A study published by Growth Memo, based on research from Citation Labs and Clickstream Solutions, tracked 48 participants through 147 major-purchase tasks — televisions, laptops, appliances, car insurance — comparing behaviour in classic Google search versus Google AI Mode. One number changes everything about how brands should think about their marketing funnel.
In classic search, 56% of participants built their own shortlist from multiple independent sources. They clicked, compared, triangulated. They consulted multiple pages and arrived at their own conclusions. In AI Mode, only 8 out of 147 tasks produced a genuinely self-built shortlist. And 64% of AI Mode participants clicked nothing at all. They read the AI's text, identified their finalists from what the AI presented, and stopped there.
The comparison process didn't shrink. For most participants, it didn't happen at all.
What this means for the marketing funnel
The traditional funnel — Awareness, Consideration, Evaluation, Conversion — was built on a model of active, self-directed consumer behaviour. Users discovered options, compared them across multiple sources, and formed their own judgement through a process that took time, required effort, and gave brands multiple opportunities to intervene.
That model describes what happens in classic search. It does not describe what happens in AI Mode. In AI Mode, the evaluation process has been compressed into a single moment — the AI's response. The user reads what the AI presents, accepts the framing, and moves toward a decision. The stages of the funnel have not been shortened. They have been collapsed into the AI's shortlist.
This is what the hourglass model captures more accurately than the traditional funnel. The top is wide — discovery happens across many surfaces. The middle is a narrow bottleneck — the AI shortlist, where selection occurs. The bottom widens again — conversion and retention, for the brands that made it through. The critical constraint is not at the top and not at the bottom. It is in the middle, at the moment of AI-driven selection.
The AI Shortlist is the set of options an AI presents to the user as viable choices in response to a purchase-intent query. If a brand is not in the shortlist, it does not exist in that decision.
The data on what drives shortlist inclusion
The Growth Memo study identifies the two dominant trust drivers in AI Mode. AI framing — the specific way the AI describes a product or brand — drove inclusion for 37% of participants. Brand recognition drove inclusion for 34%. They run nearly even.
The implication is structural. Where users arrived with existing brand preferences, recognition determined which brands the AI highlighted. Where they did not, the AI's own framing filled the gap — and that framing determined which brands were presented as serious options. In both cases, the brand's presence in the AI's training data, its consistency across authoritative sources, and the clarity of its value proposition in machine-readable formats determined whether it appeared at all.
The study also documented a pattern with direct implications for how brands structure their information. Where AI Mode showed explicit, retailer-confirmed prices, 85% of participants understood pricing clearly. Where it did not — in insurance and laptops — confusion and overconfidence filled the gap. Users made elimination decisions based on numbers that may not have applied to them. The AI formatted what it had available. Brands that had not made their pricing structured and machine-readable were simply not surfaced clearly.
The three layers of the hourglass
Understanding the hourglass model requires being precise about what happens at each layer — and what the relevant metric is at each stage.
Top layer — Discovery
The goal at the discovery layer is not traffic. It is Citation Presence — the degree to which a brand appears across the AI-accessible environments where pre-training data originates. This means platform distribution: YouTube for transcript-extractable video content, Reddit for community-level authority signals, LinkedIn for professional context, PR and external mentions for cross-source entity consistency. A brand that exists only on its own website is not discoverable by AI in any meaningful sense.
Middle layer — AI Shortlist (critical bottleneck)
This is where the competition happens. The Shortlist Rate — the percentage of relevant queries in which the brand appears as a presented option — is the metric that determines commercial relevance in an AI-mediated environment. It is not ranking. It is not impressions. It is selection.
The drivers of Shortlist Rate are specific:
- Citability — can the AI extract and trust your content? Is it structured, consistent, and machine-readable?
- Clarity — is your value proposition understandable without ambiguity? Does the AI know what you do, for whom, and why it matters?
- Structured information — pricing, features, comparisons, FAQs. The AI presents what it can extract. What it cannot extract, it omits.
- Authority signals — consistent mentions across authoritative external sources. Third-party validation that the brand is a serious option in its category.
The Citability Score measures readiness for this layer across 56 parameters. Most brands score between 30 and 45 before optimisation — well below the threshold for reliable shortlist inclusion. After a full Citation Rate intervention, scores typically reach 65–85.
Bottom layer — Conversion
The Growth Memo research confirms something that changes how conversion should be understood. Of the 117 participants who adopted the AI's shortlist directly, roughly 85% showed no internal verification behaviour at all. They did not click out to check the brand's own website. They did not read reviews. They accepted the AI's framing and moved toward a decision based on what they had already read in the AI's response.
Conversion is no longer where persuasion starts. It is where persuasion ends — after the AI has already done the evaluative work. The brand that was accurately and compellingly described in the shortlist has already won the consideration battle before the user clicks anything. The brand's website, at that point, is confirmation — not discovery.
The strategic shift in one sentence
The old model was: Rank → Click → Convert.
The new model is: Be accessible → Be cited → Be shortlisted → Be chosen.
These are not the same optimisation problem. The first is an SEO problem. The second is a citability problem — and it requires a different measurement framework, a different content strategy, and a different understanding of where the brand's marketing budget should be directed.
Shortlist Optimization as a discipline
The emerging discipline is Shortlist Optimization (SLO) — a sub-layer of Search Optimization that focuses specifically on being selected by AI systems, not just discovered by them. The distinction matters because discovery and selection require different inputs.
Discovery requires presence across accessible surfaces. Selection requires structured clarity — the kind of content that an AI can extract, interpret, and reproduce accurately in a shortlist response. A brand can be present on YouTube, Reddit, and LinkedIn and still fail at the shortlist layer if its content is ambiguous, its pricing is opaque, or its value proposition is described differently across different sources.
"The battle is no longer for attention. It is for inclusion in AI-generated decisions."
The brands that adapt now will build citability infrastructure while most of their competitors are still optimising for click-through rates on queries that an increasing share of users never see. The window is not infinite. As AI Mode adoption accelerates — and the Growth Memo data suggests it is accelerating fast — the shortlist positions in any given category will be taken by brands that established their citability early. Latecomers will face a harder problem: displacing incumbents who have already been trained into the AI's understanding of the category.
Position one in the AI response, as one participant in the study said about their laptop decision, "was the entire decision." Being that position one is not a matter of bidding. It is a matter of citability. And the measurement infrastructure to track and improve it already exists.
Source: Growth Memo — How consumers navigate high-stakes purchases in AI Mode. Research by Citation Labs and Clickstream Solutions, 48 U.S. participants, 147 shortlisting tasks across televisions, laptops, appliances, and car insurance.