There are 4 ways brands are tracking their ChatGPT visibility right now. 3 of th
By Toddpie · June 30, 2026 · Curated by George's Blog
There are 4 ways brands are tracking their ChatGPT visibility right now. 3 of them are useless. Here's the scorecard:
F → Single prompt, screenshot the result
A keyword or a one-off prompt tells you almost nothing. People don't ask AI for products the same way twice. That screenshot is one lottery ticket. It is not data.
D → API pull
If you are tracking through an API call, you are probably getting wrong data. The API, the UI, the app, and different models return different results.
C → Keyword or single-prompt dashboards
Still built on the wrong unit. A keyword tells you what was typed. Not whether your product showed up for the need behind it.
A → Topic-level repeated sampling with a confidence band
At Vektor10 we track the topic: a need a shopper is trying to meet. "A nonstick pan that's actually safe." Under each topic sits a set of real, varied prompts tagged by intent. We run that set repeatedly and report how often your product shows up across the whole topic, with the sample size and a confidence band attached.
Rand Fishkin at SparkToro ran roughly 3,000 runs and found that AI answers almost never repeat - you're gonna get different brands, different order, different count every time. That's why we never report "rank in ChatGPT."
That number is a lottery result. Anyone selling it is guessing.
If your tracking method can't tell you how often your SKU shows up across a representative sample of real queries, you're tracking luck.