AI visibility score
A composite number some tools compute to summarise how often a brand appears in AI answers — what it is made of, why the components are not comparable, and why Attensira does not compute one.
Karl-Gustav Kallasmaa, Founder & CEOLast updated An AI visibility score is a composite number that some monitoring tools compute to summarise how often a brand appears in AI assistants' answers. It is a vendor construct, not a standard: no specification defines it, and no AI platform publishes one.
Attensira does not compute an AI visibility score, and does not measure sentiment. Its published limitations say so directly: "visibility" appears as a label on a section of the product and as a value in an enum, and is not a computed metric; nothing in the system scores whether an answer spoke about a brand kindly. This page therefore defines the industry term and then says what is measured in its place.
What a score is usually made of
Almost every implementation is a weighted blend of some subset of:
- Mention frequency — how often the brand name appears in answer text.
- Citation frequency — how often a URL on the brand's domain appears in a model's citation list.
- Citation position — how high in that list the URL sits.
- Breadth — how many prompts, models or countries produce any appearance at all.
- Sentiment — some vendors add a tone judgement, usually a model grading another model's output.
Each component is defensible on its own. The composite is where the trouble starts, because the weights are unpublished, arbitrary, and the only thing standing between raw counts and a headline number.
Why the components do not combine cleanly
They have different denominators. A mention rate is over successful runs. A citation position is over the occasions a domain was cited at all. Blending a rate with an index produces a number whose units are nothing.
The inputs are not deterministic. The same prompt to the same model can return different answers minutes apart. A score computed from a single draw per prompt is one sample presented as a measurement, and week-to-week movement in it is mostly noise. Attensira's response to this is sampling depth plus a significance test: a delta that does not clear a two-proportion z-test at the conventional significance threshold is reported as no proven change rather than as a movement — which means the change could not be proven, never that it was zero.
Detection carries systematic error. Attensira's limitations page is explicit that a mention is a case-insensitive substring match of brand name or domain against answer text, with no entity resolution and no word-boundary check. A brand called Arc matches "march" and "search". Averaging does not remove that error, because it is systematic rather than random — and a composite score removes the reader's ability to see it at all.
Nothing downstream is attributable. No product in this category can join an answer a model gave someone to a visit or a signup later. Presenting presence in answers as a single score invites the reading that it is a performance number. It is not; it is a presence number. See zero-click search.
What Attensira reports instead
Two measurements, each with its meaning stated:
ShareOfVoice is the brand's own mention rate: successful runs in a window whose answer named the brand, over all successful runs in that window. Despite the name, it is not a share of a fixed total. Competitor rates use the same denominator, so one answer naming a brand and three competitors raises four independent rates, and adding them produces a total that is not bounded by the whole and means nothing.
AvgPosition is the mean index of a URL within a model's own citation list, computed per source domain. It answers "when this domain is cited, how far down the list does it sit". It is not a leaderboard position, not a brand ranking, and not comparable to a search result position.
Both are reported with their sample size, and both distinguish not measured from measured zero — a model that was never queried is not a model that rejected you.
How to read a vendor's score
If a tool gives you one, ask four questions before acting on it: which prompts, how many draws per prompt, which denominator, and what happens to the number when a model is untracked. A score that cannot answer those is a chart, not a measurement.
The underlying discipline is unchanged by the score's existence. Google's AI features guidance states there are no additional requirements to appear in AI Overviews or AI Mode and no special optimisations necessary — eligibility runs through ordinary indexing and snippet eligibility. What is left to do is what GEO has always been: be fetchable by the named agents, write passages that survive extraction, and check what is actually being said about you. See AI search for which assistants do the fetching.
Frequently asked questions
Is it a standard metric?
No. Every vendor defines its own, with unpublished weights; two vendors' scores are not comparable.
Does Attensira compute one?
No — and it has no sentiment metric either. It reports mention rates and per-domain citation positions.
Why is a composite number risky?
It hides sample size and detection noise, and blends quantities with different denominators.
Do the AI platforms publish one?
No. Google reports AI-feature traffic inside overall Search Console search data, not as a separate visibility figure.
Terms related to AI visibility score
The practice of getting a source reproduced inside an AI-generated answer, introduced as a named paradigm in a 2023 research paper.
Search where a model composes the answer and fetches pages through named crawlers, rather than returning a ranked list of links for you to read.
How an AI answer attributes what it says to the pages it read, and why a citation is a distinct outcome from a click or a mention.
A search that ends without the reader visiting any website, and the measured gap between sessions that show an AI summary and those that do not.
Google's AI-generated summary at the top of a results page, and the snippet controls that decide whether your page can appear inside one.