From Ask-a-Friend to Ask-an-Engine: Mapping the Local Discovery Funnel Across Nine Verticals

One funnel, four stages

Strip away the vertical detail and local buying looks the same everywhere. A person becomes aware they need a provider, narrows the field to a short list of candidates, verifies that a candidate can be trusted, and books. Awareness, shortlist, verification, booking. The stages have not changed. What changed is who runs each one.

In the older pattern, a trusted human ran the first two stages for you. A friend named a plumber, so awareness and shortlist collapsed into a single recommendation, and you spent your effort on verification and booking. Today the awareness and shortlist stages are increasingly run by a system that has already read what other people think and returns a synthesized version of it: a ranked list, a three-result map pack, a star-rating aggregate, or a generated answer that names a few businesses inside it.

This is why the shift matters operationally rather than philosophically. If the engine builds the shortlist, then being absent from the engine's answer is not the same as being the fourth-best option. It is being left out of the consideration set before the buyer applies any judgment of their own. The funnel still exists, but the top of it has been re-platformed onto software, and the evidence below traces that re-platforming stage by stage and vertical by vertical.

Stage one, awareness: the engine replaced the friend

The clearest evidence that the first step is now a search comes from the vertical with the longest-running, best-instrumented buyer survey: residential real estate. The National Association of Realtors reports that looking online is the first step for a large and growing share of buyers, in the range of 41 to 47 percent across recent survey years, ahead of contacting an agent or asking a person.

That same survey contains the counterweight, and it is the reason this piece resists a simple death-of-word-of-mouth story. Even with search leading the awareness stage, roughly 88 percent of buyers still purchase through an agent or broker. The engine now precedes and directs the trusted human relationship rather than abolishing it. Awareness moved onto software first; the closing relationship moved last, or has not moved at all in some verticals.

What real estate shows structurally, the broader consumer data shows in aggregate. Consumer use of AI tools specifically to find a local business grew sharply in a single year, and general-population chatbot adoption climbed at the same time. The awareness stage is no longer served by one channel. It is spread across classic search, the map, review platforms, and a new AI-answer surface that did not meaningfully exist a few years ago.

Stage two, shortlist: the map pack ranking and the AI answer build the list

Once a buyer is aware, something has to narrow the field. For local-intent queries, the dominant narrowing device has been the local map pack, and its pull on attention is measurable. Aggregated Google local-search behavior studies report that searchers click the local three-result pack around 44 percent of the time, against roughly 29 percent for organic links and 19 percent for paid, and that position inside the pack matters, with the top spot drawing a materially larger click share than the second or third.

The magnitude here deserves a careful label. The direction, that the map pack captures a disproportionate share of local clicks and conversions, is well supported. The precise figures are aggregated from industry local-SEO research rather than a single disclosed primary methodology, so treat the exact percentages as indicative rather than settled. What is not in doubt is that the map pack is a shortlist-building machine, and that appearing in it is closer to being recommended than to merely being listed.

The newer shortlist builder is the AI answer. When an answer engine responds to a local query with a few named businesses, it has performed the same narrowing the map pack does, but by different rules, from different sources, and with even less transparency. That divergence is important enough to take on its own below.

How AI picks local businesses, and why it disagrees with the map

The convenient assumption is that whoever wins the map pack also wins the AI answer. The available evidence says otherwise. A business ranking in Google's top local-pack results has been reported to have less than even odds of also appearing in AI local recommendations, and visibility inside a large language model's local suggestions is described as far harder to obtain than a map-pack ranking, with one vendor putting the gap at roughly thirty times.

The mechanism behind the disagreement is that the two systems read different material. Google's AI Overviews and AI Mode draw primarily on the Google Business Profile as the local data source, with a review platform cited in roughly a third of AI local searches specifically for review synthesis. An answer engine is not re-running the map-pack algorithm. It is assembling a response from the sources it can retrieve and trust, which is why a business can be structurally strong on one surface and invisible on the other.

This evidence sits in the emerging tier. The sharpest figures come from a single vendor's proprietary study, so the exact thirty-times gap should be read as directional. The underlying claim, that classic local ranking and generative local recommendation now diverge measurably and must be measured separately, is consistent with the broader classic-versus-generative split the rest of this research program tracks. For a business, the practical consequence is blunt: winning the map does not buy you the answer, and you cannot know your position on the answer surface without sampling it directly.

Stage three, verification: what a buyer can check before booking

Having a shortlist is not the same as trusting it. The verification stage is where the buyer decides whether a candidate is safe to hire, and it is the stage where the synthesized signal has quietly taken over the work a personal reference used to do. The foundational evidence here is Michael Luca's study of Yelp, which found that a one-star increase in rating produced a 5 to 9 percent increase in restaurant revenue, an effect concentrated in independent restaurants rather than chains, because chain buyers already carry a quality prior from the brand.

The Luca result is the evidentiary spine of the whole verification stage: it shows that an aggregated public signal, a star rating built from many strangers, now moves revenue in the way a trusted recommendation once did. It is established evidence, from a genuine natural experiment, and it is specifically strongest for independent, owner-operated businesses, which is precisely the buyer this research program serves.

Verification also has a hard ceiling that varies by vertical, and it is most acute in the regulated trades. In aesthetic medicine, buyers routinely cannot verify the credential that actually governs their safety, such as board certification or the practice's physician-supervision structure, from what they can see online. The peer-reviewed evidence on provider selection shows that narrative reviews and the content of those reviews measurably shift which provider a patient picks, which means the buyer substitutes the signal they can read for the credential they cannot. The verification stage, in other words, does not always verify the thing that matters. It verifies the thing the engine can surface.

Stage four, booking: where the funnel converts or leaks

The final stage is the one owners feel in their revenue. A buyer who has become aware, been shortlisted by an engine, and verified a candidate still has to complete a booking, and the funnel leaks here in ways that have nothing to do with search. A slow site, an unanswered call, a booking form that asks too much, or a delivery platform that intermediates the transaction can lose a buyer who had already chosen.

The vertical evidence shows the booking stage is itself increasingly channel-mediated rather than direct. In restaurants, roughly three-quarters of traffic is now off-premises, and the choice of ordering platform has itself become a comparison-shopping decision. In home services, shared-lead marketplaces route the same homeowner to several competing contractors at once, so the booking a business thought it had earned is contested at the moment of contact. Each of these inserts another piece of software between the shortlist and the confirmed job.

The reading here is that the engine now shapes who gets considered and how the booking is completed, and that a business can win the first three stages and still lose at the fourth. Which is why a diagnosis that stops at rankings is incomplete. The question is where the specific buyer, in the specific vertical, actually drops out.

The nine verticals, mapped to the funnel

The single funnel holds across every vertical this research program studied, but the strength of the evidence and the location of the tightest constraint differ. The list below maps each vertical to the funnel and states the confidence plainly, because a synthesis that spans findings of unequal strength has an obligation to say which parts rest on peer-reviewed work and which rest on industry-reported estimates.

• Real estate: awareness is search-first (41 to 47 percent), yet the agent still closes (about 88 percent). Established. The commission settlement has also moved the buyer-agent selection signal into the open, changing what buyers weigh at verification.
• Med-spa and aesthetics: the verification stage has a hard credential-legibility ceiling; buyers cannot see the supervision structure that governs safety and lean on review content instead. Established on the verification gap and the review-content effect; the funnel mapping is our synthesis.
• Dental: reviews, insurance, and location dominate verification before a new-patient booking. The review-and-choice mechanism is established; vertical-specific journey percentages in circulation could not be traced to a primary source and are deliberately not cited here.
• Solo and small legal: an affirmative truthfulness standard governs what a firm can claim, which constrains the awareness and verification stages more tightly than in unregulated trades. Established on the rule; the AI-answer application is analysis, not adjudicated law.
• Restaurants: verification runs on the star aggregate (the Luca effect), and booking has shifted off-premises onto delivery platforms. Established on both the ratings-revenue finding and the off-premises majority.
• Fitness and wellness: membership behavior is well documented, but discovery behavior specifically is a genuine white space; no primary study was found describing how consumers search for a studio. Mapping offered as a hypothesis, labeled as needing primary data.
• Professional services and B2B: the shortlist stage has gone mostly self-directed and invisible before contact, with buyers using AI tools to build and vet vendor lists. Emerging; the sources are vendor-sponsored and directionally consistent rather than settled.
• Local retail and ecommerce: near-me plus in-stock intent has made the physical store a first-class answer-engine result at the shortlist stage. Emerging; the supporting statistics are secondary-sourced industry panels.
• Home services: the map pack drives the shortlist, and shared-lead marketplaces contest the booking. The map-pack pull is established in direction; the marketplace acquisition-cost figures are contractor-anecdote sourced and flagged as needing primary data.

The evidence, graded: established, emerging, mixed

A synthesis is only as trustworthy as its willingness to grade its own inputs. The funnel model above draws on findings of three different strengths, and collapsing them into one confident narrative would overstate what the evidence shows.

The established layer is the load-bearing one: the ratings-revenue natural experiment, the peer-reviewed provider-selection research, the long-running real-estate buyer survey, and the disclosed adoption figures from national survey work. These carry the argument. The emerging layer, the AI-versus-map-pack citation gap, the AI-tool adoption jump, and the self-directed B2B journey percentages, points in a consistent direction but rests on single-vendor or vendor-sponsored studies, so its numbers are treated as indicative. A third set of figures, particularly home-services acquisition costs and dental journey percentages, is flagged as needing primary data and is not stated as fact.

The reason to publish the confidence labels rather than hide them is that the verticals where the buyer cannot verify quality on their own, legal, dental, and medical aesthetics, are exactly the verticals where a credential-accurate signal is structurally necessary. In those markets, dishonest visibility work violates platform policy and professional-conduct rules at the same time. Grading the evidence by confidence level is not a stylistic choice here. It is the only method that functions where the buyer has no other way to check.