In its established form, attribution maps observable touchpoints — clicks, sessions, campaign exposures — to conversions and revenue across a discoverable journey. Last-click, first-touch, and multi-touch models all assume the relevant contacts can be observed and linked to the same path.
In AI-mediated journeys the click chain breaks early: a buyer often forms a view inside an LLM answer where there is no click and no referrer. Rhinegold therefore extends classical attribution with observable language-system signals — mentions, citations, recommendations — and separates three levels that are routinely blurred. Attribution asks which contacts are credited with an outcome. Contribution analysis asks which factors plausibly contributed. Incrementality asks what additional effect would not have occurred without the intervention. Where strict point-attribution is not identifiable, the analytical focus moves to contribution and, where the data permits, incrementality estimation under counterfactual reasoning1.
Attribution becomes useful the moment the question shifts from "which channels drive clicks?" to "which language-system signals correlate with closed-won business?" It is the bridge that keeps Semantic Intelligence from collapsing into vanity metrics.
Last-click and first-touch models fail in AI-mediated journeys because they assume a discoverable click chain that no longer exists2. Multi-touch attribution overstates what is observable when relevant touchpoints happen inside an LLM and cannot be linked to the journey. Marketing-mix modeling has limited visibility into specific LLM-mediated touchpoints unless suitable proxy variables are modeled. The honest position is that strict single-touchpoint attribution is rarely identifiable here; what works is contributory attribution under counterfactual reasoning. Note that difference-in-differences is a method for estimating incremental effect, not an attribution model1 — it belongs in a different column.
The measurement gap is real, but it is not total. What remains observable: citations and mentions across controlled prompt sets; branded-search movement; referral and direct-traffic patterns; CRM-linked lead quality; SQL and closed-won development; intervention timing; and matched peer or control clusters. Professional steering reasons from these rather than from a click chain that no longer survives the journey.
Against last-click attribution, which assigns full credit to the closing channel. Against marketing-mix modeling, which is channel-level and econometric, with limited resolution for the LLM layer. Against Citation Rate and Mention Rate, which are inputs to attribution, not attribution itself. And against difference-in-differences, which is an incrementality method, not an attribution model.
There is no industry standard for attributing economic outcomes to LLM-mediated touchpoints. This is methodologically open ground, and a 2026/27 focus area for rhinegold.