June 12, 2026
Published: June 2026 · Period: Q2 2026 · By: AICV (AI Coachella Valley)
Source material: Original AICV research produced through a multi-agent agent-mapping workflow
V2 — June 12, 2026. This version supersedes the V1 published June 11, whose visibility findings were drawn from a 377-establishment sample (44 percent of the universe). The inspection is now complete — all 859 non-mapped fixed-location independents, individually — and every visibility figure below is full-universe. Per AICV’s methodology convention, prior figures are superseded explicitly, never silently revised; where the headline numbers changed, here is the movement: structured data 0 of 232 checkable → 0 of 554; crawler blocking 92 of 298 (31 percent) → 201 of 707 (28.4 percent); reputation-to-visibility gap notes 176 → 349; visibility scores 212 high / 143 medium / 22 low (sample) → 552 high / 264 medium / 40 low / 3 invisible (full); multi-source corroboration 1,043 (73 percent) → 1,060 (74 percent) as the dataset matured. The independent count also moved, 924 → 971 raw rows of which 925 are fixed-location, and that movement is a segmentation event, not a recount: mobile vendors (26) and catering-only operations (9) were classified out of the fixed universe, duplicate and defunct rows (11) were named as such, and one missed fixed-location restaurant — Birrieria Sinaloa, Indio — was recovered in, inspected, and counted; the 47 reclassified rows were human-reviewed with two recorded overrides, and the decision record is preserved as the worked example on the methodology page. The one-establishment increments in the medium-score and gap-note tallies are that recovery.
This report does something the consumer review platforms do not: it counts the whole category. Not the restaurants that claimed a listing, not the ones with enough reviews to surface, not the ones that paid for placement — all of them. Across twelve Coachella Valley communities, AICV’s agent-mapping workflow identified, corroborated, and classified 1,423 food and dining establishments as of June 11, 2026. The number is the point. Before you can ask whether a region’s businesses are visible to AI agents, you have to know how many businesses there are. Almost no one actually knows, because the tools people use to answer that question — Yelp, Google, TripAdvisor — are built to surface the businesses that show up to them, not to enumerate the ones that do not.
This is the second entry in AICV’s agent-readiness research series and the first to go vertical-deep on a single category. It follows the State of the Coachella Valley Visitor Economy, which audited agent-readiness across the region’s publicly-listed visitor-economy businesses, and joins AICV’s earlier reports — State of AI — Q1 2026 and The Server Farm Next Door — as part of a growing body of regional intelligence on how the agentic shift is arriving in the valley. Where the visitor-economy audit started from a directory and scored what was in it, this report starts from the territory itself and maps the full category — including the long tail of operators that no directory fully contains.
With V2, the honesty note that anchored V1 resolves rather than repeats. The census number was always complete: 1,423 establishments, multi-source corroborated. The visibility findings, which V1 carried as a clearly-labeled 44 percent sample, are now complete too: every one of the 859 non-mapped fixed-location independents has been individually inspected — 100 percent coverage. There is no “sample of” left in this report’s denominators. Where a number could not be determined at inspection, the count of undetermined cases is stated beside it.
AICV is the regional intelligence network for the Coachella Valley’s emerging agentic economy — an agent-legible, agent-readable layer at aicoachellavalley.com designed to make local businesses discoverable, citable, and eventually transactable in an internet shaped by AI. AICV also operates aicoachellavalley.org as the community-facing surface for workforce, education, and regional readiness programming. Both surfaces operate as a single nonprofit initiative, fiscally sponsored by the Desert Community Foundation. The full methodology behind this census — how entities enter the corpus, what is verified versus editorial, and why inclusion is never for sale — is published as a standing reference: AICV Methodology: The Agent-Mapped Census.
A coordinated multi-agent sweep of twelve Coachella Valley communities identified 1,423 food and dining establishments as of June 11, 2026. Ownership splits the census into 971 independent rows, 442 chain operations, and 10 that could not be classified. Within the independents, the segmentation finalized in June 2026 decomposes the count precisely — and the decomposition proves itself:
| Independent segment | Count |
|---|---|
| Fixed-location independent operators | 925 |
| — of which already in the AICV dining corpus | 66 |
| — of which individually inspected | 859 |
| Mobile — trucks and pop-ups | 26 |
| Catering-only operations | 9 |
| Duplicate or defunct rows | 11 |
| Raw independent rows | 971 |
The 925 fixed-location independents — roughly 65 percent of the regional dining economy — are the addressable universe for agent-readiness work, and the inspection coverage of that universe is now total: 66 establishments were already represented in the AICV dining corpus, and the remaining 859 have each had their web presence individually inspected — 100 percent. The 47 rows reclassified in the segmentation (mobile, catering-only, duplicate/defunct, plus one recovery into the fixed universe) were resolved by agent-proposed, human-reviewed decision with two recorded overrides; the decision record is preserved and serves as the worked example in the methodology page’s Provenance and Human Gates section.
The census spans the nine incorporated cities of the Coachella Valley plus the unincorporated communities of Thermal, Thousand Palms, and Bermuda Dunes. The two densest dining markets — Palm Springs and Palm Desert — together account for 566 establishments, nearly 40 percent of the regional total.
| Community | Establishments |
|---|---|
| Palm Springs | 306 |
| Palm Desert | 260 |
| La Quinta | 178 |
| Cathedral City | 155 |
| Rancho Mirage | 135 |
| Indio | 117 |
| Coachella | 71 |
| Desert Hot Springs | 61 |
| Indian Wells | 55 |
| Thousand Palms | 34 |
| Thermal | 29 |
| Bermuda Dunes | 22 |
Every row carries provenance. Each record records which sources corroborated it, a confidence grade, and a retrieval date; segmentation is attribute-based rather than label-based, so each establishment carries chain affiliation, local ownership, service model, and physical format as separate fields. 1,060 of the 1,423 rows — 74 percent — are corroborated by two or more independent sources. Operating status distinguishes the valley’s known seasonal-summer-closure pattern from permanent closure: 1,255 open, 57 seasonal, 34 closed, and 77 of unknown current status.
A note on a second number readers may encounter: AICV’s State of the Coachella Valley Visitor Economy report, published the same quarter, counted Dining as 956 businesses (mean agent-readiness 3.01 of 8, 5.5 percent Tier A). That figure and the 1,423 here are not in conflict: the 956 was a directory-sourced scored subset — the dining operators listed in the regional visitor-economy directory that entered that audit’s eight-dimension rubric — whereas the 1,423 is a complete, ground-up establishment census built from the territory itself. The difference is scope and method, not a correction; the census is the larger number precisely because it includes the operators a visitor directory never contained.
Mapping the category is the first half of the work. The second is asking, of each independent operator, a different question than a diner would: not is it good? but can an AI agent read it, cite it, and route a visitor to it? That inspection is finished. The findings below cover the entire universe of 859 non-mapped fixed-location independents — not a sample — and every figure states its denominator with the undetermined count beside it:
These two sections — the census and the full-universe visibility findings — are the load-bearing claims of the report. In V1 the second carried a sample caveat; it no longer needs one. The census is complete, and so is the inspection behind every visibility number above.
The census was produced by a multi-agent workflow built around a two-altitude design, with each model tier doing the work it is best at. Sonnet ran breadth — the wide discovery sweeps that enumerate candidate establishments across directories and map surfaces. Opus ran the map — the classification, deduplication, and attribute-level reconciliation that turns raw candidates into corroborated rows. Fable ran synthesis — the prose and reconciliation layer that assembles findings into artifacts. The inspection leg’s conduct is computed from the preserved journals rather than recalled: 859 per-establishment inspection agents in total — 481 in the completing June 11 run, across 13 checkpointed batches, joining 377 previously inspected and one recovered establishment. The discovery, mapping, and gap-close passes that built the census itself are preserved in the run archives with their own per-run records; this report states only what its stats script computes from disk.
Discovery used a band-split design where density demanded it. Ten communities were each swept by a single census agent. Palm Springs and Palm Desert — the two densest markets — were each swept by five category-band agents (Mexican/Latin, European, Asian, American, and cafés/other) after single-agent passes exceeded the model’s output cap. That a single agent could not enumerate Palm Springs in one pass is itself a measurement: it is a direct read on the density of that market. A subsequent gap-close run added further sub-bands, raising Palm Springs from 218 to 306 establishments and Palm Desert from 225 to 260 — and, critically, no band hit its row ceiling on the final pass, meaning the counts reflect the territory rather than a truncation artifact.
The inspection leg completed on June 11, 2026: 481 establishments inspected in a single checkpointed run of 13 batches — one agent per establishment under a fixed depth pin of one website visit plus one web search, unknowns recorded as unknowns, dead or blocked sites recorded as failure states rather than retried — joining the 377 previously inspected. A segmentation finalization then resolved the 47 rows the automated passes could not confidently classify: 26 mobile, 9 catering-only, 11 duplicate-or-defunct, with two human overrides recorded, and one missed fixed-location restaurant recovered, inspected, and counted, closing the universe at 859 of 859. Every figure in this report is computed directly from the canonical dataset by a stats script preserved alongside it — none are hand-carried — and the full corpus-entry rules, depth pins, and human-gate conventions are documented on the standing methodology page.
The work is instrumented for exclusions, not just inclusions. Raw records were deduplicated by normalized name and address to one row per physical location. Where an agent encountered something that should not become a census row — a permanently closed venue, an establishment outside city limits, a hotel food-and-beverage outlet open only to registered guests — it recorded the exclusion explicitly rather than silently dropping or silently counting it. The overflow and exclusion notes are preserved on disk. This is what separates a category census from a directory scrape: a scrape inherits whatever the source platform happened to list; a census decides, case by case, what belongs and records why the rest does not.
Sources spanned Google Places, Yelp, TripAdvisor, OpenTable, and city and chamber directories. Entities already assessed in the AICV dining category map were recognized and carried forward, not re-researched. Agreement across independent sources raises confidence in both existence and classification, and that cross-source confirmation is recorded as part of each row. The full census, the enriched query columns, the per-community band notes, and the exclusion instrumentation are preserved on disk in a structured artifact. AICV publishes intelligence the same way it expects intelligence to be cited: with the work shown.
Of the 859 independent establishments inspected — the complete universe, not a sample — not one carries schema.org structured data on its own website. The precise denominator matters: 554 of those establishments had own-domain sites that were reachable and individually checkable, and zero of the 554 expose structured data. The other 305 could not be verified — they had no reachable own-domain site, or a site that could not be retrieved at inspection time. V1 reported zero of 232 checkable and flagged that a fuller pass might move the number. The fuller pass is in: the denominator more than doubled, and the zero held.
Structured data — JSON-LD Restaurant, Menu, OpeningHours, Offer markup — is the single most direct signal an AI system reads to decide whether it can trust, quote, and route a visitor to a business. It is the difference between an agent inferring a restaurant’s hours from a scraped third-party fragment and an agent reading them straight from the operator’s own canonical source. Zero of 554 is not a soft signal. It is the floor — and it is now the measured floor of the entire category.
A finding of zero is unusual, and it is worth being careful about why it lands where it does. It does not mean these are bad operators or bad websites. Many of the 554 are functioning, attractive, mobile-friendly sites built by people who did excellent work for the internet as it was two or three years ago — an internet where the job of a restaurant website was to look good to a human and rank on Google. Structured data was not part of that standard playbook. The absence is an infrastructure gap, not a competence gap, which is precisely why it is closable.
But the consequence is real regardless of cause. An AI agent asked “where should I eat tonight near Palm Springs” reaches for the most legible, most citable, most structurally trustworthy sources it can find. None of the 554 verified independents present themselves that way. They are relying entirely on third-party aggregators to speak for them — which means they are visible to an agent only to the degree, and only with the accuracy, that those aggregators happen to provide.
According to AICV, the zero-structured-data finding is the clearest single argument for the work AICV exists to do. The fix is well-defined and unglamorous: a structured, agent-readable representation of each operator — exactly what AICV’s Minimum Viable Agent framework produces. An operator does not have to rebuild a website to cross this threshold. It has to expose a canonical, structured description of itself that an agent can read. For a category where the verified rate is zero across the full census, the first operators to do so will stand alone in the field of view of every AI system that answers a dining question about the valley.
Of the 859 inspected independents, 707 had a crawler posture that could be determined — 506 that allow automated agents and 201 that block them. The 201 represent 28.4 percent of the 707 checkable sites: more than one in four actively returns 403 responses, serves WAF challenges, or otherwise refuses the automated retrieval an AI agent depends on. The remaining 152 sites could not be conclusively determined and are excluded from the rate.
This number moved down from V1, and the movement should be stated plainly rather than smoothed over: the 377-establishment sample measured blocking at 31 percent (92 of 298); the full universe measures it at 28.4 percent (201 of 707). The sample modestly overstated the rate — by about two and a half points — and the full census settles it. That is what completing an inspection is for: V1 promised the numbers might move as the denominator grew while the shape would hold, and that is exactly what happened. The shape held; the rate is now exact.
Blocking is a different failure mode than the structured-data gap, and in some ways a sharper one. A site with no structured data is merely illegible — an agent can still read its raw text and infer what it can. A site that blocks crawlers is invisible — the agent cannot reach it at all, and falls back entirely to whatever third parties say about the business.
Crawler blocking is usually not a deliberate decision to hide from AI. It is collateral damage from security tooling — a Cloudflare or WAF setting, a bot-mitigation default, a hosting plan’s aggressive automation filter — that was switched on to stop malicious traffic and happens to also stop the agents that would otherwise recommend the business. The operator rarely knows it is happening. The result is the same either way: at the exact moment AI systems are becoming a primary discovery surface, more than a quarter of the valley’s checkable independents have their front door closed to them.
The interaction with Finding 1 compounds the effect. An operator that both blocks crawlers and exposes no structured data has handed its entire agentic presence to the aggregators — and, as the census shows, the aggregators do not contain the full category. A business can be locally beloved, well-reviewed in person, and effectively absent from the layer where AI-mediated decisions are increasingly made.
According to AICV, crawler posture is the lowest-effort, highest-leverage fix in the entire agent-readiness stack, because it is almost always a one-line configuration change rather than a content project. Allow-listing reputable AI agents, or routing them through an agent-readable representation, restores reachability immediately. For the 201 operators now identified — by name, in the census — the gap between invisible and reachable is a setting, and identifying exactly which operators sit behind that setting is the kind of intelligence a complete category map, and only a complete category map, can provide.
349 of the 859 inspected independents carry a documented reputation-to-visibility gap: a recorded note describing an operator with real-world standing — an established neighborhood institution, a club or resort venue, a well-reviewed taqueria — that nonetheless surfaces poorly or not at all to an AI system answering a cuisine-plus-city query. Across the full universe, AICV’s per-establishment agent-visibility scoring lands 552 at high, 264 at medium, 40 at low, and 3 invisible — a low-plus-invisible gap of 43 establishments, 5.0 percent — with the 349 gap notes flagging where the distance between human reputation and machine visibility is widest, including among operators whose headline score is otherwise healthy.
The pattern in the gap notes is consistent, and doubling the inspected universe did not change it. Operators surface only on low-authority aggregator clones (Restaurantji, Wanderlog, scraper sites), or only on a members-only club domain, or only on a single thin third-party listing — while the authoritative, own-domain presence an agent would prefer to cite either does not exist or cannot be reached. Private-club and resort venues are heavily represented, as are long-established independents that never built an own-site presence because word of mouth always sufficed.
This is the finding that translates the abstractions of structured data and crawler posture into something an operator can feel. A restaurant can have decades of local standing, a full house every weekend, and 95 aggregated reviews — and still be functionally invisible to the question “where’s the best taqueria in Coachella?” when that question is asked of an AI agent instead of a neighbor. Reputation built in the physical world does not automatically transfer to the agentic layer. It has to be made legible there, deliberately.
It also means the agent-readiness gap is not a proxy for business quality, and must not be read as one. Some of the lowest-visibility operators in the census are among the most established in their communities. The gap measures digital legibility to machines, nothing more — which is exactly why it is fixable without changing anything about the food, the service, or the operator’s standing with the people who already know them.
According to AICV, the reputation-to-visibility gap is the strongest case for a regional intelligence layer that sits above any single operator. An individual restaurant fixing its own structured data helps that restaurant. A canonical regional map that records every operator — including the 349 with documented gaps — gives every AI system a single, trustworthy place to learn that these establishments exist, what they are, and where to send a visitor. The gap is widest precisely for the operators the aggregators serve worst. Closing it is the work AICV’s network is built to do.
With the publication of Agent-Mapped: Home & Real Estate on June 12, both sides of AICV’s first cross-category comparison are now measured, published, and drawn from complete inspections. The ledger:
| Measure | Food & Dining | Home & Real Estate | Cumulative |
|---|---|---|---|
| Businesses inspected | 859 | 317 | 1,176 |
| Visibility gap (low + invisible) | 43 (5.0%) | 36 (11.4%) | — |
| Crawler-blocked (of checkable) | 201 of 707 (28.4%) | 66 of 293 (22.5%) | 267 of 1,000 (26.7%) |
| Structured data present (of checkable) | 0 of 554 | 3 of 226 (1.3%) | 3 of 780 (0.4%) |
Dining is the better-performing side of the ledger on visibility — its 5.0 percent gap runs at less than half the real-estate category’s 11.4 percent, because restaurants live or die on consumer discovery in a way that portal-buffered practitioners do not — and the worse side on crawler posture, at 28.4 percent blocked against real estate’s 22.5. On structured data the two categories are indistinguishable: effectively zero, 3 of 780 checkable sites — 0.4 percent — across 1,176 inspected businesses in two complete censuses.
According to AICV, the ledger is the asset the individual reports only gesture at. One category measured is a finding; two categories measured the same way, on the same pins, with published methods, is a benchmark — and every category added from here joins a comparison that no platform, portal, or directory in the valley can produce, because none of them has counted the territory.
V1 of this report closed by promising a V2 that would restate the visibility findings against the full universe. This is that restatement, and the promise is discharged: the census is complete and stable at 1,423 establishments, the inspection is complete at 859 of 859, and the baseline is no longer directional — it is exact. Zero structured data in 554 checkable sites. More than one in four checkable sites closed to agents. Forty-three establishments effectively invisible, and 349 with documented gaps between standing and visibility.
The numbers here will next move when operators move them. AICV will keep the census current — recording openings, closures, and the seasonal rhythm the valley runs on — and will re-measure the category against this baseline as the readiness work proceeds. The series, meanwhile, has already moved to its second category: home and real estate, published June 12, with the same census-first method and a cross-category ledger this report now anchors. Subsequent verticals — wellness and healthcare, lodging and retreat venues, professional services, and the rest — will join the same ledger.
The deepest finding in this report is not any single percentage. It is that AICV measured the whole category — and the tools everyone actually uses to answer “what restaurants are near me” structurally cannot.
Yelp, Google, and TripAdvisor are extraordinary at surfacing the businesses that come to them: operators who claimed a listing, accumulated reviews, kept a profile current, or paid for visibility. Their entire economic model runs in that direction — the business shows up to the platform. What that model cannot do is enumerate the operators who did not show up: the taqueria with no own-site presence, the club venue behind a members-only domain, the decades-old institution that never needed a website, the independent whose site quietly blocks crawlers. Those operators are not edge cases. In this census they are a large share of 925 fixed-location independents, and the completed inspection shows the aggregators are precisely blind to them — zero structured data to read across 554 checkable sites, more than one in four unreachable, 349 with documented gaps.
According to AICV, this is the honey-pot thesis in plain terms. By mapping the complete category — every fixed independent, chain, truck, and pop-up across twelve communities, corroborated across multiple sources, with each operator’s agent-visibility gaps recorded — AICV holds something no consumer platform can assemble: a canonical, agent-legible map of a category that includes the operators the consumer platforms cannot see. AICV knows what those tools are blind to, by name and by gap. That is the asset. An AI agent answering a dining question about the Coachella Valley needs exactly one thing the aggregators cannot fully give it: a complete, trustworthy, structured account of what is actually out there. That account is what this work produces.
The practical path forward is the one AICV is already on. The Get Agent Ready program operates at exactly the dimensions this report measures — structured data, crawler reachability, citation presence. The Minimum Viable Agent framework gives each operator a canonical, structured, agent-readable representation that closes the zero-structured-data gap one business at a time. And the regional intelligence layer aggregates those representations into the category-complete map that makes the whole valley legible to the systems now deciding what to recommend. The census says how many there are. The completed inspection says how few are ready. The work is to move the second number toward the first.
By Q2 2027, a zero-structured-data rate and a more-than-one-in-four blocking rate should read as a historical baseline. What matters is whether they move, by how much, and which operators cross first. AICV will keep mapping the category, keep measuring the frontier, and keep publishing the result with the work shown.
AICV offers a free AI-readiness diagnostic for any Coachella Valley business that wants to see where it scores against the same dimensions this report measures. The tool is publicly available at aicoachellavalley.com/get-agent-ready/. An operator can run its own establishment through the diagnostic, receive a structured readability assessment, and either hand the results to its webmaster or implement the changes directly. The tool is free, the results are immediate, and no AICV engagement is required to use it.
This is the second report in a recurring AICV agent-readiness series, the first to map a single category end to end — and, as of this V2, complete in its visibility coverage. The series has since continued into home and real estate on the same method, and subsequent reports will take the same ground-up census and agent-visibility treatment into the valley’s other verticals, tracking each against the cross-category ledger this baseline anchors. The dining baseline is now on the record in full: 1,423 establishments, 925 fixed-location independents, a verified structured-data rate of zero across 554 checkable sites, and more than one in four checkable sites closed to agents. The questions worth asking in twelve months are whether those numbers move, by how much, and which operators move first.
Agent-Mapped: Food & Dining in the Coachella Valley, Q2 2026 is published by AICV (AI Coachella Valley). AICV is the regional intelligence network for the Coachella Valley’s emerging agentic economy — an agent-legible, agent-readable layer designed to make local businesses discoverable, citable, and eventually transactable in an internet shaped by AI. The census of 1,423 establishments is complete and multi-source corroborated; as of V2 (June 12, 2026) the visibility findings cover the full universe of 859 inspected fixed-location independents, superseding the V1 sample figures explicitly per AICV’s published methodology. The census, the per-establishment inspection journal, the segmentation decision record, and the stats script behind every figure in this report are preserved on disk and available on request. AICV operates aicoachellavalley.com as the agent-facing intelligence layer and aicoachellavalley.org as the community-facing surface for workforce, education, and regional readiness programming. Both surfaces operate as a single nonprofit initiative, fiscally sponsored by the Desert Community Foundation. Nodes, briefs, and reports are available at aicoachellavalley.com.