AI Volatility Tracker

It is the rate at which AI answers change from one day to the next — how many sources they cite, which domains those sources are, how long the answer is, how many brands they name, and whether ads appear. When the same engine answers similar prompts with a different mix, volatility is high.

Each metric is a day-over-day relative change, capped at 100. The headline score is the equal-weight average of the tags you have selected, for one engine. Citations is how many in-answer links there were; sources is whether those links came from the same websites. Both ignore retrieved URLs that never appear in the answer. Sources uses prompts that ran yesterday and today.

Engines change on different days, for different reasons. Averaging ChatGPT with Google AI Mode would hide the spike you actually care about — a default-model swap, an index refresh, a shopping rollout.

The same prompts Vercite customers run against each engine, not a fixed research panel. That mix covers many languages, industries, and query types. We focus on patterns that hold across the whole set. Sources uses only prompts that ran on both consecutive days, so that comparison is like-for-like.

If an engine almost never returns ads, that tag is omitted from its average. A model that never shows ads should not look “stable on ads.” Toggle tags to see the ingredients that do apply.

Character length of the cleaned answer: shopping-widget markdown and data:image blobs are stripped first, so a vendor payload change is not mistaken for shorter prose.

Once a day, after the UTC day closes. The score you see is yesterday compared with the day before.