save the actual search mode, visible model choice and personalization conditions. The business's country does not prove the interface's location. If the service's region is not controllable or visible, record that explicitly. This conditions record is more useful than trying to guess afterward why the same product appeared in a different set of recommendations.
Why an unchanged panel matters
In AURA GEO, the manifest fixes the questions, AIs, country, language and questionSetHash. This preserves the composition of a specific run rather than replacing it with a modified current list. Editing a question calls for a new check. If the panel grows, show the new panel's results separately or compare only the overlapping subset. A share of linked answers across five questions cannot be compared without explanation to the share across twenty different questions.
What to do with unavailable checks
In an educational example, suppose the first panel has six actual answers and two tasks in the second are blocked because sign-in is missing. You cannot record those two states as ‘no link’ and declare a decline. Show completed answers, blocked checks and the same tested subset. If the comparison is incomplete, save precisely that conclusion. An operational execution failure differs from a change in answer content and needs its own next step.
A practical repeat-check sequence
Open the original run, choose the same platform and repeat the unchanged questions. Use separate neutral conversations where possible and save the observable session conditions. Do not add the brand only in the second run or ask the model to reconsider its answer to get the desired result. Each new actual answer gets its own record; resaving the same answer uses the previous stable key to avoid artificial duplicates.
Reading the result in the product
AURA provides the original texts and sources, so a change can be checked for a specific question. Establish first whether a link really appeared, the brand context changed or merely a different source set was used. Then compare this with the work log. Answers may differ even under the same conditions; repeatability does not mean guaranteed stability. Prepare a brief conclusion stating what is comparable, what changed and which further run or correction is justified now.
Conditions and limitations
- One panel's result is not the volume of all queries and does not guarantee a future AI answer.
Explore the details
A comparable, verifiable conclusion from the saved sample with a specific next action.
Compare run conditions