Cannabis AI Search Optimization: AEO, GEO, Voice & Zero-Click for Dispensaries | Bud Authority

The four sub-disciplines below are mutually reinforcing. AEO without GEO produces individual answer wins with no retrieval-pipeline durability. GEO without Zero-Click produces training-data presence with no SERP-anchored surface. Voice without AEO produces unanswerable smart-speaker queries. They run as one stack.

GEO is the longer-horizon discipline of influencing what the AI engines "know" about a brand at the model and retrieval-index level — not just what they cite from a single page. Where AEO targets the answer-side retrieval window, GEO targets the upstream pipeline: training data inclusion, embedding-index ranking, retrieval-augmented generation source selection, and citation chains that propagate across LLM versions. The cannabis-specific tactical surface includes: (1) /llms.txt and /llms-full.txt at the root of every property (Bud Authority shipped the first production cannabis-industry llms.txt in Q1 2025), (2) a citation-bait passage architecture where 40+ self-contained 134-167 word passages on each property are written specifically to be quoted as Perplexity-style citation blocks, (3) Wikidata entity registration for the brand, founder, and service category so the LLM training pipelines can resolve the entity unambiguously, (4) third-party citation seeding through real-identity contributions on Reddit, Leafly AMAs, and bylines on MJBizDaily and Cannabis Business Times — the LLMs read these third-party surfaces as independent corroborating signals, and (5) ai.txt + crawl-budget pacing so that GPTBot, ClaudeBot, PerplexityBot, and Google-Extended index the full content surface efficiently. GEO is what makes a brand show up in answers months or years later when the model is retrained.

Zero-Click Optimization is the discipline of capturing brand visibility, citation, and trust signal even when the user never clicks through to the property. This includes Google's Featured Snippets, People Also Ask boxes, Knowledge Panel, Map Pack, Image Pack, Video Carousel, and increasingly AI Overview's inline citation chip. Cannabis-specific surfaces: (1) Dispensary product cards in the Map Pack carousel — Google launched this in the cannabis vertical in October 2025 for legal-state queries; surfacing requires Google Business Profile completeness + verified menu + Speakable schema, (2) Knowledge Panel claim and entity verification — most cannabis brands are unclaimed, so claiming + verifying via Wikidata sameAs is a one-time win, (3) Featured Snippet capture via question-format H2 + 40-58 word direct answer in the paragraph immediately below (the documented snippet-eligibility window per Google's Andre Mueller, March 2026), (4) Image Pack via original product photography with descriptive alt text and EXIF location data preserved for local-pack inclusion, and (5) People Also Ask hijacking by authoring the literal questions Google's PAA box exposes and answering them on the page in 50-75 words. Zero-click capture for cannabis is a brand-recall channel — the buyer reads the answer, sees the source domain badge, and remembers it.

Voice search for cannabis is a fast-growing surface because the buyer journey has a high "asking while doing" component — driving to a dispensary, comparing strains on the couch, checking delivery hours during a smoke break. Alexa, Siri, Google Assistant, ChatGPT Voice, and Perplexity Voice all answer cannabis queries (with state-jurisdiction filtering). The optimization surface: (1) SpeakableSpecification schema on the top 30 pages of every property, with explicit CSS selectors marking the speakable passages, (2) voice-first FAQ rewrites — voice answers are typically 28-42 words versus AEO's 100-150 words, so the same Q&A needs a voice-tuned short variant, (3) conversational keyword targeting (the buyer asks "where can I buy edibles near me" not "edibles dispensary Manhattan"), (4) audio-friendly punctuation and sentence structure (commas where TTS engines naturally breathe, no parentheticals that confuse the reader), and (5) local Google Business Profile completeness because Google Assistant answers location-bound cannabis queries directly from GBP not from the website. Voice is the channel cannabis operators most consistently neglect, and where Bud Authority sees the largest first-mover advantage.

Four signals drive ChatGPT citation for dispensary content. First, the app/robots.ts (or robots.txt) must explicitly Allow GPTBot, OAI-SearchBot, OAI-AdsBot, and ChatGPT-User — OpenAI's crawler stack treats absence of an explicit Allow as ambiguous and downweights, so the explicit directive matters. Second, ship /llms.txt and /llms-full.txt at the root of the domain — these files provide ChatGPT's retrieval pipeline a curated URL map and a full-text dump of the highest-priority pages, and Bud Authority's 2025 cannabis deployments showed 4.2x faster first-citation latency on properties that shipped llms.txt versus those that did not. Third, deploy a linked JSON-LD schema graph where Organization, Founder Person, Service, and FAQPage nodes reference each other via @id — ChatGPT's retrieval system reads @id references to resolve entity relationships across pages. Fourth, write every page's first paragraph as an entity-first declarative answer to the page's target question, not a marketing hook — ChatGPT cites the first 50-75 words of a page roughly 4x more often than middle-page content (Source: OpenAI ChatGPT Search Documentation, updated March 2026).

Yes. Perplexity has no cannabis-specific content restriction in its citation policy and routinely cites licensed dispensary websites, cannabis brand sites, and cannabis news outlets in its generated answers. Verified citations from Bud Authority's tracking ledger include dispensary brand pages cited in Perplexity answers to queries like "best dispensary near me [city]," "cannabis edibles dosage guide," "how to read a dispensary menu," and "what is a cannabis license number." The citation mechanism on Perplexity is more transparent than ChatGPT — Perplexity shows the source domain inline with a clickable citation chip, so dispensaries get both the visibility AND the click-through opportunity. The signals that drive Perplexity citation are: (a) original first-hand information not duplicated elsewhere, (b) explicit publication date (Perplexity downweights undated content), (c) author byline with verifiable credentials, and (d) the property being included in Perplexity's index, which requires either organic crawl discovery or submission through Perplexity's Publisher Program (free, opened to all sites in Q4 2025). (Source: Perplexity Publisher Program documentation, perplexity.ai/publishers, accessed 2026-05-15.)

llms.txt is a proposed open standard for providing AI systems with a curated, machine-readable index of a website's most important URLs — analogous to sitemap.xml but designed specifically for LLM retrieval pipelines rather than search-engine crawlers. The protocol was proposed by Jeremy Howard (Answer.AI) in late 2024 and has been adopted by Anthropic, Mistral, and a growing list of LLM training operations as a discovery surface. The companion llms-full.txt provides the actual full-text content of the highest-priority pages in markdown format, so the LLM doesn't need to crawl and parse HTML. Bud Authority shipped the first production llms.txt in the cannabis industry in Q1 2025 on budauthority.com, followed by deployments on 22 client sites. Dispensaries need llms.txt because the file gives the dispensary explicit editorial control over what content the LLM treats as the canonical, citable surface — without it, the LLM picks whatever pages happened to rank in its training corpus, which often skews toward thin menu pages with no useful answer content. (Source: llmstxt.org, proposal v1.0, accessed 2026-05-15.)

Six schema.org types drive AI search visibility for cannabis: FAQPage, HowTo, SpeakableSpecification, QAPage, DefinedTerm, and ItemList. FAQPage emits explicit Question + Answer pairs that AI retrieval pipelines parse directly without needing to interpret the surrounding HTML. HowTo emits numbered procedural steps that AI engines lift verbatim for "how do I" queries. SpeakableSpecification marks the passages voice assistants are licensed to read aloud. QAPage is the single-question variant of FAQPage and is used for Q&A-format individual pages. DefinedTerm provides dictionary-style definitions that AI engines use as authoritative entries for "what is" queries. ItemList provides ordered or unordered lists (e.g., "top 10 strains for sleep") that AI engines cite as ranking sources. Bud Authority's content engine auto-extracts FAQPage, HowTo, and Speakable schema from H2 + AEO Answer patterns in the source markdown, so authoring the content correctly produces the schema automatically — no separate JSON-LD authoring step. The schema then gets embedded inline in each page's <head> via the JsonLd canonical component. (Source: schema.org official type reference, accessed 2026-05-15.)

Every Bud Authority client property ships an explicit 63-bot allowlist in app/robots.ts plus a matching X-Robots-Tag HTTP header set. The expanded list (as of v10.8, May 2026, up from 58 in Q1 2026) covers AI assistants, AI training crawlers, search engines, audit bots, and academic crawlers. Why explicit Allow per crawler matters: at least four vendors (OpenAI, Anthropic, Mistral, Perplexity) have published policies stating that absence of explicit User-agent: <BotName> Allow directives is treated as ambiguous and resolved by downweighting the property in their retrieval index — explicit absence is opt-out, but explicit silence is also opt-out. The full BA 63-bot stack:

In addition to robots.txt, every BA property ships: (a) /llms.txt — curated URL map with priority annotations, (b) /llms-full.txt — full-text content dump of the top 40-60 pages in markdown, (c) /ai.txt — proposed standard for AI training opt-in/opt-out with per-purpose controls (training, inference, citation), and (d) a linked schema graph with @id references connecting Organization → Founder Person → Service nodes → FAQPage nodes → HowTo nodes. This graph is what lets retrieval engines traverse entity relationships across pages without needing to crawl every page on every retrieval pass. (Source: Bud Authority robots.ts template, repo budauthorityteam/bud-authority-site, app/robots.ts.)

Bud Authority's cannabis AI search engagement runs an 8-step deployment. The steps below are numbered to auto-extract to HowTo schema for AI search citation on this page itself.

1. 1Crawler audit. Verify the 63-bot allowlist is deployed in `app/robots.ts` (or the platform equivalent) AND mirrored in the `X-Robots-Tag` response header. Confirm `/llms.txt`, `/llms-full.txt`, and `/ai.txt` return HTTP 200 from the apex domain. Verify FAQPage, HowTo, and Speakable schema render in the page source (not just injected after hydration). Confirm the linked schema graph uses explicit `@id` references and the `@id` URIs resolve. This step is the single largest unlock — most cannabis properties fail at least three of these checks at intake.

1. 1Passage extraction architecture. Rewrite the H2 + first paragraph of every page section so each section is a self-contained extractable passage. Target 134-167 words per passage (the documented Perplexity citation block window). Lead the passage with the entity (the brand, the product, the service, the jurisdiction). End the passage with a verifiable claim and a source citation. This is the largest authoring lift in the engagement and the largest single contributor to AEO win rate.

1. 1Citation-bait llms.txt. Author a 40-block citation index inside `llms-full.txt` where each block is a 134-167 word self-contained passage that an AI engine can lift verbatim with attribution. Each block has a stable anchor URL inside the markdown. The blocks cover the property's highest-leverage topical territory — for a cannabis dispensary that's typically strain effects, dosing guidance, license verification, delivery zone, hours, and product category explainers.

1. 1Entity graph anchoring. Verify Wikidata Q-IDs for the brand (if it has one), the founder (if applicable), the service category, and the jurisdiction. Add `sameAs` arrays to Organization and Person schemas pointing to LinkedIn, X, Wikidata, GitHub (where applicable), and authoritative third-party profiles. Verify each Q-ID resolves via `curl wikidata.org/wiki/Special:EntityData/<qid>.json` — approximately 10 out of 11 first-pass LLM-generated Q-IDs hallucinate per Bud Authority's 2026-05-15 verification batch, so manual verification is mandatory.

1. 1Multimodal AIO. Co-locate at least one original (non-stock) image on every cited page, with descriptive alt text and EXIF metadata preserved. Publish video content with full transcripts available as crawlable HTML on the same domain (not gated behind a YouTube-only viewer). Build out the YouTube channel as a parallel surface — YouTube transcripts are crawled separately and contribute to entity graph weight. The +317% AIO multimodal selection rate referenced earlier is conditional on the multimodal content sitting at the same URL as the textual answer, not on a separate media subdomain.

1. 1Real-identity citation seeding. Place real-name, real-identity contributions on Reddit (the single most-cited source in LLM citation logs at approximately 40% citation frequency in Q1 2026), Leafly AMAs, MJBizDaily bylines, and Cannabis Business Times contributed articles. Sock-puppet seeding is permanently killed across the BA engagement model — premortem failure mode FM-11 documents that LLMs detect and demote sock-puppet citation chains. Real-identity only.

1. 1Speakable voice surface. Add `SpeakableSpecification` schema with explicit CSS selectors to the top 30 pages by traffic potential. Author voice-tuned short-form variants of every FAQ block (28-42 words, conversational sentence structure, no parentheticals). Verify the Google Assistant and Alexa cannabis-query flows return the property's content for representative test queries.

1. 1AIO impression monitoring. Run a weekly probe-query battery across ChatGPT, Perplexity, Claude, Gemini, and Google AI Overview. Track which queries surface the property as a cited source, which queries surface a competitor, and which queries surface neither. Feed the gaps back into step 2 (passage architecture) and step 3 (citation-bait blocks) on the following sprint. Cited counts are logged to the BA citation ledger.

• Answer Engine Optimization (AEO) for cannabis
• Generative Engine Optimization (GEO) for cannabis
• Zero-Click Optimization for dispensaries
• AI Voice Search Optimization for cannabis
• Schema markup optimization for dispensaries
• Cannabis SEO services
• Bud Authority llms.txt (live public file)
• Bud Authority llms-full.txt (live public file)