If you are responsible for how your business appears in AI answers, you need a clear definition of what you are trying to observe. AI search visibility describes whether and how a brand appears when people ask an AI assistant about its market. The term has no standard definition and no standard unit, so the first useful thing to know is what any given number is counting.
What is observable
Only one thing can be observed from outside these systems: ask a question, record the answer. Everything a visibility product reports is built from that.
From a recorded answer you can establish four things.
- Whether the brand was named, and in what context, since being named as an option and named as a warning are both mentions.
- Which sources were cited, which are usually third-party pages rather than the brand's own site.
- Which competitors appeared alongside or instead.
- How all of that changes across repeated runs over time.
What cannot be observed from outside is why. No product can see how an assistant selected its sources, what weighting it applied, or what it consulted and discarded. Anything describing a mechanism inside the service is inference, and should be read as such.
Why there is no rank
Search returns an ordered list, so position is meaningful. An answer is prose. Many answers have no ordered list. Where a list appears, the next run may change its order or name different options.
Answers also vary between runs for the same prompt, because generation is not deterministic and retrieval changes. This is normal behavior, covered in why repeated answers vary.
So a product promising your brand a stable position in every answer is describing something these systems do not provide. What remains consistent across answers is presence, absence, comparison against competitors, and direction over time.
The measurements that are worth reporting
Presence rate. The share of runs of a given prompt in which the brand was named, over a stated period. It needs its denominator: presence in seven of ten runs over a month means something, and "we appear in ChatGPT" does not.
Competitive comparison. Which brands appear where you do not, across the same prompt set. This is fair regardless of variability, because both brands face the same conditions.
Cited sources. The pages answers rely on for your topics. This is the most directly actionable output, because it is a list of places worth appearing on.
Change over a series. Movement across weeks, which outlasts run-to-run noise in a way that any single reading does not.
What it cannot tell you
It cannot attribute traffic or revenue. The path from an answer to a visit is frequently indirect or absent entirely, since a good answer may resolve the question with no click at all. That is covered in why answer visibility is not traffic.
It cannot prove why something changed. A brand appearing this month and not last month could reflect your work, a competitor's work, a change to the service, or ordinary variation.
It cannot guarantee an outcome. No work makes an assistant name a brand, because nobody controls the systems being measured.
What the measurement is for
Two things, and being clear about them prevents buying the wrong tool.
The first is knowing where you stand, which mostly means knowing where you are absent. Absence across the questions your buyers ask is the clearest and most actionable finding this measurement produces.
The second is finding the material that answers are built from. When an answer cites six pages for a question in your market, those pages are being used to inform your buyers. Whether you appear on them, whether what they say is current, and whether they accept submissions are three concrete questions with concrete work attached.
That second use is the connection between measurement and action, and it is why visibility tracking on its own tends to produce reports rather than results.
Where to start
Write down the ten questions your buyers ask, in their words, including the ones about your brand by name. Check what the assistants say today. Read the answers rather than counting them.
Most teams find something specific and fixable in the first hour: an out-of-date description, a comparison page that lists competitors and not them, or a roundup that everyone in the market cites.
Start a visibility review with the buyer questions and appearances you intend to track. You can then explain what the results show and choose a relevant next task.
See how to select a prompt set, how citations differ from mentions, or how AEO, GEO and SEO relate.