What replaces keyword rankings as the metric for AI search?
Nothing maps one to one, but the closest workable metric is citation share: across a fixed set of customer questions, the share of answers in which you are named or linked, and who appears instead when you are not. A rank is a position in a list that exists. Citation share is a rate across a sample. It is noisier, it needs repeated observations, and it cannot be checked once a month and trusted.
Why rank does not translate
A ranking assumes a stable, ordered list that everyone sees. Generated answers have none of those properties. They blend several sources, vary between runs of the identical prompt, and differ by user and location.
Vendors who report an AI ranking are either reporting the position of a citation inside one answer on one run, or they are reporting something they made up. Ask how many observations produced the number. The answer is usually one.
There is also no equivalent of an impression. You cannot know how many people asked a question at all, so you have no denominator for opportunity. Every metric in this space is a rate over a sample you chose, which makes the choice of sample part of the methodology rather than a detail.
Defining citation share so it survives scrutiny
The plain version is appearances divided by questions run, computed per assistant: if your panel has sixty questions and you are cited in nine answers on a given assistant, citation share is fifteen percent for that assistant on that run.
The better version weights questions by commercial value, so a question that precedes a booked job counts more than a definitional one. Whichever you choose, publish the definition next to the number. A metric nobody can reconstruct gets ignored the first time it is inconvenient.
Track who is being cited instead of you
The most useful column is not yours. It is the list of sources occupying the answers you want. Frequently it is not a competitor at all - it is a directory, a review aggregator, a manufacturer's support page or a national franchise's content library.
That changes the strategy. If a directory owns the answer, the play may be to be well represented inside that directory rather than to try to displace it. If a manufacturer owns it, the play is the local and decision-stage layer they will never write. This is the same reasoning behind how we structure AI search programs.
Track the type of source as well as the name. A market where answers come mostly from directories calls for different work than one where answers come mostly from operator websites, and the mix tends to shift over time.
Keep an outcome metric above it
Citation share is a leading indicator, not a result. It can rise while nothing changes in the business, and it can be gamed by choosing flattering questions.
Keep booked jobs and revenue from organic and direct sources as the metric above it, so the leading indicator always has to explain itself against something real. That hierarchy - outcome first, diagnostics beneath it - is how revenue reporting should be organized regardless of channel.
Topics: metrics · citation share · rankings · KPIs
Have a version of this question about your own business?
The useful answer usually depends on which systems you run and how they're connected. That's a conversation, not a blog post.