Two surfaces, two measurements

The commercial reality of AI buying behaviour is fragmenting along two surfaces. On one, the consumer talks to a frontier-model assistant and the model responds with recommendations drawn from its training data and connected retrieval. On the other, an autonomous shopping agent acts on the consumer's behalf - searching, reading product pages, comparing, selecting. The two surfaces are real, both are measurable, and they do not always agree. This paper reports the first controlled cross-method study comparing the two. Three brands across three consumer categories - a mass-market mascara, a major international car rental brand, a leading global online travel platform - were measured under matched conditions on each method: AIVO Meridian's structured Buying Journey Probe (BJP) and the AIVO Agentic Shopping Journey (ASJ). The same persona, the same brief, the same constraints, the same category. Only the measurement methodology varied. Four findings. 1 Where the methods agree, they agree on details - same brand displaced, often the same named winners. That convergence is not noise. It is cross-method validation of a real underlying commercial outcome. Half of the cells in our test fell here. 2 Where the methods disagree, the disagreement is interpretable: chat-based measurement leans on training-data dominance and retrieval ranking; agentic measurement leans on what live web search actually surfaces. These are not errors - they are different commercial surfaces. 3 In one cell of our test, the same brand wins decisively on one method and loses decisively on the other, against the same brief. A brand acting on one measurement alone would draw the wrong strategic conclusion. 4 Four failure modes emerge across the combined data set. Two are visible to either method alone. The third - category recategorization - is most clearly visible when both methods are run together. The fourth - methodology-divergent verdicts - is by definition only visible when both methods are run.