A Better Dashboard for an AI-Mediated Journey
The subject of this series is the changing relationship between AI-mediated search, consumer product discovery, and the traditional attribution systems used to interpret them. The central question is now clear: what can a conventional funnel see when an AI Overview captures the first attention on a search page and an assistant forms a shortlist before a shopper reaches a product page? The answer is not that the journey has vanished, nor that every AI-mediated journey is faster. It is that the visible boundary has moved. A defensible dashboard must detect that movement before it reallocates credit, budget, or confidence.
One visibility gap, two observations
The two studies illuminate different sides of the same problem. The eye-tracking study observes attention inside a search interface. When an AI Overview appears above traditional ranked results, it receives substantially more fixation and saccade time than the ranked-results region. Yet the familiar top-left or golden-triangle pattern remains, and the first ranked link still attracts the greatest attention and click activity within the traditional list. The generated layer changes the first encounter without making the rest of the page irrelevant.
The passive-log study observes what remains visible around a consumer journey. Its most stable result is a change in the pre-intent channel mix among journeys with observed LLM exposure: visible LLM share rises, and visible review or UGC share also rises after adjustment. The evidence does not show a corresponding, robust acceleration of the full journey. Count-based compression, time compression, and time-to-intent differences weaken or become imprecise after adjustment, and the within-user comparison shows higher visible LLM share without higher compression.
These observations are related but not interchangeable. More gaze on an AI Overview does not directly measure a change in product attribution. A changed pre-intent channel mix does not tell us how long a person looked at a summary or whether they trusted it. The first study establishes an interface-level attention shift. The second establishes a journey-level trace shift. Together, they support a visibility model: evaluation can become more concentrated inside a generated layer while the external path becomes less representative of the work that preceded its visible pages.
That model changes what a dashboard should treat as a performance event. A traditional result may receive less first-glance attention but remain important as a source of detail or verification. A product page may receive a direct visit after an assistant has narrowed the choice set, making it consequential without making it the origin of discovery. A review page may appear more or less often in the visible trace without revealing whether its ideas were summarized, compared, or replaced inside the conversation.
The correct unit of interpretation is therefore not the isolated event. It is the relationship among interface exposure, visible sequence, and the decision the analyst wants to make. If the decision is whether the measurement environment has changed, an observed LLM touchpoint and a rising pre-intent LLM share are meaningful signals. If the decision is whether consumers are deciding faster, the same signals are insufficient without evidence of the unobserved comparison and effort inside the interface.
When a shorter path should trigger an alarm
A sudden increase in direct product-page visits can look like a success story. It may suggest that shoppers are skipping low-value browsing, moving with greater intent, and reaching commercially useful pages more efficiently. But the same pattern can occur when shortlist formation and comparison have moved into an assistant. The first visible product page is then the remainder of the journey, not necessarily its beginning.
The same ambiguity applies to a shorter apparent time-to-intent. A dashboard begins its clock at the first observable consideration event.