The following themes emerged from think-aloud transcripts collected during free exploration (Tasks 3 through 8) and open-ended follow-up probes (Tasks 10, 11, 13, 15, and 16). Themes are ordered by frequency and consistency of expression across participants. Finding 5 is worth reading as the through-line: it is the one that most directly addresses whether and how an AI module can earn user trust in this context.
Finding 1 — Vehicle-specific facts belong at the top
The most consistently expressed preference was for factual, vehicle-specific information (condition, mileage, price, key specs) to appear early in the page, before any promotional or AI-generated summary copy. Participants described a mental model that proceeds from objective facts toward subjective impressions, not the reverse. Options that led with promotional copy, particularly A and, in some readings, C, were described as "getting in the way," "like an ad," or requiring unnecessary scrolling before reaching needed information.
"We're talking about a used car, so the condition of the used car — information related to that — is always paramount and should be at the top. Then you can have your generalized information about the model."
Desktop participant, Option B preference
Finding 2 — Used-car context heightens scrutiny of condition information
Participants consistently framed their evaluation through the lens of used-car purchase risk. Condition status, mileage, Carfax/inspection records, and vehicle preparation details carried significantly more evaluative weight than they would on a new-car listing. Options that buried or omitted condition information, particularly D, lost credibility on this dimension, creating a notable tension: D was preferred for brevity and hierarchy, but some felt it was incomplete without condition detail. Condition information should be considered table stakes for the AI merchandising block.
"If it's not listing anything about the specific condition of the used car, I am not going to consider it. AI can and does hallucinate. For things like this — real, important information — just hire somebody. You need accuracy."
Mobile participant, Option B preference
Finding 3 — Enumerated features beat paragraph summaries
Bulleted or categorized feature lists (as in Option D's "Features" section) were widely preferred over the prose-heavy "Why You'll Love This" format in Options A and C. Participants described scanning feature lists rather than reading them; prose summaries were harder to skim and perceived as more marketing-oriented. The label "Why You'll Love This" was specifically flagged by several participants as too sales-forward. Some noted they treat copy written in that register the same way they treat advertising and skip it by default. "Highlights" or unlabeled feature grids were more neutral and trusted.
"When we see it laid out like [a promotional block], our brain might automatically discount it, throw it away as an advertisement and not even realize it's related to this specific car."
Desktop participant, Option B preference
Finding 4 — Brevity vs. comprehensiveness is a platform-mediated tension
Desktop users were more tolerant of comprehensive content and in some cases actively valued it. Option C's length was seen as justified by information density on desktop. Mobile users showed much lower tolerance for scrolling past content to find facts, which explains Option D's dominant mobile performance. This suggests a responsive or adaptive content strategy may be warranted: desktop can accommodate more depth in the AI merchandising block, while mobile benefits from aggressive content prioritization.
"Option D was just straight to the point. It wasn't overwhelming. I liked the bullet points, I liked the order. At this point, if I'm going to go in to test drive it, I don't need too much information."
Mobile participant, Option D preference
Finding 5 — AI-generated content is conditionally accepted, with a clear scope boundary
When participants learned the page content was AI-generated, the majority responded neutrally or with mild acceptance. Importantly, this acceptance was not passive indifference: it was conditional on what the AI was understood to be doing. Participants were comfortable with AI doing aggregation and synthesis work, pulling together information that already exists into a readable summary. They drew a sharp line at factual claims. Condition status, specifications, and vehicle history were domains where several participants explicitly said AI should not be the source, citing the risk of hallucination and the stakes of the purchase.
This finding points toward a meaningful design constraint for the module: AI-generated copy works best when it is clearly doing descriptive and summary work, not when it is presented as a source of ground-truth vehicle facts. The discomfort with "Why You'll Love This" is related to this: it positions AI as making a personal recommendation, which feels like a different and less trusted register than summarizing what a vehicle has. The practical implication is that the AI module should be scoped to synthesis and description, with structured data handling all factual claims.
"I don't think it would really make a difference to me. When used ethically and in the right manner, AI really serves its purpose well — getting a lot of information in one place."
Desktop participant, Option D preference
Finding 6 — Price visibility confusion (adjacent issue)
Multiple participants across both conditions expressed confusion about the relationship between the price displayed at the top of the page and a price breakdown shown lower on the page. Several noted that two different price figures appeared on the same VDP, with one appearing to reflect a new-car MSRP and one a used-car AutoNation price, which created significant confusion. This finding is outside the direct scope of the AI merchandising block study but surfaced consistently enough to warrant flagging as a separate issue for the VDP product team.
