Four signals, four meanings
Four observations answer four different questions about the same AI response:
- Name presence: Is the firm named at all? Presence can be positive, neutral, subordinate or negative. It proves only that the name appeared.
- Attribution or citation: Is a claim or passage connected to the firm's material? This establishes source use, not automatic endorsement of the firm as the best answer.
- Source placement: Does the firm appear among the sources through which a buyer can discover and verify the answer? Placement, prominence and the role of the source matter alongside the existence of a link.
- Semantic position: How does the answer frame the firm - as the authority, a supporting source, one option, a weaker alternative or the subject of criticism?
The useful unit is therefore not a mention or citation in isolation. It is the firm's observed role on one buyer question, on one engine, across repeated responses.
A link is not a verdict
A citation can support one factual sentence while the answer recommends somebody else. A firm can be recommended without a visible citation. A name can appear only because the question named it. None of those observations, alone, establishes the firm's complete AI position.
Delphic's four-outcome measurement model uses these signals differently when measuring unbranded discovery, the authority role received when discovered, named-brand evaluation and direct competitive standing. This framework establishes the prior rule: do not infer commercial authority from the presence of a name or link.
The public model explains the distinctions. Applying them consistently is a separate task. Delphic keeps private the detailed definitions and rules used to classify difficult or borderline cases consistently across engines and runs.
What a single score erases
A single visibility number can hide commercially opposite results:
- a link may support a statistic while a rival receives the recommendation;
- a prompted brand mention may be reported as if the firm were discovered unprompted;
- frequent appearances may all place the firm in a subordinate role;
- criticism may increase a mention count while weakening the position that count appears to celebrate.
The arithmetic can be accurate while the conclusion is wrong. Preserving the four signals makes the result interpretable before it becomes a dashboard score or a claim of progress.
Before trusting a visibility report, ask four questions:
- Does it separate positive, neutral, subordinate and negative appearances?
- Which claim does each citation support, and what role does the answer give the firm?
- Are unbranded discovery, named evaluation and direct comparison kept separate?
- Which commercially different outcomes disappear inside the headline score?
These questions test whether the result preserves the public distinctions. They do not disclose the detailed rules used to resolve borderline cases.
Bad categories reward the wrong work
A weak observation model can reward the wrong work. It can celebrate a link that gives a rival the recommendation, treat a prompted brand mention as unbranded discovery or mistake frequent but subordinate appearances for authority.
A precise model makes the evidence usable. It shows what AI credited the firm for, what role the firm occupied and which different question requires measurement next.