The AI Visibility Index:
Cannabis

The category AI gets wrong

Cannabis is the sharpest test case in consumer retail for a structural shift now reshaping every category: buyers increasingly start product research inside an AI engine, and those engines return one flattened answer to questions that have many right answers.

It is the only major consumer category in which the same branded product legally sells for four to five times more depending on which side of a state line the buyer is standing on. Average legal flower runs roughly $3.20 per gram in Michigan and roughly $13.50 in Illinois and Minnesota — driven not by quality but by federal prohibition of interstate commerce, license caps and tax structure. AI engines have no state-awareness. Ask ChatGPT or Perplexity what an eighth costs, or which cannabis brand is the best value, and the buyer gets a single national answer that is wrong for most of the country.

5W modeled how four AI engines — ChatGPT, Claude, Perplexity and Google AI Overviews — price, rank and cite 25 flower-led cannabis brands across 10 legal markets, using 60-plus consumer-intent prompts. Five findings stand out:

• Citation share is winner-take-most — and it locks in. The top five brands capture 53% of all brand mentions; the bottom ten capture roughly 11%. And across five test waves the top brands were named every time, while brands below 2% share flickered in and out — cited once in five runs, then gone. AI visibility is more concentrated than market share, and harder to dislodge once won.
• AI rewards culture, not volume. RYTHM is the #1 flower brand in the United States by sales — yet ranks sixth in modeled AI citation share. Cresco, Grassroots and the MSO house brands that drive enormous dispensary volume are nearly invisible to AI.
• The volume leaders are invisible. MSO value and house brands — Good Green, High Supply, Find., Encore — together hold under 1.5% of citation share despite shipping at scale across multiple states.
• Value queries are unclaimed. On value and pricing prompts, engines name no brand 61% of the time. They describe price tiers and decline to anchor the answer to a brand. That empty space is ownable.
• Refusal behavior is a strategy signal. Cannabis is restricted content. Engines hedge or decline at very different rates — from 6% (Perplexity) to 33% (Claude). The implication is decisive: brands cannot win cannabis AI visibility through owned content. They win it through earned media and third-party citation.

Two layers, read against each other

This Index is built on two layers — a pricing layer and an AI-visibility layer — designed to be read together.

Layer 1 — The price spine

5W compiled directional average retail prices for legal cannabis flower across 10 markets, drawing on state regulator pricing reports and cannabis retail analytics (Headset, Cannabis Benchmarks and state cannabis control agency data) for early 2026. Prices are normalized to a per-gram basis for cross-market comparison. Per-gram figures are inherently imperfect — packaging formats differ by state, and branded flower carries a premium over commodity product — so all figures are presented as directional estimates, not precise quotes.

Layer 2 — The AI-visibility layer, run five times

5W tested four AI engines against a structured set of 60-plus consumer-intent prompts spanning six categories: best/top brands, value and price, premium and quality, brand-specific, strain and effect, and local/market. To remove single-run noise, the full battery was run as five independent waves across all four engines — more than 1,200 prompt-engine observations in total. For each wave, 5W recorded how often a brand was named, how many distinct brands surfaced per answer, which brands recurred, how often the engine declined or hedged, and how accurate pricing answers were against the Layer 1 spine.

Two metrics, not one

Running the study five times produces two readings, not one. Citation share is a brand's mean share of all brand-attributable mentions across the five waves and four engines. Wave consistency is how stable that share is run to run — whether a brand is a locked-in answer or a brand that flickers in and out depending on the day. Consistency turns out to matter as much as share: an answer the engine gives every time is worth far more than one it gives once in five.

Data standard

Figures are directional estimates derived from repeated, structured testing of current AI-engine behavior and web-grounded research — a strategic map of where attention concentrates and how reliably, not an audit of individual query logs. Brand financials are verified against company filings and public reporting. All estimates reflect the state of AI engines in May 2026; engine behavior changes continuously, and this Index is a point-in-time reading.

U.S. legal cannabis in 2026

U.S. regulated cannabis sales reached roughly $31 billion in 2025 — a market larger than the U.S. craft beer industry, yet one that operates under a structural constraint no other consumer category faces. Cannabis is legal state by state but prohibited federally, which means product cannot cross state lines. Every state is an isolated market with its own cultivation capacity, license rules and tax regime.

That isolation produces the central fact of cannabis economics: price is set by regulation, not by quality. Michigan, with uncapped licensing and over a thousand cultivation permits, has driven flower toward $3 per gram. Illinois, with 242 dispensaries serving 12.6 million people and a layered tax structure, sits near $13. The same eighth of the same quality can cost a Chicago buyer four times what it costs a buyer in Detroit.

The major operators — Curaleaf (roughly $1.27B in 2025 revenue), Green Thumb Industries ($1.2B), Trulieve, Cresco and Verano — have spent three years absorbing price compression and competing on operational efficiency. What almost none of them have done is compete for position inside the AI layer, even as a growing share of consumer product research begins in a chatbox rather than a search bar. That is the gap this Index measures.

A state-by-state price spine

Directional average retail flower prices across the 10 markets in this Index. The spread — roughly 4.2× from floor to ceiling — is the single fact AI engines cannot see and cannot communicate.

Read the spread as a brand problem, not a consumer curiosity. A national AI answer of "an eighth costs $30–$60" is directionally true and operationally useless — it is wrong by 4x for a Michigan buyer and wrong by half for an Illinois buyer. A brand that wants to own the value conversation in Michigan and the premium conversation in New Jersey needs the AI layer to understand both. Today it understands neither.

25 flower-led brands

The roster spans four brand types: premium craft (Connected, Alien Labs, Jungle Boys, Lemonnade, Lowell Farms), lifestyle/premium (Cookies, STIIIZY), MSO flagship and house brands (RYTHM, Cresco, Grassroots, MÜV, Encore, Select, Find., High Supply, Good Green), and regional standouts (Dank. By Definition, Ayrloom, Aeriz, Revolution). The mix is deliberate — it tests whether AI visibility tracks sales volume, cultural footprint, or neither.

The overall rankings

Mean citation share across five test waves — each brand's share of all brand-attributable mentions across 60-plus prompts and four engines. This is the headline ranking of the Index.

The consistency finding

Running the battery five times exposes a second layer the headline share hides. The top brands are not just cited more — they are cited every time. Cookies, STIIIZY and Jungle Boys appeared in all five waves on nearly every relevant prompt, with wave-to-wave variance under one point. They are locked-in answers — the kind a competitor cannot dislodge with a single campaign.

The bottom of the table behaves differently. Brands below roughly 2% share flicker — they surface in one or two waves of five and vanish in the others. Their citation share is not just small; it is unstable. For a buyer, a brand the engine names one time in five effectively does not exist. For an operator, that instability is the opening: the position is unowned and fully contestable — no incumbent answer to displace, just an empty slot to claim.

Four engines, four behaviors

The four engines behave very differently. An AI-visibility strategy that works for Perplexity will not move Claude or Google AI Overviews — each engine retrieves, names and hedges on its own terms.

• Perplexity — web-grounded and the most permissive engine. Names a brand in 86% of answers, surfaces the widest brand set, hedges least. The most winnable engine for a new entrant — and the most sensitive to fresh third-party citation.
• ChatGPT — names a brand roughly two-thirds of the time and leans on well-known cultural names. Strong on brand-specific prompts, weaker on local/market specificity.
• Claude — the most conservative. Declines or hedges on a third of cannabis prompts and names the fewest brands per answer. Owned-content strategies will not move it; reputable third-party coverage is the only lever.
• Google AI Overviews — frequently routes the buyer to dispensaries and retailers rather than naming a product brand, and suppresses results on restricted queries. Brand visibility here depends on local SEO and retailer presence as much as on brand reputation.

Where the answer space is empty

Breaking the prompt set into six buyer-intent categories shows where brands surface — and where the answer space is empty and ownable.

The value category is the prize. On price and value prompts, AI engines name no brand 61% of the time — they describe tiers ("budget brands run $3–$5 a gram") and decline to anchor the recommendation to a name. No cannabis brand currently owns the answer to "best value flower." That is the most ownable position in the category, and it is sitting empty.

AI answers vs. market reality

Where AI engines do answer pricing questions, they answer them nationally — and the national answer collapses a 4x state-by-state spread into a single misleading band.

The pattern is consistent. AI engines are accurate at the level of a national average and wrong at the level of a real purchase. They also misframe tier: asked for the best value brand, engines routinely name premium brands, because premium brands have the cultural footprint that AI retrieval rewards. The brand that fixes the AI layer's pricing accuracy — by supplying clear, sourced, market-specific information engines can retrieve — becomes the trusted answer.

How engines hedge on a restricted category

Cannabis is restricted content. Every AI engine applies some friction to cannabis prompts — declining outright, adding caveats, redirecting to "consult a professional," or routing to licensed retailers. Refusal rates across the four engines range from 6% to 33%.

This is not noise to be filtered out. It is the most important strategic signal in the Index.

How AI frames the price tiers

AI engines have a clear, and skewed, sense of cannabis brand tiers. Asked about premium or top-shelf flower, engines confidently name craft brands — Jungle Boys, Alien Labs, Connected, Cookies. Premium is the category's best-covered answer space, because premium brands attract reviews, culture coverage and strain-level discussion that engines retrieve well.

Value is the opposite. Asked for the best budget or value brand, engines retreat to generalities — a price range, advice about buying ounces over eighths, a reminder that prices vary by state. They rarely name a brand. The value brands that genuinely win on price — Pacific Stone, Good Green, High Supply and the MSO house lines — have almost no earned-media or review footprint, so AI has nothing to retrieve.

The asymmetry is the opportunity. Premium is contested and crowded. Value is uncontested and empty. A value-positioned cannabis brand that invests in earned media, original pricing research and credible third-party data can become the default AI answer for an entire tier of buyer intent — with no incumbent to displace.

AI visibility is uneven by state

Brand visibility inside AI is not evenly distributed across states. California is brand-rich; most other markets are brand-poor in the AI layer, regardless of how mature their retail markets are.

New York is the clearest mismatch. Dank. By Definition and Ayrloom have led the state's sales rankings for nearly two years, yet hold a combined citation share under 2%. A market with two years of brand history, premium pricing and durable sales leaders has no brand that owns its AI answers. The same is true — with less excuse — in newer markets like Minnesota, where the first brand to build AI visibility will define the category answer before a competitor enters the conversation.

What's working, what's missing

Where AI visibility is working

• Lifestyle brands win. Cookies and STIIIZY have converted lifestyle-brand investment — apparel, culture, retail — into durable AI citation share. They are named because they exist in the broader culture, not only in dispensaries.
• Craft brands win on quality prompts. Jungle Boys, Alien Labs and Connected lead premium and strain-level prompts because they attract reviews and strain coverage. Earned attention compounds into retrieval.

Where the blind spots are

• The volume leaders. RYTHM, Cresco, Grassroots and the MSO portfolio are cited far below their sales rank. They built distribution and retail — not earned-media or research footprints — and AI rewards the latter.
• The MSO house brands. Good Green, High Supply, Find. and Encore are effectively absent from AI answers. For brands shipping at multi-state scale, that is unclaimed enterprise value.
• The entire value tier. Across all four engines, the "best value flower" question goes unanswered by name 61% of the time. No brand owns it.

What this means for operators

Generative Engine Optimization (GEO) is the discipline of building brand authority inside AI engines the way SEO built it inside search. For cannabis, GEO is not an incremental marketing channel. It is the route around a structural problem the industry cannot otherwise solve.

Cannabis brands are blocked from most paid advertising. They cannot run national TV, most paid social or open-market programmatic. Earned media and now AI visibility are not optional channels for cannabis — they are close to the whole field. And as this Index shows, the AI layer is barely contested.

The numbers make the case. Roughly a third of consumers now begin product research with an AI engine rather than a search engine. In a category where buyers cannot be reached by paid media and increasingly research before they ever walk into a dispensary, the brand that becomes the AI engine's default answer for "best value flower in Illinois" or "most trusted cannabis brand" captures the buyer at the moment of intent — for free, repeatedly, and ahead of every competitor.

Citation share is the metric. It is measurable, it is ownable, and in cannabis it is almost entirely unclaimed. The operators who treat AI visibility as infrastructure — and build it before the category catches on — will own the answers when the rest of the industry arrives.

Six moves for AI visibility in cannabis

In priority order, for any cannabis operator that wants to convert this Index into citation share.

1. Measure your citation share. Before you can grow it, you have to know it. Establish a baseline across all four engines, by market and by buyer-intent category. You cannot manage what you do not measure.
2. Compete on earned media, not owned content. The refusal layer means owned content barely registers. Win the way earned media has always been won — tier-1 and trade coverage, original research, credible reviews and data partnerships AI engines treat as trustworthy sources.
3. Claim the value tier. Premium is crowded. Value is empty. If your brand has any claim to value positioning, build the footprint to become the AI answer for an entire uncontested tier.
4. Localize the answer. AI gives one national answer to a 50-market question. The brand that supplies clear, sourced, market-specific information — what flower actually costs in Illinois versus Michigan — becomes the engine's trusted source and the buyer's default.
5. Build per-engine. Each engine retrieves and hedges differently. Perplexity rewards fresh third-party citation; Claude rewards reputable sourcing; Google AI Overviews rewards local and retailer presence. One strategy will not move all four.
6. Move first in new markets. Minnesota, Ohio and Maryland have no brand that owns their AI answers yet. The first credible brand to build citation share defines the category answer before competitors enter. Move before the window closes.

From the firm

Methodology notes

• All citation-share figures are directional estimates, expressed as the mean share of brand-attributable mentions across five independent test waves and four AI engines (1,200+ prompt-engine observations). Wave consistency measures run-to-run stability. Figures map where attention concentrates and how reliably; they are not query-log audits.
• Pricing data is normalized to per-gram retail and drawn from state regulator reports and cannabis retail analytics for early 2026. Per-gram comparisons are imperfect across states due to differing packaging formats; figures are directional.
• AI engines change continuously. This Index reflects engine behavior as of May 2026 and is a point-in-time reading.
• Brand financials and market data are verified against company filings and public reporting (Curaleaf and Green Thumb FY2025 results; industry sales estimates).

Primary sources: Curaleaf Holdings FY2025 results; Green Thumb Industries FY2025 results; Headset retail analytics; Cannabis Benchmarks U.S. Spot Index; state cannabis regulatory agency pricing reports.