A ChatGPT screenshot is not an AI-visibility audit

I keep seeing the same screenshot shared as proof: ChatGPT named a product once, so "we're visible."

One lucky answer is a story. An audit is a log.

If you only capture the wins, you can't tell:

• whether you show up for the queries that actually matter
• whether you were named, cited, or just vaguely gestured at
• whether yesterday's mention still holds after a model refresh

What changed my mind was treating it like a tiny experiment, not a highlight reel:

1. Write 10–20 prompts a real buyer would type (not brand vanity searches).
2. Run them on a schedule for 30 days.
3. Log the outcome every time: nobody / competitor / you-but-not-cited / you-cited.
4. Only then decide what to ship on the page.

Screenshots are great for morale. They're terrible as a strategy input, because they have no denominator and no falsification rule.

Curious how others here are logging AI answers without turning it into a full-time job — spreadsheet, tags, something else?