There is no ChatGPT ranking to track.
The question sounds reasonable and has no correct answer. Understanding why changes what you should ask for instead.
Can you track your ranking inside ChatGPT?
A search engine returns a list. The list is ordered, largely stable within a location and time, and two people asking the same thing usually see something close to the same result. Position ten means something.
An assistant composes an answer. It is generated per conversation, varies between two runs of an identical prompt, changes with the wording, the account, the region, the model version and whatever the system retrieved at that moment. There is no ordered list to occupy a position in. A vendor reporting your rank inside a model is reporting something else and labeling it a rank.
What can be tracked honestly
- Whether the business is named at all when a question in its category is asked, and how the answer describes it. That is an observation about one answer on one date, not a position.
- Whether the evidence a model would retrieve exists and is reachable. This is the part that is fully observable, fully in the client's control, and the part that actually moves outcomes.
- Supported platform data. Where a platform reports exposure and clicks, those numbers are real and dated and can be trended.
- Citations and mentions in public, which can be found and re-found.
What a sample is worth, and how it should be reported
Asking an assistant a category question and recording what it said is a legitimate observation. It is worth something when the question set is fixed in advance, the wording never changes, the date and model are recorded, and enough runs exist that a single answer is not read as a trend.
Change the wording and the series resets, because the new string is a different question. That is why a tracked question here is never quietly reworded to sound better: a reworded question with an unbroken chart is the most convincing wrong number a report can contain.
Why the wrong question is expensive
A business that buys rank tracking inside a model buys a number that cannot be verified, cannot be reproduced and cannot be acted on. When it moves, nobody can say why. When it does not move, nobody can say what to do. Both outcomes lead to spending on whatever the vendor sells next.
The better question is the one with a checkable answer: can these systems find, parse, verify and quote the evidence about this business, and is that evidence getting stronger or weaker? That is what ongoing AI visibility tracking reports, and every part of it has a source.
Three questions for anyone selling rank tracking in an AI model
- What exactly was asked, and has that wording ever changed? If the prompt set is not fixed and published, the series cannot mean anything over time.
- How many runs is each data point, and on which model version? One run of a probabilistic system is an anecdote. A number with no model version attached cannot be compared with itself three months later.
- What does the number do when you change nothing? A metric that moves on its own between two identical weeks is measuring variance, and any improvement it later reports is indistinguishable from the same variance.
These are not gotchas. A vendor with honest answers to all three is measuring something real and narrow, and saying so. The ones worth avoiding are the ones whose answer is a chart.
This is one question beneath AI visibility tracking, the ongoing work Digilu does through The Observatory. The free point-in-time baseline is AIOInsights. Digilu cannot make a private AI model recommend a business. It can make the public evidence clearer, stronger and easier to verify, then track whether visibility improves.