Short answer: answers do not have ranked positions. A brand is named, mentioned in passing, or absent, and separately cited or not cited. A number that looks like a rank is almost always the order of appearance inside a single sampled answer, which is not stable enough to trend.
Why the metric is tempting and wrong
Rank tracking worked in search because a SERP is an ordered list that is broadly stable within a day. An answer is generated text. Ask the same question twice and the order of brands can change without anything about those brands changing. Build a chart on that and you will spend your time explaining noise to whoever reads it.
If a metric can move 30% overnight with no action taken and no cause identifiable, it is not a metric. It is a sampling artefact.
What to measure instead
| Metric | Definition | Why it holds up |
|---|---|---|
| Mention rate | Share of a fixed question set where the brand appears | Averaging over many questions absorbs per-run variance |
| Recommendation rate | Share where the answer actively suggests the brand | Separates being referenced from being chosen |
| Citation rank among sources | Where your domain sits among all cited sources | An ordinal position over a real population, not a share that drifts |
The conditions that make even these trustworthy
- A fixed, versioned question set with no brand names in the questions. A question containing your name will be answered with your name; that is an echo, not visibility.
- Identical detection for every brand, yourself included. Any asymmetry moves in your favour, because you wrote the detector while looking at your own name.
- Missing runs recorded as null, never zero. A week with no collection is not a week of zero visibility, and rendering it as 0% turns a flat trend into a fake decline.
What a good report looks like
Two rates and one ordinal, over a question set you can show someone, with the runs that failed visible rather than hidden. It is less exciting than a rank chart. It is also the only version that survives being checked.