One firm, two very different answers
A buyer asks, “Is this firm a credible specialist?” and AI describes its strengths. The same buyer then asks, “Who is the authoritative source for this problem?” and AI credits a rival, the media or a republisher. The first answer reflects named Brand evaluation. The second tests unbranded Discovery and the Authority role received when discovered. One can be strong while the other is weak.
This is easy to miss because both questions belong to the same broad topic. But a question about market direction, one about method and one comparing advisers require different knowledge. They can draw on different sources and give the same firm different roles.
A brand can carry useful associations across those questions. It may be named often or evaluated positively when the buyer supplies its name. That does not create one transferable position. Delphic therefore separates four outcomes: whether the firm is discovered unprompted, the authority role it receives when found, how it is evaluated when named and how it stands against a specified rival.
The useful unit is a question family: one commercially meaningful buyer need expressed in several natural ways that resolve to the same substantive answer. Position is observed for that family on each engine and across repeated runs. A category-wide score can summarise those records. It cannot replace them without hiding where the firm is present, subordinate or absent.
What one result establishes
Question-level loss describes what happened; it does not yet explain why. Absence may reflect source availability, answer fit, public evidence, a stronger competing source or ordinary output variation. Those possibilities require diagnosis after the position has been established.
Repeated-measurement research has found material variation in source selection across runs within a 24-hour window. One prompt and one response are therefore too weak to establish a stable pattern. The source repeatedly receiving the credit can still be identified, but only for the measured family, engine and period.
The same granularity prevents the opposite mistake. Another source's strength on one question does not mean it has captured the whole category. The loss is bounded. So is the opportunity.
The loss is local, but it is real
A firm's search traffic can look healthy. Its brand can receive warm answers when named. Yet on the unbranded questions that create new demand, another source may be discovered first, credited as the authority and given the route to the buyer.
That is the commercial wound hidden by a brand-wide score. The firm is not invisible everywhere; it is missing where a particular piece of expertise should have earned attention. Delphic locates those losses question by question, so the firm can distinguish a real position from a comforting average.