A screenshot of ChatGPT isn't a diary

If you ask ChatGPT once, screenshot the answer, and call that "our AI visibility," you measured a mood. Same prompt, new chat, different shortlist. SparkToro's jitter numbers are why a screenshot isn't a diary.

A useful first measurement is smaller than a month.

Pick five questions a real buyer would type — not "what is [my brand]." Buyer questions. The ones that spend.

Run each one in a fresh chat, on more than one engine if you can. For every answer, write one row:

• the question
• the date
• you / a named competitor / nobody

Stop. That is a mention rate, not a rank. You are not optimizing a page yet. You are checking whether the model even has a slot for you when the money question is asked.

If the row is "nobody," the next hour is better spent rewriting the first screen of one page so a model can extract who you help — not adding llms.txt and hoping.

If you only need that first row, not a 30-day lab, we put the sheet up as a free file: The 5-Prompt Citation Snapshot.
https://normbrytande.gumroad.com/l/quwzbl

Five buyer-intent prompts. One log. Three buckets. Not a citation promise.