Win-Rate℠ and the Linkage Gap: Why Being Cited Is Not Being Chosen
Being mentioned by an AI assistant and being recommended by one are not the same event. The gap between those two moments is where revenue actually moves, and closing it has become the central problem of brand measurement in an AI mediated market.
Across more than 20,000 multi-turn buying journeys covering over 200 brands, AIVO found that most brands models know well are not the brands ultimately recommended when a buying decision is made. The displacement rate sits at 87.3 percent, documented in a DOI-anchored working paper on AIVO Standard. We call this the Linkage Gap, and Win-Rate℠ is the measure of how often a brand survives it.
Win-Rate℠ measures whether a brand is carried through to the final recommendation rather than merely mentioned during the conversation. It is distinct from ARQ, AIVO's composite headline metric, which weighs recommendation success alongside resilience under challenge and head to head performance against competitors. Win-Rate℠ answers a narrower question, and it is often the one with the greatest commercial consequence. At the moment a model has to choose, does this brand survive.
Where the gap actually opens
For most categories that question has a practical, uncomfortable answer. Brands can appear in the large majority of probe turns and still lose on Win-Rate℠ if they are not the brand a model actually names when pressed. This usually happens at the comparison turn, the moment a model stops describing a category and starts constructing a justification for choosing one named option over another. Brands with strong product documentation and weak comparative material are the ones most exposed here, because a model has plenty to say about what the brand is and very little it can assemble into a justification for choosing it.
A win at this stage can also be more fragile than it looks. Turn by turn tracking of what kind of evidence still supports a brand's position, its own marketing pages, third party analyst material, or independent commentary, often shows that a brand's apparent win at the decision turn is resting entirely on material the brand publishes about itself, with every independent source having already dropped away earlier in the conversation. That recommendation has no independent support left, and it is exposed the moment a competitor arrives at that turn with evidence the model can actually use.
Two categories of proof
Two patterns from the corpus illustrate why the gap persists once it opens. One brand in the corpus carries an extreme Linkage Gap, cited constantly across probe turns and chosen almost never, the clearest documented case of being known without being wanted. A different brand shows the inverse pattern, strong on AI-RQ, AIVO's reputation instrument, but weak on Win-Rate℠, a brand a model understands well and still fails to recommend. One loses because it lacks comparative evidence despite high visibility. The other loses because its reputation profile fails to convert into recommendation. The two patterns require different remediation and would look identical on a metric that only tracked presence.
The most persistent illustration in the corpus is a defunct brand still being recommended above live competitors months after it stopped trading. It is a useful corrective to the assumption that AI recommendation tracks the real world closely. Models recommend the argument they can reconstruct most convincingly, not necessarily the market as it exists today.
From measurement to remediation
None of this is static. Audits are re-probed on a fixed cadence against a frozen prompt cohort, platform set, and scoring methodology, so a later measurement can be attributed to work actually done rather than to the underlying models shifting in the background. Remediation itself follows a consistent order once a Linkage Gap is diagnosed. Structural readiness, whether a model can reliably read the brand at all, comes first, because nothing else can be attributed until that is solved. Then the specific turn where displacement is occurring, most often the comparison or criteria turn, where independent evidence is what is missing. Then the weakest individual platform, where movement will be most visible against a flat baseline.
Closing a Linkage Gap is not about producing more content. It is about producing the comparative evidence a model needs at the exact moment it has to justify one brand over another. That is a different brief than most search or content teams are currently resourced for, and it is the brief that determines whether a brand's Win-Rate℠ moves.
What it costs to leave open
In categories where AI assisted research already shapes a meaningful share of buying decisions, a low Win-Rate℠ stops being a measurement question and becomes a commercial planning question. AIVO models that exposure directly as Revenue at Risk, built from published revenue, a discovery share, the measured Win-Rate℠ gap, an estimated AI influenced share of the category, and a conservatism dampener applied throughout to avoid overstating exposure. The figure is explicitly modeled rather than measured, and it sharpens the moment a brand supplies its own segment level data. The underlying measured findings do not depend on it either way.
The market is beginning to standardize adjacent measurement problems. AIVO's AI Traffic Attribution Convention, currently under review by the Media Rating Council and IAB Tech Lab, addresses attribution rather than recommendation, but both reflect the same underlying shift: AI mediated commerce requires measurement infrastructure designed specifically for AI rather than inherited from search.
AI visibility gets a brand into the conversation.
Win-Rate℠ determines whether it leaves with the customer.
Comments ()