The short version
Assistant answers in the region are getting shorter and citing fewer sources, which raises the cost of not being one of them. The set of brands quoted for a given commercial question is small and slow to change, so being early to a topic still matters more than being loudest.
Two consequences follow. Visibility now looks closer to a winner-takes-few distribution than to the gradual slope of a results page, so the gap between being cited and being nearly cited is far wider than the gap between position three and position eight ever was. And because the cited set changes slowly, the cost of entering a topic rises the longer you wait.
Why a quarterly cadence
Monthly is noise and annually is too late. Assistant behaviour shifts with model and retrieval releases, which arrive on roughly a quarterly rhythm, and the content and corroboration work you do in response takes about that long to surface. A quarterly review is the shortest honest feedback loop available.
It also forces the discipline of writing down what you saw last time. Most teams trying to track this have an impression rather than a record, and an impression cannot tell you whether anything moved.
Running the ten-question audit
The audit is a spreadsheet and an afternoon. Ten questions, three assistants, one row per question per month, recording which sources were named and whether you were among them. It is unglamorous, and it is the only reliable way to know where you stand, because no rank tracker exists for this.
Choosing the questions
Take them from sales calls, not from a keyword tool. You want the question as a buyer phrases it before they have learned your category’s vocabulary. Ten is enough. Teams that start with fifty abandon the exercise by the second month, which leaves them with one data point and no trend.
Recording the answer
Log the sources named, not the prose. The prose varies every time you ask and tells you very little. The source list is the stable signal, and it is the thing you are competing to join.
Reading the result
You are looking for three patterns. Sources appearing across every question are the regional authorities worth pursuing. Sources appearing once are noise. And questions where the set shifts month to month are the ones still open to you, which is where the quarter’s effort should go.
What we would do this quarter
Pick the ten commercial questions your buyers actually ask, check what the assistants currently say, and fix the pages that should be cited and are not. Then go earn mentions on those exact topics. Ten questions done properly beats a hundred tracked passively.
In practice that sequences into four moves: audit the ten questions, pick the three where the citation set is still moving, fix those pages so the answer sits directly under the heading, then spend the rest of the quarter earning mentions on those three topics specifically. Spreading the same effort across ten topics reliably produces nothing on any of them.
What changes fast, and what does not
Answer length and formatting change with almost every model release, and chasing them is wasted effort. What changes slowly is which sources an assistant trusts for a given topic. Build against the slow thing.
This is also why page-level work has a ceiling. You can write a genuinely quotable answer in a morning. Becoming one of the sources a model reaches for takes quarters, and no amount of on-page work shortcuts it.
If you are already cited
Being in the set is less stable than it looks. The citation list for a question is short, and it reshuffles whenever a source that corroborated you goes quiet or a competitor earns a mention you did not. Presence is a position to defend rather than a result to bank.
Defending it is mostly maintenance. Keep the answer on the page current, because a page that contradicts what an assistant said about you last quarter is a liability rather than an asset. Keep the external mentions alive, since a reference from four years ago carries less weight than one from four months ago. And watch the questions adjacent to the one you own, because that is where a competitor builds the corroboration that eventually displaces you.
How to report this to a board
Resist inventing a score. Report the citation set for your ten questions, your presence in it, and the direction of travel since last quarter. That is three lines and it is defensible.
A composite “AI visibility index” is not defensible, and the first person who asks how it is calculated will establish that in front of everyone. An honest three-line answer survives the question. A number you cannot derive does not.