When the Machine Recommends a Nursery: Local Visibility for Garden Centers and Plant Nurseries

Introduction

A plant is a difficult thing to sell over the internet. It is alive, heavy, and fragile; it is perishable; and its quality is judged with the hands and the eyes at the moment of purchase. A shrub bought online arrives crushed, leggy, or wrong for the region, and a flat of annuals cannot survive a slow delivery. So garden centers and plant nurseries remain one of the most stubbornly local corners of retail. When someone wants plants, they do not open a shipping cart. They ask where to buy plants near them, and then they drive.

That physical reality shapes the visibility terrain in a way the rest of retail does not share. In categories that ship easily, the machine answers a "near me" query by naming Amazon, a national chain, or a marketplace, and the local shop is crowded out before it is even considered. Plants resist that gravity. The question this series holds constant is therefore especially pointed here: when a shopper looks for a nursery, who does the machine name, and are independent local shops eligible to be named at all?

This report treats that question empirically. It reads the surfaces a nearby buyer actually meets, the Google local map pack, the organic results, and the Google AI Overview, across five US metropolitan areas, and it classifies who appears: an independent local shop, a national chain, an online marketplace, or a directory that lists everyone and answers nothing. It then examines the independents that surfaced and asks whether their public presence is engineered well enough to keep being named. The buying mode, seasonal and perishable and local by necessity, is the lens throughout, because it explains both why the terrain favors independents and why so many of them still forfeit the surface they should own.

Background and literature

Economists have long sorted products by how a buyer can verify quality. Nelson distinguished search goods, whose attributes can be judged before purchase, from experience goods, which must be consumed to be assessed (Nelson, 1970). Darby and Karni added credence goods, whose quality a buyer cannot fully judge even after consumption, such as expert diagnosis and repair (Darby and Karni, 1973). A live plant sits across these categories in an unusual way. Its immediate condition is a search attribute a shopper can inspect on the bench, which is exactly why the purchase stays physical; yet whether it will thrive in a given yard, soil, and climate is closer to a credence attribute that depends on the seller's horticultural expertise. This is the theoretical seat of the independent nursery's advantage: staff who know which plant belongs in which spot supply a credence assurance the warehouse shelf cannot.

The deeper problem beneath a credence purchase is information asymmetry. Akerlof showed that when a buyer cannot verify quality, good sellers and bad ones pool at the same price and trust collapses toward the worst case (Akerlof, 1970). A shopper choosing a nursery faces exactly this uncertainty before the drive, and resolves it with proxies the seller cannot fake cheaply. Spence formalized the escape: a costly, hard-to-imitate signal separates the able from the rest (Spence, 1973). Years of accumulated reviews, a complete and consistent profile, and a resolvable identity are precisely such signals, expensive to accumulate and easy for the engine to read, which is why the machine can lean on them to build its shortlist. The independent that has earned these signals is legible to the algorithm as a safe recommendation; the one that has not is indistinguishable from the lemons.

The information a buyer gathers before acting has its own structure. Information foraging theory models search as a cost-benefit hunt in which people follow the strongest available scent of value and abandon a patch when the return falls (Pirolli and Card, 1999). A map pack of three named, rated businesses is a far richer scent than a page of blue links, which is why local intent so often ends inside the pack. Choice overload compounds the effect: presented with too many undifferentiated options, people defer or disengage (Iyengar and Lepper, 2000). The engine dissolves the overload by pre-selecting a shortlist, and being on that shortlist, or absent from it, is the whole game.

Reviews are the currency that decides the shortlist. Luca's study of Yelp found that a one-star increase in rating moved revenue by roughly five to nine percent for independent restaurants, with no such effect for chains whose reputations are already fixed (Luca, 2016). The asymmetry is the point: a chain carries its reputation into every location, while an independent must build its signal review by review, so the marginal review is worth far more to the local shop. That mechanism travels directly to nurseries, where an independent with genuine review depth reads to the engine as a trusted local entity. Channel research on showrooming and webrooming describes buyers who research in one channel and buy in another (Verhoef, Kannan, and Inman, 2015). For plants the webrooming pattern dominates: the buyer researches online and completes the purchase in person, which makes online visibility the front door to an in-store sale rather than a competing channel.

Two further bodies of work frame the moment. The pandemic produced a documented surge in gardening participation and a measurable link between gardening and wellbeing that has partly normalized rather than fully receded (Gulyas, Caton, and Edmondson, 2024; Feller and colleagues, 2026), enlarging the buyer pool this study observes. And the emerging science of generative engine optimization shows that content and entity signals, clear structure, citations, and corroborated facts, lift how often a business is included in generated answers by up to roughly forty percent (Aggarwal and colleagues, 2024). That finding matters here because the surfaces measured in this study, from the map pack to the chat answer, are increasingly assembled by machines that reward the resolvable and ignore the ambiguous. The theory converges on a single operational claim: the nursery the machine can identify and corroborate is the nursery it names.

The demand terrain

Before asking who wins the surface, it is worth establishing what shoppers type, because the language reveals the mode. Real US monthly search volumes, captured on 2026-07-22, show that buyers speak in plain proximity, not in trade vocabulary. The phrase "nursery near me" drew about 550,000 searches a month, "plant nursery near me" about 246,000, and "garden center near me" about 135,000. The industry's own preferred self-description, "independent garden center", drew about 720, and wholesale trade terms such as "container grown plants" and "balled and burlapped trees" sat in the low hundreds.

The gap is not a rounding difference; it is roughly three orders of magnitude. For every shopper who searches the way the trade press writes, hundreds search "nursery near me". That has a direct operational consequence. A nursery that fills its website and profile with horticultural nomenclature is optimizing for language almost no buyer uses, while the enormous proximity demand is decided by local signals, an accurate profile, a resolvable location, and review depth, that have nothing to do with plant Latin. The demand is huge, local, and phrased in the vernacular. The only open question is who the machine hands it to.

Findings: who the machine names

Across the five metros we attempted 40 metro-by-prompt cells. Thirty-three returned usable SERP data and seven returned an empty result set for long-tail phrasings such as "independent garden center [city]", a small finding in itself, since the trade's own phrasing returns little. Within the 33 usable cells the local map pack was present on 31, or 94 percent. The pack is the surface that answers this category.

The composition of that pack is the headline. Of 89 classified local-pack slots, 88 were independent local nurseries and one was a chain, a 98.9 percent independent share. National big-box garden departments and online marketplaces were almost entirely absent from branded "garden center near me" and "plant nursery near me" packs. This is the perishability effect made visible: because the machine reads these queries as requests for a nearby place to buy living inventory, it surfaces the businesses built around that inventory, and those are overwhelmingly independents. The named shops were real and well-reviewed, from Gethsemane Garden Center in Chicago and North Haven Gardens in Dallas to Dig It Gardens in Phoenix and the 14th Street Garden Center in the New York metro.

The organic top ten told a more contested story. Of 212 classified organic results, independent and brand sites held 65.1 percent, but directories and aggregators such as Yelp, Yahoo Local, and listing hubs took 23.1 percent, chains 9.4 percent, and marketplaces 0 percent. So while independents own the map pack, roughly one organic slot in four is held by a page that lists nurseries rather than being one, an intermediary that captures the click and monetizes the buyer's attention on the way to the shop.

The AI Overview was conspicuous by its rarity. Google returned an AI Overview block on only 4 of 33 queries, about 12 percent, and even then as an asynchronous placeholder whose reference content we did not retrieve under budget. For local, proximity-driven garden queries, Google today answers with the pack, not with a generated paragraph. The observation is time-stamped and consequential: the AI-answer risk that dominates shippable categories has not yet reached the nursery's door, which hands independents a window to secure the local surface before it does.

Finally, eligibility, and here the picture sharpens into a paradox. The independents that appeared were not marginal. Across 39 unique named shops the median rating was 4.7 and the median review count was 170, with a range from 3 to 954 reviews. Review depth, not chain budget, is the sorting signal, and the thin-review tail is what gets buried. Yet the on-page picture was weaker. In an audit of eight independent nursery homepages, JSON-LD structured data was present on five, but not one carried a LocalBusiness, Store, or GardenStore schema type; the markup found was generic WebSite and Organization data from a default theme. Zero of eight declared the local-place entity, with address, hours, and geography, that engines and machine answers use to resolve which business is which. The independents are eligible through their Google Business Profiles and their reviews, and at the same time invisible as entities on their own sites, strong where the platform vouches for them and silent where they could vouch for themselves.

The AI answer engines: who ChatGPT, Perplexity, and Gemini name

The Google capture above measures the classic surfaces. To see whether the synthetic answer engines route the same category to the same businesses, the study was extended with a second grid on 2026-07-23: three engines (ChatGPT, Perplexity, and Gemini, each with web search enabled) were asked two question types, a reputational prompt ("best local plant nurseries and garden centers in [metro]") and a purchase-intent prompt ("where to buy plants for my garden in [metro]"), across three of the five metros (New York, Dallas, Chicago), twice each (k=2), for 36 captures in all. Every capture returned a usable answer. The full text of each was read and the businesses it named were classified by the same scheme used for the Google surfaces.

The headline is that the chatbots agree with the map pack. All three engines answered these queries by naming independent local nurseries, not big-box garden departments, and they did so in every metro. Each engine's lead picks were independents: Chelsea Garden Center and Urban Garden Center in New York; North Haven Gardens, Ruibal's Plants of Texas, Nicholson-Hardie, and Redenta's in Dallas; Gethsemane Garden Center and Christy Webber Farm & Garden in Chicago. Home Depot and Lowe's surfaced only as an occasional budget aside inside a few Perplexity intent answers, never as the recommendation, and big-box garden departments were otherwise absent. The one chain to recur was Calloway's Nursery, a regional Dallas-Fort Worth garden-center chain, named in Dallas alongside the independents. The perishability effect that shapes the map pack reaches into the generated answer: asked for a nearby place to buy living inventory, the machine names the independent.

Where the engines diverge is in how they source the answer, which matters for a business that wants to be named. ChatGPT returned structured cards (business name, rating, review count, address) and cited the nurseries' own websites; all 58 of its citations pointed at primary shop domains, none at directories. Perplexity named independents just as readily but leaned on the aggregator layer in its citations: of 171 citations, the largest single source was Reddit (36), followed by YouTube and city and lifestyle media (Curbed, CultureMap, Secret NYC, TrustAnalytica, Dallasites101), the same intermediary tier that taxes the organic surface. Gemini named the most independents of the three and framed them explicitly against the chains ("skip the big-box home improvement stores"), but it masks every one of its 425 citations behind a Google redirect (vertexaisearch.cloud.google.com), so its sources cannot be classified from the URL and were read only through the businesses named in its prose. That masking is a real measurement limit, recorded here rather than papered over.

Across the two runs the anchor names were stable while the long tail moved. ChatGPT was close to identical run to run, repeating the same shops with the same review cards and swapping at most one name. Perplexity and Gemini repeated their lead picks reliably but resequenced the tail and substituted fresh shops between runs, the expected signature of a non-deterministic system. Read against the study's thesis, the finding is favorable and provisional: the chatbots, like the map pack, name independents today, so the category's structural advantage reaches into synthetic answers, for now. But the same eligibility logic governs which independents recur, depth of reputation and a resolvable identity, and the entity gap that leaves these shops undeclared on their own websites is precisely what will decide whether the next generation of answers keeps naming them.

Discussion

The pattern is coherent once read through the buying mode. Because live plants are perishable and local by necessity, the machine treats "nursery near me" as a request for a place, and the local pack is how it answers a request for a place. Chains and marketplaces, which win shippable categories by aggregating supply and reviews at national scale, have little purchase on a query the engine has already decided is about proximity to living inventory. That is why the independent share of the pack is near total. Structurally, this is the most winnable category in the series for independents, and the theory predicts it: the credence assurance a shopper wants for a plant, will it thrive here, is exactly what a knowledgeable local seller supplies and a warehouse cannot.

The cross-surface reading strengthens that conclusion rather than complicating it. Google's map pack and all three synthetic answer engines converged on the same independents in the same metros, even though each was built by a different company from a different index. Convergence from independent systems is the signature of a real underlying signal, not a quirk of one ranking algorithm. What the engines share is not code but a dependence on the same public evidence, the profile, the reviews, the corroborated identity, so a shop that is legible on those signals tends to be named everywhere at once, and a shop that is not is omitted everywhere at once. The advantage is therefore structural, not surface-specific, which is the reassuring half of the finding.

But winnable is not won. Two leaks drain the advantage. The first is the directory tax on the organic surface: nearly a quarter of the top ten is held by aggregators that rank for the buyer's query and then route, list, or monetize the attention. Every directory slot is a place a shop's own site could have stood, and the same intermediary tier reappeared in Perplexity's citations, so the tax is not confined to classic search. The second and deeper leak is the entity gap. A business the engine cannot confidently identify is a business it recommends with less confidence, and the on-page audit found independents strong on the profile and reviews yet silent as entities on their own websites, with no LocalBusiness schema and no corroborated address and hours in a form the machine reads directly. The generative-engine literature is explicit that resolvable, corroborated entities are the ones machine answers include (Aggarwal and colleagues, 2024). As AI Overviews and chat answers extend into local, the shops that have not declared themselves will be the ones the next surface omits, and the omission will happen quietly, at the moment of decision, where the owner cannot see it.

This is precisely the terrain the RavenEye model is built to read and hold. The Visibility Corpus is the growing body of measured, dated observations of who gets named across engines and metros; this study is one contribution to it, and it is what lets a claim like "independents hold 98.9 percent of the pack but zero of eight declare a local entity" be stated with a date rather than asserted from intuition. Search Surface Optimization is the practice of engineering the four things that decide the outcome, the profile, the entity, the reputation, and the technical surface, as one coordinated system rather than a pile of listing tasks. And the Machine-Readiness Score is the single measured number that tells a nursery where it stands and whether the gap costing it customers is eligibility, entity, reviews, or the directory layer. The finding here is not that independents are losing. It is that they are positioned to win and mostly have not engineered the surface to lock it in.

Implications for retailers

For an independent garden center or nursery, the strategic reading is unusually favorable and unusually specific. The map pack is yours to lose, and the levers that decide it are within reach without outspending a national chain. The work is to be the most resolvable, best-reviewed, most completely described local entity in your area, and to keep being it through the season and past it.

Concretely, that means a Google Business Profile that is complete and current rather than claimed and forgotten; a business identity that reads identically across every directory, so the engine resolves you to one confident entity; a LocalBusiness schema on your own site that declares your address, hours, and place, so the machine can read you directly; a steady flow of real reviews answered in your own voice; and a technical surface that loads and works the moment a buyer arrives. These are offered as a remedy for a measured gap, not as a guaranteed ranking. Local placement personalizes by proximity and prominence that no firm controls, and reviews must be earned from real customers under FTC rules. What can be engineered is every signal that legitimately moves the outcome, and for this category those signals are close at hand and mostly unattended.