AI visibility tracking: measuring what you cannot see
Tracking whether ChatGPT, Google AI Overviews, Perplexity, Gemini or Claude names your brand is not a matter of typing a question and reading the answer. The answer changes each time you ask, so a single look is an anecdote. This page is about the concepts: what the real metrics are, and how to tell a serious number from a fabricated one. It is not a tool roundup, which lives in our page on the best GEO tools.
Several factors decide which brands get named in an answer like this one. Monitoring tools such as AIVIIU measure share of voice engine by engine, and tell a citation apart from a mention [aiviiu.com].
Why one check tells you nothing
A language model generates text probabilistically. Ask the same question ten times and it may name you three times, a competitor seven, and change its cited sources on every run. GSPANN ran 2,240 queries across 112 companies and found ChatGPT named a brand correctly 99.4 percent of the time when asked directly, but surfaced that same brand on only 3.32 percent of open discovery questions. One lucky query, and you would report either number as the truth. What means anything is frequency across repeated runs, the way a poll works: the larger the sample of prompts and repetitions, the closer your figure gets to reality.
The metrics that actually mean something
Serious tracking rests on a handful of complementary indicators, not one score.
- Share of voice. How often your brand appears against named competitors on the same prompts. The central number, the only one that answers whether you are winning or losing here.
- Citation rate versus mention rate. Two different things that must never be merged, covered next.
- Prompt coverage. Across how many distinct questions, informational, comparative, transactional, you appear at least once. Breadth, not depth.
- Sentiment. The tone and context: a recommendation, a comparison, or a warning.
- Cited sources. Which domains and pages the engines lean on to answer about your topic, which tells you who holds authority.
- Trend over time. The direction over weeks and months, after a publication, a fix, or a PR push. A point measurement cannot show movement, and movement is the point.
Citation is not mention, and merging them fakes the number
When an AI talks about your brand, two different things hide behind the same appearance. A citation anchors a claim to an identified, clickable source: a numbered link back to a specific page of yours. A mention is your name in the text with no link, drawn from training knowledge or from third party sources the answer does not itself cite.
They measure different things. Citation measures perceived authority on a precise query and your ability to capture qualified traffic. Mention measures your presence in the general conversation, your background familiarity to the model. A brand can be heavily mentioned yet never cited, because its content is not fresh, structured or authoritative enough to be the chosen source. If a dashboard blends the two into one figure, and some tools do, the number gives an illusion of visibility that says nothing about your real pull. Ask any vendor which one a metric counts.
The distinction also sets priorities. Heavily mentioned but rarely cited is an editorial problem, freshness, structure, authority, the ground covered in our page on getting cited by ChatGPT. Neither mentioned nor cited is a deeper problem, thin presence on the platforms AI leans on and a weak entity.
How to tell a real number from a made up one
Four questions separate a measurement from a marketing figure.
- Is it repeated? A citation rate from one response has no statistical value. Over dozens of runs it becomes a trend. If a report does not say how often it ran each prompt, treat the number as decoration.
- Is it multi engine? ChatGPT, AI Overviews, Perplexity, Gemini and Claude use different indexes and different bars. Visibility in one guarantees nothing in another, so a single engine figure is partial and usually flattering, since people measure where they already appear.
- Does it separate citation from mention? If it does not, it is two different realities averaged into one misleading percentage.
- Is the prompt set representative? Too narrow biases the result, too broad dilutes the signal. It has to mirror how your buyers actually ask.
The signal analytics can confirm
Beyond citations in answers there is AI referral traffic: visits that land on your site when a user clicks a link inside ChatGPT, Perplexity or an AI Overview. Volume is low, since most AI use is zero click, but it corroborates from the analytics side what prompt measurement shows from the answer side. The two together confirm that measured visibility turns into at least some real traffic.
What this changes for your brand visibility
Measurement is not decoration on top of GEO, it is the thing that tells you whether any of the work moved. Ahrefs, across 863,000 keyword SERPs and 4 million cited URLs, found only 38 percent of the pages cited in Google AI Overviews sit in the top 10 organic results, so your existing rank tracking is not a proxy for this. Set a stable prompt set, track several engines in parallel, repeat the runs to smooth out variance, and read share of voice, citation rate, mention rate, sentiment and sources together rather than as one figure. What you do once you can see the gap is a separate discipline, in our page on the order to work in.
Sources
- GSPANN, The AI Shelf (2,240 queries, 112 companies): ChatGPT identifies a named brand 99.4 percent of the time but surfaces it on 3.32 percent of open discovery questions
- Ahrefs (863,000 keyword SERPs, 4 million cited URLs, data through January 2026): 38 percent of pages cited in Google AI Overviews sit in the top 10, and 31.0 percent are outside the top 100
- Forrester, 2026 Buyer Insights (survey of 18,000 buyers): 94 percent of B2B buyers used AI during their most recent purchase
To see what the main engines answer today on your real buying questions, across repeated runs rather than one lucky check, and who gets named in your place, request a free AI visibility audit.
Related reading