Who the Machine Names When Flowers Cannot Wait: Order Gatherers, Local Florists, and the Time-Critical Visibility Terrain

Introduction

Few retail purchases carry the combination of pressures a florist order does. The buyer is usually racing a deadline, a birthday that is today, a funeral tomorrow, an anniversary already half-forgotten, so the decision gets made fast. The product is perishable and cannot be stockpiled, which rules out the warehouse-and-ship model that reshaped so many other categories. And the emotional weight runs high, because flowers are rarely bought for the self; they are a message sent to someone who matters. This is a category bought in a hurry, for someone else, with feeling, and that precise shape of demand governs how its visibility terrain behaves.

That shape has bred a distinctive intermediary. For decades the floral trade has contended with the order gatherer, an operator that markets itself as a local flower shop, collects the payment, then relays the order to a real florist to fill, keeping a cut for having stood in the middle. Order gatherers reach the buyer through the wire services that route floral orders between shops, and through advertising that borrows local-sounding names, city-specific page titles, and stock arrangement photography. The practice is documented well enough that at least one US state, Texas, passed a law against misrepresenting a floral business's geographic location. The middleman is not a hypothetical; it is a structural feature of how flowers are sold.

The question this study holds constant across the series takes on a sharper edge here. When a shopper looks for a florist, who does the machine name, and where does the real local shop stand against an intermediary engineered to look exactly like it? A category defined by urgency is precisely where a buyer is least able to check whether the name at the top of the results is a neighborhood shop or a call center three states away.

Data and method. We captured Google's results for five real shopper prompts across five metros on 2026-07-22, separating the two surfaces a shopper actually sees: the local map 3-pack and the organic listings beneath it, and requesting an AI Overview on every query. Each result was classified by operator type. To situate the capture in the category's demand, we also read Google's reported monthly search volume for florist near me, the dominant buyer query, across the twelve months to June 2026: demand peaked in February 2026 at about 673,000 searches and bottomed in November 2025 at about 301,000, a roughly 2.2 times swing that confirms this is a category whose demand crests around Valentine's Day, which is why visibility has to be earned before the season rather than during it. We report plainly what we could and could not reach: the Google result surfaces were captured directly and dated, at one snapshot per query on the primary grid, with a separate five-capture intra-day stability recapture on 2026-07-23 reported in the stability panel; the standalone chat engines (ChatGPT, Perplexity, Gemini) were captured separately on 2026-07-23 and are reported in their own section below, while Bing Copilot remained uncaptured. Nothing we could not reach is estimated.

Background and literature

The florist purchase sits awkwardly across the classic economic typology of goods. Nelson separated search goods, whose quality a buyer can judge before purchase, from experience goods, whose quality is revealed only in consumption (Nelson, 1970). A bouquet is partly an experience good for the recipient, but for the sender it edges toward a third type. Darby and Karni named the credence good, whose quality the buyer cannot verify even after the fact, and showed how that gap opens room for fraud (Darby and Karni, 1973). The person ordering funeral flowers from four states away never sees what arrived; they cannot confirm the stems were fresh or the arrangement matched the photograph. That unverifiability is the exact condition Akerlof modeled, in which information asymmetry lets lower-quality or misrepresented offerings crowd out honest ones (Akerlof, 1970). Emons sharpened the point for markets where the seller is also the expert who defines what the buyer needs, proving that such settings carry a standing incentive to overstate, misrepresent, and defraud unless some market mechanism disciplines it (Emons, 1997). The order gatherer is a textbook exploitation of credence: the buyer cannot check, so the appearance of locality is allowed to substitute for the fact of it.

The emotional register of the purchase deepens the exposure. Sherry's anthropology of gift exchange framed a gift as a social transaction in three stages, gestation, presentation, and reformulation, in which the object stands in for the relationship between giver and receiver (Sherry, 1983). Flowers are a near-pure instance: they are chosen less for their material properties than for what their arrival will say, which is why the sender fixates on the act of delivery and rarely audits the goods. A buyer optimizing a symbol under time pressure is a buyer who will accept the first operator that looks credibly local, because the felt cost of getting the gesture wrong dwarfs the effort of verifying who fills it.

Time pressure then compounds both problems. Pirolli and Card's information foraging theory holds that searchers follow the strongest available scent of value at the lowest cost and abandon deeper evaluation once the cost of continuing rises (Pirolli and Card, 1999). A shopper who needs flowers delivered this afternoon carries a very high cost of continuing, so the first credible-looking result captures the click. Dhar and Nowlis showed experimentally that time pressure makes shoppers less likely to defer or keep searching, pushing them to resolve a high-conflict choice quickly rather than hold out for a better one (Dhar and Nowlis, 1999). Iyengar and Lepper added that an abundance of options can demotivate choice altogether (Iyengar and Lepper, 2000); under deadline the shopper does not want a long list, they want one trustworthy name, which loads enormous weight onto whichever operator the machine surfaces first.

Reviews are the mechanism the market evolved to price credence. Dellarocas described how digital feedback systems industrialized word of mouth, turning scattered private experience into a public signal a stranger can act on before any transaction (Dellarocas, 2003). Luca then measured the consequence directly: displayed reputation moves real revenue, and the effect concentrates among independent businesses rather than chains with established brand equity (Luca, 2016). For a local florist the review corpus is close to the entire basis on which a stranger decides to trust it with an emotional errand. The US Federal Trade Commission's rule on consumer reviews, effective 2024, treats fake, incentivized, and suppressed reviews as violations precisely because that signal is so load-bearing (FTC, 16 CFR Part 465).

The newest layer is the generated answer. Aggarwal and colleagues introduced generative engine optimization, showing that structured, well-cited, entity-clear content is likelier to be included when an engine composes an answer rather than returning a list (Aggarwal et al., 2024). Whether that layer is even active for a given category is itself an empirical question, and one this study measured directly for florists rather than assumed.

The market these forces act on is large and fragmented. IBISWorld valued the US florists industry at 7.9 billion dollars in 2025 across 41,649 businesses (IBISWorld, 2026), and the Society of American Florists counted roughly 12,154 traditional retail florist shops in 2022 (SAF, 2022). A fragmented field of small independents, selling a credence-heavy symbol under deadline, is a terrain built for intermediary capture. The rest of this report measures where that capture actually occurs, and, just as importantly, where it does not.

Findings: the map pack belongs to local shops

The clearest result in the study is also the most reassuring for independents. Across all 25 query cells, the Google local map 3-pack, the block of three businesses shown with a map, was composed entirely of independent local florists. Every one of the 75 pack slots captured, in all five metros and on both the reputational and the transactional prompts, went to a real local shop. No order gatherer, wire-service brand, grocery chain, or directory appeared in a single pack position anywhere in the grid.

This is a strong signal that local shops are not merely eligible to be named; on the surface Google reserves for local intent, they are the only operators named at all. The pack runs on proximity, prominence, and the Google Business Profile, and for this category those mechanics resolve cleanly to the neighborhood shop. For an independent the pack is therefore winnable, and it is also the surface where the credence problem is most directly solved, because it presents the shop's own name, rating, and reviews to the buyer before a single click.

Ranking in the pack, however, is downstream of legibility. Before an engine can name a shop it has to read it as a confident entity. To probe that, we audited the publicly observable structured data of six real independent shops. Readiness was uneven: four of the six carried solid local-business markup with opening hours, price range, and in two cases aggregated ratings, while two did not, including a celebrated Dallas boutique whose modern storefront served commerce and product markup but no detected local-business schema at all. The irony of the set is that the most heavily marked-up site belonged to a shop running on a wire-service template, which shipped structured data by default. Presence in the pack shows the shops are being resolved; the audit shows that resolution rests on signals many shops have not deliberately engineered, and could lose.

Findings: the organic layer is where the middleman lives

Beneath the map pack, the organic listings tell a different and query-dependent story. Pooling all 460 organic slots captured, independent shops held about 74 percent, with the remainder split between order-gatherer and wire-service properties, national chains and grocery floral, and directory or editorial pages. But the aggregate hides the effect that matters, which appears only when the organic results are split by what the shopper was trying to do.

On reputational queries, where the shopper is comparing options (best florist, florist near me, local flower shop), independent shops held about 76 percent of organic slots, order gatherers and wire services held about 7 percent, and the rest was mostly directories and best-of media. On transactional queries, where the shopper is under deadline (same day flower delivery, flower delivery near me), the order-gatherer and wire-service share rose to about 18 percent, roughly two and a half times its reputational level, and the combined non-local commercial share, adding grocery and chain floral, climbed from about 8 percent to about 26 percent. The independent share of organic fell correspondingly. In Dallas, the same-day-delivery query returned 1-800-Flowers, From You Flowers, and Kroger's floral service in its highest commercial organic positions, with the marketplace UrbanStems close behind, all sitting above or among the local shops.

This intent effect was consistent across every metro. The transactional intruder share of organic ran from about 20 percent in New York and Atlanta to about 32 percent in Phoenix, never once falling to its low reputational level. The pattern is not a quirk of one city; it is how the category's organic surface behaves under deadline.

The synthesis of the two surfaces is the study's central observation. The order-gatherer hijack is real, but it is contained to one surface and one intent: it lives in the organic listings for the urgent, buy-now query, which is precisely the query that defines florist demand. The map pack directly above those listings stays entirely local. So the shopper most exposed to an operator engineered to look local is the one who skips the pack, in a hurry, and forages for the first credible name below it, while the real shop sits both in the pack above and further down the same organic list. The intrusion is not everywhere; it is aimed at the exact moment of least scrutiny.

A related structural fact surfaced during capture. Many independent shops run their own storefronts on wire-service e-commerce platforms and describe themselves as an FTD or Teleflora florist, so the intermediary layer is embedded inside the independents' own technology stack. Independence in the shopfront does not always mean independence in the plumbing. These organic readings rest on a single snapshot per query and are tiered accordingly.

Findings: the answer layer is emerging, not yet the battleground

We requested a Google AI Overview on all 25 florist queries in the primary grid. Google served one on none of them. Across every prompt and every metro, the generated-answer box that increasingly sits atop results for informational queries did not appear for florist shopping intent on the 2026-07-22 capture date. A separate k of 5 stability recapture the following day, however, surfaced sporadic AI Overviews in a small share of cells, so the layer is better read as beginning to flicker in than as permanently dark.

This near-absence is a finding, not a gap. It means that for this category, as of the primary capture, the machine is not yet compressing the florist decision into a single named recommendation at the top of Google, even as the recapture shows it starting to appear. The decision still happens where this study measured it, in the map pack and the organic listings, which is good news for independents that hold the pack. It also sets a baseline. The literature says the generated layer, where it does appear, rewards machine-legible entities and names very few businesses. As Google begins serving AI Overviews for florist queries, the shops that are illegible as local entities today, the ones the structured-data audit flagged, are the ones most likely to be left out of the sentence tomorrow. The narrow window buys time to fix legibility, not permission to ignore it.

The Google AI Overview is one answer surface; the standalone chat engines are another, and those we did capture. On 2026-07-23, with web search enabled, we put both the reputational and the same-day prompts to ChatGPT, Perplexity, and Gemini across three of the five metros, and the next section reports who each one named. Copilot on Bing remained uncaptured this round and is named as such.

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

Between the primary Google grid and this analysis we captured the standalone answer engines directly. On 2026-07-23, with web search enabled, we put both the reputational prompt (best local florists in [metro]) and the transactional prompt (where to buy flowers for same-day delivery in [metro]) to ChatGPT, Perplexity, and Gemini across three of the five metros, New York, Dallas, and Chicago, at two runs each, a grid of 36 captures that all returned. The most striking result is how differently the three engines resolve the same question. ChatGPT answered every one of its twelve captures as a short list of named local shops, each presented with its address, its review count, and a link to the shop's own website, and it named more than thirty distinct independent florists across the three metros. It named no order gatherer, no wire-service brand, and no directory in any answer, and, critically, it stayed fully local even on the same-day prompt, surfacing neighborhood shops such as Ruscus Flowers and Superior Florist in New York, Estrella's Flower Shop and In Bloom Flowers in Dallas, and H.M. Floral Studio and Ashland Addison in Chicago rather than routing the urgent order to an intermediary.

Perplexity behaved like the Google organic layer it draws on. On the reputational prompt it named a strong roster of local independents, Julia Testa, PlantShed, Starbright and Élan Flowers in New York, Avant Garden and Dr Delphinium in Dallas, Nerine and Flowers for Dreams in Chicago, and grounded them in best-of media and directories rather than the shops' own sites, citing Modern Luxury, Petal Republic, Thursd, Time Out, Reddit and D Magazine. On the transactional same-day prompt, however, the order gatherer intruded exactly as it does in Google organic: in all three metros Perplexity named and cited wire-service and national delivery brands, 1-800-Flowers, FTD, From You Flowers, ProFlowers, UrbanStems, The Bouqs Co., and the marketplace Floom, mixed in among a few local shops. The same intent effect this study found on Google's organic surface reappears inside the chat answer, and it is concentrated in the engine that leans hardest on the open web.

Gemini named local independents most generously of the three, describing them in editorial prose organized by neighborhood and style, and it led with local shops even on the same-day prompt, adding only a national delivery brand or two such as UrbanStems or Floom rather than handing the query to them. Its citations, however, could not be classified: every source Gemini returned was masked behind a single Google redirect host (vertexaisearch.cloud.google.com), so the engine was read by the businesses named in its text, not by its links. Consistency separated the engines further. ChatGPT was the most repeatable, returning essentially the same set in a reshuffled order across its two runs; Perplexity's top named set overlapped substantially between runs; Gemini reshuffled the most, holding a stable core of anchor shops (Élan and PlantShed in New York, Dr Delphinium and McShan in Dallas, Flowers for Dreams and Ashland-Addison in Chicago) while swapping much of the surrounding list, including introducing fresh names such as fleursBELLA and Field & Florist on the second run. Across engines the picture is convergence on a handful of anchor independents per metro and divergence across a long tail of everything else.

Read against the study's thesis, the chat engines mostly extend the Google finding rather than overturn it. Independents win the map pack, and two of the three chat engines also name them first, ChatGPT exclusively and Gemini predominantly. The order-gatherer hijack that this study located in Google's transactional organic listings does reappear in the synthetic answers, but it is not uniform: it is concentrated in Perplexity's same-day answer, largely absent from Gemini, and absent altogether from ChatGPT, whose profile-grounded answers behaved like the local map pack rather than the organic layer. This is a single-day, small grid at two runs, and synthetic answers are non-deterministic and drift over time, so the reading is directional. But the direction is clear enough to matter: the engine a shopper reaches for changes whether they are handed a real local shop or an operator engineered to look local, and the same-day query is where that difference is sharpest.

Discussion

The two Google surfaces tell a coherent story once read through the buying mode. The map pack is Google's answer to local intent, and its proximity-and-prominence mechanics resolve cleanly to the neighborhood shop, which is why it was entirely independent. The organic layer is a more open contest, and there intent is decisive. On the reputational query the shopper forages with time to spare, sampling directories and best-of lists and rewarding shops with earned reputation, and the credence problem is partly solved by the visible reviews Luca and Dellarocas identified as the market's substitute for verification. On the transactional query the forager takes the first credible scent, deferral collapses under deadline as Dhar and Nowlis predicted, and verification never happens, which is the exact seam the order gatherer is built to exploit and the standing incentive Emons described. The hijack concentrates where scrutiny is weakest, which is not an accident of the market but its design.

This reframes the order-gatherer debate away from morality and toward measurement. The intermediary does not win the pack and does not dominate the reputational query; it takes a slice of the organic listings at the moment of least scrutiny, by being more legible and more present on the transactional query than a busy independent has bothered to be. That is a diagnosis with a remedy, not a grievance. With AI Overviews absent from the primary grid and only beginning to surface on recapture, the map pack and organic listings are still very nearly the whole board. As the generated layer switches on for florists, the same legibility deficit the structured-data audit found would convert directly into absence from the answer, because that layer names so few businesses that being unreadable is indistinguishable from being unranked.

The chat engines let us see the mechanism twice over, because each one inherits the bias of the surface it leans on. ChatGPT grounded its answers in the shops' own profiles and behaved like the map pack, naming only independents even under a same-day prompt. Perplexity leans hardest on the open web, so it reproduced the open web's fault line: local independents on the reputational query, order gatherers and wire services on the same-day one. Gemini sat between them, leading local but masking every citation behind a Google proxy. The same intent effect thus appears three times, on Google organic, inside Perplexity's transactional answer, and nowhere the engine is grounded in first-party entity data, which says the fix is not engine-specific. A shop that is legible as a confident local entity is the shop each of these systems can name, whichever surface the buyer reaches for.

This is where the RavenEye model applies directly. We read a market's terrain as a Visibility Corpus, the map of where attention actually sits for a category and a place, and we treat the remedy as Search Surface Optimization, the disciplined engineering of the signals that decide who gets named across classic search, the map pack, and the answer layer, measured to one number, the Machine-Readiness Score. For florists the corpus reading is unusually actionable: hold the pack, which is winnable on profile and reviews; contest the transactional organic query, where intermediaries have simply engineered legibility and presence harder than the shops they relay to; and get entity-legible now, while the answer layer is still dark for this category. The remedy is not to out-shout the middleman but to become the more confident local entity on the surfaces and queries the terrain shows are decisive.

None of this is a guarantee. Placement in a pack or a generated answer is influenced by proximity, personalization, and undocumented model behavior that no firm controls, and these readings are single dated snapshots. What the terrain shows is where the advantage sits, and for this category it sits on holding the pack, on the transactional organic query, and on entity legibility, which is a more actionable finding than any single headline number.

Implications for retailers

For an independent flower shop, the practical reading of this study is specific. First, the map pack is yours to hold, and the levers are an accurate, complete Google Business Profile and a real, current review corpus answered in your own voice, because those are the signals both the pack and any future answer layer lean on when they decide who to trust with an emotional purchase. Second, the transactional organic query is where you quietly cede ground to operators engineered to look local, and closing that gap is an entity and legibility problem, not a budget problem: one consistent set of business facts across the web, local-business structured data a machine can parse, and a same-day-delivery presence that resolves to you rather than to a keyword-domain intermediary.

Third, legibility is the precondition for everything else, and the moment to fix it is now, while the generated-answer layer is still absent for this category. A shop that serves no local-business schema is asking the machine to guess what it is, and machines under deadline do not guess, they name the clearer option. That is a fixable defect, and fixing it does not touch the craft that makes the shop worth choosing.

Holding the pack and being named more often is achievable. Being guaranteed a spot is not, since placement depends on proximity, personalization, and model behavior no outside firm controls. The work is to engineer every signal that can legitimately be moved and to measure the movement, including where it is flat. That is the remedy this study points to, offered as engineering, not as a promise of lift.