How often, and how favourably, a brand shows up inside AI-generated answers. It is becoming its own reporting line because search rank no longer predicts it.
Also known as AI search visibility, LLM visibility, model presence, brand presence in AI answers
AI visibility is the measurement of whether a brand appears when a model answers a question in its category, and what the model says when it does. It covers presence (are we mentioned at all), position (are we first or fourth in the list the model produces), framing (are we the recommendation or the cautionary comparison), and citation (does the answer link back to our page or to a review site describing us).
It is treated separately from SEO because the relationship between ranking and being cited turned out to be loose. A page can rank well and never be quoted. A brand can be recommended constantly on the strength of third-party coverage it does not control. And with AI Overviews cutting organic traffic somewhere in the 18 to 47 percent range depending on category, the traffic that used to prove visibility has stopped arriving.
Methodology is the unsolved part. Answers vary by user, by session, by model version, and there is no equivalent of a rank tracker that everyone agrees on. Most current practice is sampling: run a fixed prompt set across several models on a schedule and score the outputs, accepting that you are measuring a distribution rather than a position.
Clients have already started asking for it in monthly reporting, and most agencies do not have a defensible answer for how they would measure it.
It is a reporting gap that is being filled by whoever answers first. The request is arriving from client-side marketing directors who read the same trade coverage everyone else did. Whichever partner produces a credible methodology gets to define the metric, and defining the metric is how a channel becomes a retained scope.
The levers sit outside the media plan. Improving AI visibility usually means third-party coverage, structured data, documentation quality and consistent naming across the web. Almost none of that is paid media, and a lot of it is work agencies have historically pushed to the client’s own team.
Sampling honestly is the hard part. It is very easy to build a prompt set that flatters the client and a scoring rubric that always trends up. Locking the prompt set, versioning it, and reporting the variance rather than a single clean number is what separates a real measure from a comfort metric.
A performance team loses a quarterly review badly: organic sessions are down 31 percent, rankings are flat, and they have no explanation the client accepts. The following month they build a crude tracker, forty buying-intent prompts run weekly across three models, scored by an analyst for mention, position and sentiment. The first read is uncomfortable. The client appears in 12 percent of relevant answers while its main competitor appears in 61 percent, almost entirely on the back of a comparison article on a review site nobody at the agency had been tracking. The resulting plan contains almost no media: fix the product documentation, pursue three specific third-party placements, and standardise how the product line is named across every property. It is the first plan the agency has presented that the SEO lead, the PR lead and the content lead all have to sign.
The automations and agents module of the workshop teaches you how to build AI workflows that compress the busywork without taking the craft out of the studio.