How discovery worked before agents: browse many, compare some, buy one
For roughly 25 years, the consumer discovery model was a three-stage process. A buyer had a need. They entered a search query. They received a results page with ten or more options, scrolled through several, clicked on two or three, compared them on their own, and eventually bought. Every step of that evaluation was human-directed.
The implication for brands was simple, if expensive: presence meant possibility. Getting onto the first page of Google results, or into the first row of a category page on a marketplace, meant a buyer might see you. They might not choose you, but visibility bought a shot. The entire discipline of search engine optimisation (SEO) was built on this model. So was paid search, display retargeting, comparison shopping engines, and affiliate review sites. They all competed for the human's attention at the browse-and-compare stage.
The buyer's consideration set was whatever they personally constructed. Ten tabs open, three price comparisons, two YouTube review videos. The funnel was wide at the top and narrowed slowly as the buyer did the work of narrowing it.
| Stage | Classic browse model | Agent-mediated shortlist model |
|---|---|---|
| Who does the evaluation? | The buyer, across multiple tabs and visits | The AI agent, before the buyer sees any results |
| How many options enter evaluation? | 10-20 per search results page; buyer controls depth | Hundreds of indexed sources; agent controls all of it |
| What does the buyer receive? | An open list to browse and compare | A closed shortlist of 3-5 options |
| What is the buyer's primary task? | Discovery and comparison | Verification of the shortlist they received |
| What signals influence rank? | Keywords, page authority, ad spend, structured data | Structured data accuracy, entity clarity, review corpus, citations |
| Can you buy visibility? | Yes, via paid search and shopping ads | Not directly; shortlist position is determined by agent trust, not bid |
That model is intact for a large share of buyer journeys today. Google search still processes billions of queries daily. But a growing segment of buyers, the ones most comfortable with AI tools, now start differently. They describe a need to an agent and receive a verdict. The browse stage never happens.
The funnel reframe: agent evaluates hundreds, buyer verifies three
Roger Dunn named the pattern in Microsoft Advertising's May 2026 post on agentic commerce: "This is what I've been calling the Shortlist Economy. When someone asks ChatGPT, Copilot, or Gemini for a product recommendation, the AI assembles a shortlist, usually three to five options, and that shortlist becomes the only consideration set that matters. If you're not on it, you don't get to make your case."
The diagram below shows the structural difference. On the left side, hundreds of candidate brands flow into the agent's evaluation. The agent processes structured data, review signals, entity information, and citation sources. It then outputs a shortlist. The buyer receives three to five options. Their task is to verify, not to discover.
The critical observation is that consumers do not abandon traditional channels after receiving a shortlist. Dunn's analysis notes that the vast majority of buyers still verify AI recommendations before purchasing. They take that shortlist to Google, to Bing, to a brand website, to YouTube. But the verification search is a different kind of search. They are searching for the specific brands the agent named, not starting a category search from scratch. Being on the shortlist determines which brands benefit from that verification traffic.
For brands in Singapore, Australia, the United States, Canada, and Malaysia, the segment of buyers who start with an AI agent rather than a search engine is growing, faster in some categories than others. High-consideration purchases with a long research phase, including financial services, insurance, software, and professional services, are where agent-mediated discovery is appearing first, because those buyers have the most to gain from an agent doing the comparison work for them.
What brand means when the first audience is a model
Brand, in the classic sense, is an emotional and associative construct. Decades of advertising have made certain names feel safe, aspirational, or familiar. That feeling is a human feeling. AI models do not have feelings about brands. They have data about brands.
Dunn identifies two distinct trust layers that now operate in sequence. Machine trust comes first: can an AI agent find the brand, understand what it sells, and have confidence that its data is accurate and current? That is about structured data, reviews, entity consistency, and citation signals. It is operational, not emotional. Human trust comes second: when the buyer arrives to verify the shortlist, does the brand have the credibility and experience to close the deal?
The mistake operators make is assuming that strong human trust automatically produces machine trust. It does not. A brand with 30 years of history, strong net promoter scores, and a large paid media budget can still be invisible in AI-generated shortlists if its entity data is inconsistent across the web, its schema markup is absent, or its content does not answer the specific questions buyers are actually asking agents.
For service businesses like leapbuzz clients in banking, insurance, and fintech, the equivalent of product attribute data is entity clarity: a well-defined, consistently expressed description of what the firm does, for whom, in which markets, and with what track record, backed by citations from credible third-party sources. An agent evaluating "AI marketing consultancy Singapore" will prefer the firm whose entity description is consistent, specific, and externally cited over the one with a beautiful website and vague positioning.
This is where leapbuzz's AI marketing strategy work intersects directly with the shortlist problem. Positioning clarity is not just a brand exercise anymore. It is a machine-readability problem.
The four signals that determine shortlist inclusion
Dunn's analysis identifies a ranked set of signals that AI agents use to decide which options make the shortlist. The ranking is based on observed agent behaviour, not declared algorithm documentation, so treat it as a working model rather than a published specification.
- Product or service truth as machine-readable data. AI agents reason over structured attributes: dimensions, compatibility, features, use cases, service scope, target audience. If those attributes do not exist as machine-readable data, the brand is not a candidate. In Dunn's words: "In the old world, poor data meant lower conversion. In the agentic world, poor data means you never enter consideration." For service firms, this translates to structured schema markup on every page, consistent service descriptions, and FAQ content that directly answers the questions buyers are asking agents.
- Review corpus quality, not quantity. AI systems synthesise review sentiment to answer specific questions. Dunn notes that one detailed review explaining how a product performed in a real scenario is worth dozens of generic five-star ratings. Third-party endorsements, expert mentions, and certifications act as trust multipliers that agents increasingly weight. For B2B service firms, this means published case studies, industry citations, and named client references where permitted.
- Fulfillment reliability and consistency. As agentic commerce matures, delivery reliability becomes a scoring input. For service firms, the equivalent is delivery consistency: do you do what you say, on the timescale you commit to? Public-facing evidence of delivery quality, including testimonials and structured results pages, signals this to agents.
- Brand authority as a verifiable digital identity. Traditional brand equity still matters, but its mechanism is shifting from emotional halo to verifiable digital identity. An organisation with a clear founding date, named leadership, verified social profiles, and consistent entity data across platforms is more legible to an agent than one without those signals, regardless of media spend.
| Signal category | For product businesses | For service businesses |
|---|---|---|
| Machine-readable data | Schema.org Product markup, merchant feed accuracy, attribute completeness | Schema.org Service markup, FAQ schema, Organization entity data, JSON-LD graph completeness |
| Review corpus | Detailed use-case reviews on Google, platform marketplaces, and independent review sites | Published case studies, named testimonials, industry citations, results pages |
| Fulfillment reliability | Delivery track record, returns rate, inventory accuracy | Delivery consistency, publicly evidenced outcomes, client retention signals |
| Brand authority | Third-party endorsements, certifications, press mentions | Named leadership, verified entity profiles, external citations from credible sources |
The signals above are exactly what generative engine optimisation (GEO) targets. The companion post on GEO fundamentals covers the implementation detail. Microsoft's Clarity Citations Reporting (released 17 June 2026) is now the first free measurement tool that shows your share of authority in AI-generated answers, giving operators a scoreboard for shortlist performance. The post on Clarity AI citation tracking covers that measurement layer in detail. The underlying academic framing on citation-based content signals comes from Princeton's GEO research (Aggarwal et al., 2023, arXiv:2311.09735), which found that adding verifiable statistics and citations lifts AI citation rates by approximately 37 percent. Google's own documentation on AI Overviews eligibility (published at blog.google) identifies authoritative, well-structured content as the primary factor in AI-generated answer inclusion.
The operator to-do list: structural preparation, not campaign tactics
The shortlist economy is not a campaign problem. Running more paid search or increasing social ad spend does not improve shortlist inclusion, because agents do not consume ad placements. The preparation is structural.
Dunn's recommendation for product businesses applies directly to service firms: product data enrichment and agent readiness is the highest-leverage investment available right now. "Unglamorous, I know. But it's the highest-leverage investment available right now. In the age of AI recommendations, your product data is your shelf placement." For service firms, entity data is the equivalent of product data.
Dunn explicitly flags the first-mover dynamic: "the competitive advantage in AI visibility is forming now, and the window is measured in quarters." Brands that complete the structural work in Q3 and Q4 2026 will have a compounding advantage over those that wait until AI-mediated discovery is the obvious majority channel.
Shortlist readiness checklist (expand to review)
Eight structural items for service-firm operators entering the shortlist economy. Work through these before investing in incremental campaign spend.
- Complete JSON-LD schema graph on every public page. Organization node with named leadership, full address, and contact data. Service nodes with descriptive scope, target audience, and area served. FAQPage nodes that directly answer buyer questions in complete sentences. Every page must parse without errors.
- Entity consistency across the web. Your business name, address, phone number, service description, and founding date must match across Google Business Profile, LinkedIn, Crunchbase, and any industry directory listings. Agents cross-reference these. Discrepancies reduce trust scores.
- FAQ content written in agent-readable chunks. Each FAQ answer should read as a complete, self-contained response to a specific buyer question. Answers that assume the reader has read the surrounding article do not extract well. Aim for 80-150 words per answer, written in plain English, with the direct answer in the first sentence.
- External citations from credible sources. Academic citations, platform vendor documentation, regulator references, and named industry research cited in your content give agents confidence in your accuracy. Weasel citations ("industry reports suggest") reduce that confidence. Name the source.
- Results and proof structured as accessible content. Case studies, outcome descriptions, and anonymised client results need to be on your site, structured as readable content, not gated behind a contact form. Agents cannot read behind gates. If your proof lives only in sales decks, it is invisible to agents.
- Named, verifiable leadership. A Person schema node for the founder or named practitioners, with LinkedIn profile, credentials, and role history, is a strong entity authority signal. Anonymous firms are harder for agents to trust-score.
- Measure your current shortlist inclusion rate. Use Microsoft Clarity's Citations Reporting (launched 17 June 2026) to see your current share of authority in AI-generated answers. Run a structured set of buyer queries across ChatGPT, Perplexity, and Google AI Overviews and record which shortlists include you today. That is your baseline.
- Do not cut brand investment. Machine trust gets you onto the shortlist. Human trust closes the deal when the buyer arrives to verify. Buyers still do verification searches, and the brand the buyer already trusts gets the click. Both layers require investment.
The agentic marketing operations post covers how this structural work fits into the broader shift in how marketing teams operate inside an agent-mediated environment. The agentic marketing ops post covers the three layers of agentic marketing and the escalation-boundary design that keeps human judgement where it must stay.
leapbuzz's AI visibility optimisation service addresses the structural shortlist preparation directly: schema audit, entity consistency check, FAQ architecture, citation sourcing, and measurement baseline against the Microsoft Clarity Citations scoreboard. If you are running paid media into markets where AI-mediated discovery is already a meaningful buyer segment, the structural work is a prerequisite, not a nice-to-have.
