The short answer
Record a fixed prompt set and the exact cited URL, rather than treating one chatbot answer as a stable ranking.

A three-step workflow
Define buyer questions
Start with a stable list of questions your target buyer would plausibly ask. Give each question a permanent ID.
Save answer and sources
For each run, save the complete answer and the destination URL of every relevant citation, including missing citations.
Compare repeated runs
Compare the same prompt IDs across dates. Report changed sources and answer context before summarizing counts.
Illustrative example
Put the decision in context
Use ten questions covering discovery, comparison and purchase. Log platform, date, country, browsing state, answer and cited URL.
What to record in your working sheet
Keep the evidence beside the decision so the work can be checked and handed over. For this task, use these fields:
- Prompt ID and exact question
- Platform, mode, date and context
- Mention, citation URL and supporting passage
- Observation limits and next check
A mistake to avoid
Counting a brand mention as a clickable citation.
References and further reading
The workflow and illustrative example are SEOByIndia editorial guidance. Use the linked primary documentation for platform requirements; interfaces and policies can change.
Need help applying this to your site?
Share your website, the problem you are investigating and the evidence you already have. We can discuss a focused scope and the next useful check.
Discuss an AI visibility audit