Quick answer
Scores move between audits mainly because models update without notice, most AI answers are sampled rather than fixed so identical prompts can return different mentions, and the pages an answer draws on keep changing. One run moving is rarely meaningful; the same direction across three or more runs on a fixed prompt set is the signal worth acting on.
Key takeaways
- Model updates change how a question gets answered, with no announcement tied to your brand specifically.
- Most AI answers are sampled, not fixed, so the same prompt can return a different mention on back-to-back runs.
- One run moving is noise. The same direction across three or more runs on a fixed prompt set is a trend.
Where the movement actually comes from
Three things drive most of the change you see between audits, and none of them are a bug in the measurement. Separating them makes the next move obvious instead of guesswork.
- Model updates: providers retrain and re-tune their models on their own schedule, and a routine update can shift how any brand gets described without a public changelog entry naming yours.
- Answer sampling: chat-style answers are generated token by token with some randomness built in, so asking the identical question twice can surface a competitor one time and not the next, even with nothing else different.
- Source freshness: an answer that cites or draws on a web page will reflect whatever that page says today. A page you updated last week, or a competitor's page that changed, can move an answer without any change on your side.
Telling noise from a real trend
| What you see | Likely explanation | What to do |
|---|---|---|
| One prompt's mention flips between two runs a day apart | Normal sampling variance | Nothing. Wait for the next scheduled run. |
| Mention rate drifts in the same direction for three or more runs | A real underlying change | Look at what the answer now cites or describes, and whether a competitor's content changed. |
| A whole platform's answers change the same week for many prompts, not just yours | A model update on that platform | Check if competitors moved too. If everyone moved together, it is the model, not you. |
| A single prompt stays flat while every other prompt in the set moves | That prompt may be worded narrowly enough that one source dominates it | Review the prompt wording itself before assuming anything changed about your brand. |
A single bad-looking run is not an incident
What is actually worth investigating
- A drop that holds for two full cycles on the same prompts, not just one.
- A new omission on a prompt that used to mention you reliably, right after a competitor published something relevant.
- A citation that now points to an outdated or incorrect page about your brand.
Should I re-run an audit immediately after a bad result?
Only if you changed something specific you want to verify. Otherwise the next scheduled run on the same prompt set will tell you whether it was noise.
Does a bigger prompt set reduce this kind of noise?
Yes. A single prompt is the noisiest unit you can look at; a full set averaged over time smooths out normal sampling variance.
Cite this page
BrandWater AI Research. (2026, 5 October 2026). Why your AI visibility score moves between audits. https://brandwaterai.in/blog/why-your-ai-visibility-score-moves-between-audits
How we work
Figures are dated and linked to their sources. Where none exist we say so. Read our methodology and AI transparency pages.