ColaDigital Study · Original Research · November 2026
We asked ChatGPT, Google AI Mode, Gemini, Perplexity, and Claude the questions cannabis shoppers ask about 40 real dispensaries in 20 US cities. Then we checked more than 1,200 answers against state license records, dispensary websites, and Google listings.
By Vee Popat, founder of ColaDigital · Published November 2026 · Answers collected November 5 to 10, 2026Vee founded ColaDigital, a cannabis marketing agency, and has more than two decades of search marketing behind him.
The Short Answer (November 2026): AI assistants often got dispensaries wrong. The city's top-rated dispensary got a clear "go here" in only 45 of 100 answers, price answers mixed in old prices and shops in the wrong state, and about 1 in 10 answers repeated a damaging claim. Most of the errors we could trace started with information a dispensary can fix.
Study at a Glance
What we found
More cannabis shoppers now ask AI platforms where to buy weed. We ran a test asking five major AI tools (ChatGPT, Google AI Mode, Gemini, Perplexity, and Claude) the exact questions a customer uses when researching 40 active dispensaries. Then we checked every response against official state licensing databases, the dispensaries' web pages, and Google Business profiles.
The results were troubling. AI assistants regularly shared inaccurate details, outdated information, or facts belonging to completely different businesses. The top-rated dispensary in a city got a direct "go here" recommendation in only 45 of 100 responses that asked if a shopper should go there. And when shoppers asked who has the cheapest weed or the best deals, the answers mixed in old prices, shops in the wrong state, and the wrong tax.
For dispensary operators, where these errors come from matters most.
Most of the errors we could trace came from sources you can control: your own website, Google Maps records, unclaimed profiles on Leafly, Weedmaps, or Yelp, and outdated feedback scraped onto third-party directories.
That's why this study puts the fixes first.
| AI assistant | Right location details, with a link (higher is better) | Wrong location fact (lower is better) | Complaint answers that leaned negative (lower is better) | Changed its “best dispensary” pick (lower is better) |
|---|---|---|---|---|
| Google AI Mode | 38 of 40 | 1 of 40 | 29 of 40 | 5 of 20 |
| ChatGPT | 38 of 40 | 2 of 40 | 11 of 40 | 7 of 20 |
| Gemini | 1* of 40 | 2 of 40 | 18 of 40 | 10 of 20 |
| Perplexity | 15 of 40 | 3 of 40 | 27 of 40 | 9† of 20 |
| Claude | 29 of 40 | 5 of 40 | 10 of 40 | 5 of 20 |
* Gemini rarely showed where its answers came from, so few of its answers had a link to check. † Perplexity named a shop in one run and declined in the other in 9 cities.
For single-location dispensary operators, an inaccurate AI answer can quietly steer cannabis buyers to local competitors. AI tools get thrown off easily. During our test, a single customer review led AI assistants to claim a Cincinnati dispensary locks its doors at 9:50 PM, when it actually stays open until 10 PM!
For multi-location cannabis brands, the same errors repeat at every location. One AI assistant confused High Profile's Springfield, Illinois storefront with its Missouri dispensary, and another gave out one chain location's phone number from the state's dispensary map instead of the one on its own site.
Glowing Google reviews offered little protection. Some of the most damaging answers we saw were about dispensaries with 4.9-star averages and thousands of reviews.
More shoppers are asking AI assistants where to shop for weed instead of searching Google. And if they are searching Google, their query will likely trigger Google’s AI Overview/AI Mode.
In SOCi’s 2026 Local Discovery Index, a survey of more than 1,000 US consumers found that 52% said they used an AI tool to find a local business in the past 30 days, up from 9% a year earlier. Two-thirds said an AI tool had given them incorrect information about a local business at least once.
Other studies, such as 5W’s Cannabis AI Visibility Index, measure how often cannabis brands appear in AI answers. That can be useful for brands, but dispensary owners need a more specific answer. For example:
What does AI tell customers who ask about your shop by name?
Is it accurate, and where did the information come from?
That last question matters most because it points to something you can fix.
If I were operating a dispensary, this detail would matter most to me. Every time an AI platform got something wrong, we tracked down where it came from. The same few sources kept showing up.
| Where it came from | What went wrong | A real example | The fix |
|---|---|---|---|
| Your own website | Hours that only appear after the page loads a pop-up or widget, location pages the homepage doesn't link to, an old page left live | Perplexity cited one Tucson dispensary's website and still came back without its hours. A Peoria dispensary's older location page said it was closed Mondays and Thursdays | Type your hours, address, phone number, and rec or medical status right on one page per location (not in an image or pop-up), and link it from your homepage |
| Reviews you show on your site | AI read a customer's words as your rules | A review asking "since when does the door lock at 9:50?" became a "published policy" | Keep customer reviews clearly marked as reviews, and state your real hours right next to them |
| Profiles you haven't claimed | Old hours, old locations, and "may be closed" warnings | An unclaimed Leafly profile told two AI assistants that a 5.0-star Albuquerque shop "may be permanently closed" | Claim your Leafly, Weedmaps, Yelp, and Birdeye profiles and update them |
| Directory sites that copy Google reviews | A frozen handful of old reviews became the complaint list | A page of reviews frozen in March 2025 was still being quoted for complaints in November 2026 | Find these pages, and where you can't fix them, make sure your own site and claimed profiles say more |
| Small review pages | A few angry reviews outweighed thousands of good ones | 27 Yelp reviews outweighed roughly 12,000 Google reviews in one AI's warnings about a 4.9-star pot shop | Claim your Yelp and Leafly profiles and reply to reviews there, so a few old complaints aren't all a reader sees |
| A name that doesn't match your license | AI lost track of which dispensary was which | The Mint in Reno is licensed as Jade Cannabis Co. Claude said it couldn't find The Mint and gave a rival's address | Say both names on your site, plainly |
| Businesses with a similar name | Someone else's reviews, closures, or address | A medspa, a counseling practice, and another state's cannabis app all ended up in answers about our dispensaries | Make your city, address, and the word "dispensary" obvious everywhere your name appears |
| Old news | Settled stories came back with no date and no ending | A license suspension from 2023, over within weeks, kept appearing in answers about opening hours | Keep your own site and claimed profiles current and dated, so there's newer information for AI to find |
| Fake Google Maps listings | Spam "dispensaries" showed up as AI picks | A one-review "dispensary" in a Boston office tower led one AI's list of shops | Report them on Google Maps with "Suggest an edit" |
| Deal and price sites | Old prices, wrong state, and wrong tax | A deal "last seen" more than three months earlier. Springfield, Illinois deals that were really in Missouri and Massachusetts | Post your deals on your own site, with the date each one started |
These challenges extend beyond cannabis marketing, but dispensaries have it harder, because Google Ads doesn't allow ads for adult-use cannabis.
That doesn’t mean paid advertising is off the table. Cannabis businesses can advertise through programmatic display, online video (OLV), and connected TV (CTV).
These channels may be costly for a single shop, but can be more affordable for multi-location operators and cannabis brands. For independent dispensaries, a strong organic presence remains especially important.
We can't guarantee updating these listings alters output across every single AI platform. AI models change quickly, so this study is a snapshot in time.
However, these fixes are affordable, they are in your control, and they clean up the exact sources we watched AI tools pull from.
If it were my storefront, that is where I would start.
Keep a dated “deals” page on your website. Price is the first thing many shoppers ask AI about. List promotions, regular prices, and the dates they apply in plain text. Without current details, AI tools may pull outdated prices from Reddit threads or third-party menu sites
Give each dispensary location its own page, and link to it from your homepage. Clearly show the physical street address, contact phone number, exact operating hours, and if you serve recreational (adult-use) customers, medical patients, or both, using simple body text rather than hiding them behind images, widgets, or pop-up menus. In Las Cruces, a dispensary’s primary site featured its secondary location in another town, confusing two distinct AI platforms about its primary location. Refer to our guide on optimizing dispensary websites for AI search engines to clean up those pages further
Delete old location pages, or point them to your current page. If an outdated page still says you are closed on Mondays, AI tools can still find it and repeat it
List your official business names, nicknames, and aliases clearly. For example, saying “The Mint, legally operating as Jade Cannabis Co.” can keep AI platforms from missing your dispensary entirely
Keep review widgets clearly separate from your dispensary's policies. Otherwise, AI assistants may mistake a customer’s comment for an official rule or business detail
Claim your profiles on Leafly, Weedmaps, Yelp, and Birdeye. Then check the hours, and make sure each location shows as open. Unclaimed profiles were behind many of the “may be closed” warnings we found
Report fake dispensaries near you on Google Maps. Search “dispensary” around your address. A listing with odd characters in its name, a 24-hour "dispensary" in an office building, or a name like "Licensed Plasma Dispensary" is worth reporting with "Suggest an edit." One sat at the same address as a real Tucson dispensary
If your address previously belonged to another dispensary, check its old name too. Perplexity still listed daily hours for MedMen in San Jose, which closed in early 2024, at an address now occupied by The Cake House
Test the AI yourself, and repeat the test. Use the questions in the next section in a private browser window across all five assistants. In some cities, the “best dispensary” recommendation changed between runs of the same question as often as half the time. A single answer is not enough to draw conclusions
If your city shares a name with cities in other states, check that price answers stay local. When asked about the best deals in Springfield, Illinois, ChatGPT led with dispensaries in Springfield, Missouri, and Springfield, Massachusetts
For the most consistent results, open a private browser window and log out of each account where possible. Start a fresh chat for every question. A private window helps prevent your browser history from carrying over, but it doesn’t eliminate every influence, such as location or activity within the current session.
Use ChatGPT Temporary Chat, Perplexity Incognito, or Claude Incognito when available. Google AI Mode may still personalize results, even when you’re logged out.
In a new chat each time, ask ChatGPT, Gemini, Perplexity, Claude, and Google AI Mode:
These are the exact questions we asked, word for word, so your answers will line up with our results.
For each answer, note any inaccuracies, competitors the AI recommends, and the sources it cites. Those sources are the starting point for your fixes.
Rather not spend an afternoon checking? We’ll do it for you.
We’ll check what AI tools tell shoppers about your dispensary and what they can read on your website, then send you the first three things we’d fix. Send us your details in the form below.
Have Us Check Your Dispensary
We'll send you:
Price is usually the first thing a cannabis shopper checks, so we started there. In Sweed's 2025 Cannabis Shopper Experience Report, a survey of 955 US adults, 77% said promotions decide where they shop, and 89% said competitive pricing makes them more likely to come back. In every city, we asked each assistant, "Which dispensary has the cheapest weed in [city]?" and "Which dispensary has the best deals in [city]?"
The price answers were some of the weakest in the study. Every assistant gave at least one price answer with an old price, a shop in the wrong state, or a wrong address or tax rate. The city's top-rated dispensary came up in only 92 of the 200 price and deals answers. Perplexity declined to name a shop in 38 of its 40, but still showed a row of shops above its answer.
| AI assistant | What stood out |
|---|---|
| Google AI Mode | Long lists of daily deals. Suggested Michigan to Chicago and Cleveland weed shoppers. Gave Illinois purchase limits that changed in June |
| ChatGPT | Named shops and prices, sometimes from a price-comparison site with a deal "last seen July 31." Led its Springfield, Illinois deals answer with shops in Missouri and Massachusetts |
| Gemini | Wrong street addresses for real dispensaries. Said the Massachusetts cannabis tax is 6.25% |
| Perplexity | Declined to name a cheapest shop in 18 of 20 cities and the best deals in all 20, but still showed a row of shops |
| Claude | Opened 19 of 20 answers to each question by saying it couldn't name one shop, then quoted prices in 13 cheapest-weed and 11 best-deals answers |
For a dispensary, a price answer is about as close to a sale as it gets. But AI assistants often pull prices from deal pages, directories, and Reddit posts without making clear when those prices were current. The worst examples are in Price Answers Were Out of Date or in the Wrong State.
Not in a reliable way. When we asked, “Should I go to [dispensary] or somewhere else?”, the city’s top-rated dispensary received a clear “go here” recommendation in just 45 of 100 answers. In the other 55, assistants hedged, declined to recommend a shop, or, in 8 cases, sent the customer elsewhere.
The lower-rated dispensary received a clear “go here” recommendation 13 times and was passed over for another shop 29 times. That part was expected.
What stood out was how often assistants hedged even when asked about a 4.9-star shop. In 7 of the 20 cities, at least one assistant recommended the lower-rated dispensary while hedging on or steering customers away from the top-rated one.
ChatGPT steered shoppers away from 13 of the 20 lower-rated dispensaries toward a specific competitor, more than any other assistant. Perplexity declined to recommend a shop in 18 of 40 responses, yet sometimes recommended one anyway. In Phoenix, it explicitly urged cannabis buyers to visit a 4.2-star location while advising them to skip the city’s top-rated 4.9-star storefront.
The same assistant often contradicted itself. In 17 instances, an AI tool called a dispensary "good" in one answer, then recommended a rival or discouraged a visit when asked the very next question. The advice a buyer gets depends a lot on how they word the question.
Mostly, the answers were accurate when the dispensary’s own website clearly listed the details. Across 200 answers, 13 included a fact that contradicted an official record, or about 1 in 15.
That count excludes 47 answers, out of 800, that mixed up the dispensary with another business, as well as answers that sent customers to a competitor.
| AI assistant | Answers | Correct, with a link to the dispensary's own site or listing | Had a wrong fact | Suggested a competitor |
|---|---|---|---|---|
| Google AI Mode | 40 | 38 | 1 | 2 |
| ChatGPT | 40 | 38 | 2 | 0 |
| Gemini | 40 | 1 (it rarely shows sources) | 2 | 0 |
| Perplexity | 40 | 15 | 3 | 0 |
| Claude | 40 | 29 | 5 | 8 |
Google AI Mode and ChatGPT were the most accurate. AI Mode often pulled details directly from Google listings and dispensary websites, which worked well when those sources were accurate, but repeated their errors when they weren’t.
Claude struggled most. It couldn’t find the address of one 5.0-star dispensary, listed a website that belonged to someone else, and recommended a competitor in 8 of 40 answers to a simple question about the shop’s location. Its final eight answers, collected November 9, were all usable, though one cited a spam directory for a St. Louis dispensary’s hours.
Usually, and that included the lower-rated shops. AI assistants said “yes” in 72 of 100 answers about top-rated dispensaries, and in 42 of 100 answers about lower-rated ones. They often endorsed lower-rated dispensaries without mentioning their ratings. Gemini endorsed 13 of the 20 lower-rated dispensaries; Google AI Mode endorsed 12.
Partway through testing, Perplexity switched to one-paragraph answers and kept using them after a 47-minute break.
For three dispensaries, including one with about 1,400 Google reviews and a 4.8-star rating, it said there wasn’t enough review information to judge. To a shopper, “not enough evidence” can sound like a no.
This is where the damage showed up. Of the 200 answers to questions about complaints, 173 (87%) included at least one detail that could turn a customer away. In 95 answers (48%), the negatives outweighed the positives, with more likely to lose a sale than win one.
That’s partly to be expected because we asked about complaints. The bigger concern was how often the evidence was thin, outdated, or about another business.
One Cleveland dispensary had about 900 Google reviews and a 4.8-star rating, but just six reviews on Leafly. One of those six said “avoid,” and several assistants repeated that review in their answers about complaints.
Across all five assistants, 41 of 100 answers about top-rated dispensaries leaned more negative than positive. For lower-rated dispensaries, that rose to 54 of 100
Google AI Mode was toughest on top-rated shops, with 15 of 20 answers leaning negative. Claude was the least critical, with just 1 of 20. It more often said it “couldn’t find specific complaints”
Perplexity was critical of both groups: 12 of 20 answers about top-rated dispensaries and 15 of 20 about lower-rated ones leaned negative
Often, no. We asked “What’s the best dispensary in [city]?” twice per city, each time in a fresh chat. For four assistants, the two runs were about 30 minutes apart. Claude’s were roughly five hours apart because its free plan ran out before we could ask again.
| AI assistant | First dispensary named changed between runs | Named the city's top-rated dispensary from our pair |
|---|---|---|
| Google AI Mode | 5 of 20 cities | 32 of 40 answers |
| ChatGPT | 7 of 20 cities | 30 of 40, and the lower-rated one in 3 |
| Gemini | 10 of 20 cities | At least 10 of 40 (some of our saved answers are cut short) |
| Perplexity | Picked in one run and refused in the other in 9 cities | 22 of 40, and the lower-rated one in 23 |
| Claude | 5 of 20 cities (runs about five hours apart) | 30 of 40, and the lower-rated one in 5 |
AI Mode and Claude were the steadiest, at 5 cities each, and both named the city's top-rated dispensary in most answers. Claude managed that even with five hours between its runs. Gemini changed its first pick in half the cities. Perplexity usually showed a row of shops pulled from Google Maps and then either declined to call any of them the best or named one. Because that row lists many shops, Perplexity "named" the lower-rated dispensary in our pair far more often than AI Mode or Gemini did.
Perplexity mostly refused. But it still showed a row of shops, and those weren't always local. Its Springfield, Illinois answer showed St. Louis dispensaries in another state, its Cleveland deals answer showed Michigan dispensaries, and its first Tucson deals listing was a Phoenix shop. It also showed smoke shops and medical-card doctors, and in Springfield, Illinois, the state's medical cannabis office. Oddly, on the cheapest-weed question in San Diego and Cleveland, it did name shops and prices.
The others answered with problems we could check:
We checked the quoted prices we could against the dispensaries' own websites. All three matched: Planet 13's 30% locals discount in Las Vegas, and $20 eighths at Cookies Downtown and eighths of $25 or less at Luxury Leaf, both in St. Louis. Most other prices sit on live menus that change daily, so we've reported them as the AI gave them.
In 82 of 800 answers, about one in ten, AI assistants repeated claims about a dispensary’s honesty, safety, or legality. The claims ranged from deleted or rewarded reviews to mold, unpaid suppliers, a parking-lot robbery, and allegations about an owner. We could verify few of them. When we could, such as a 2025 state contamination notice, the answer often gave no date or outcome. Some assistants said “allegedly”; many did not.
We won’t name the dispensaries or repeat the claims. Doing so would reproduce the very harm we’re describing.
We could verify one example. Gemini described a Chicago dispensary’s routine union vote and first union contract as “union friction” and “controversies surrounding former leadership.” We found no evidence of a dispute or leadership controversy, and Gemini reversed the timeline.
These are mostly true events. What the answers usually left out was the date and how each one ended.
A 2019 hobby blog and a 2019 national news story about payments to drivers who bring in customers shaped answers about a top-rated Las Vegas dispensary in 2026
A 2023 license suspension at a Baltimore dispensary, settled and reopened within a month, came up in answers that only asked for its hours
A 2023 union drive and a federal labor board settlement came up when someone asked where a Columbia, Missouri dispensary is
A 2020 pesticide settlement against a San Jose dispensary's operator came up in its complaint and "should I go" answers
A 2017 settlement over California's Prop 65 product-warning law, against a former operator, came up in a San Diego dispensary's complaint answer
A 2025 state product advisory came up again and again for one Springfield, Massachusetts dispensary. The same advisory listed the city's top-rated dispensary too, and only one AI assistant mentioned that
We looked for dispensaries in our 20 cities that had closed for good, confirmed by the state or a dated news report, and found three. We asked all five AI assistants where each one is and if a customer should go there. This was a small side test, separate from the 800 graded answers.
| Closed dispensary | ChatGPT | Claude | Perplexity | Google AI Mode | Gemini |
|---|---|---|---|---|---|
| Sacred Garden, Las Cruces (closed September 2024) | Caught it | Caught it | Caught it | Said it's open | Gave hours |
| Paradise Exotics Distro, Albuquerque (license revoked July 2023) | Caught it | Caught it | Caught it | Caught it | Gave hours |
| MedMen, San Jose (closed early 2024) | Caught it | Caught it | Gave hours in one answer | Caught it | Caught it |
Sacred Garden, Las Cruces. Local TV station KVIA reported its last day as September 21, 2024, and Google's own listing now labels it "Out of business." Google AI Mode still said "the storefront is open daily," with hours and a phone number, citing directory pages that carry the old listing. Gemini gave its weekday and weekend hours, then told the customer to skip it because "local customer reviews and feedback have skewed mixed-to-negative in recent times." It never mentioned the closure, which was the clearest reason not to go. ChatGPT, Perplexity, and Claude all said it closed in September 2024 and named their news sources.
Paradise Exotics Distro, Albuquerque. New Mexico's Regulation and Licensing Department revoked its license in July 2023 and ordered sales stopped. More than two years later, Gemini said it's "generally open daily from 9:00 AM to 8:00 PM." The other four all brought up the revocation, and Claude warned that a directory still lists the shop as open.
MedMen, San Jose. It closed in early 2024, the year the chain went through bankruptcy, and The Cake House now holds the state license at the same address. Perplexity's "should I go" answer said it had closed, but the row of shops above that answer still showed MedMen as a place to visit, and its "where is it" answer gave daily hours of 9 AM to 9 PM. The City of San José's own list of registered cannabis businesses, dated February 2025, still names MedMen at that address.
Three closures can't give us a rate. Still, with Sacred Garden, Google's own AI assistants missed what Google Maps already knew.
Bloom Wellness, Peoria, Illinois. ChatGPT showed a "Temporarily closed" label in three of its answers, and AI Mode said it "may have temporarily or permanently closed." Its Google listing shows daily hours, its Illinois license is active (still under an older company name), and Leafly shows no closure notice. Both cited a Reddit thread from early November that we couldn't open, so there's a small chance of a short closure we can't see. Everything we can see says open.
A 5.0-star dispensary in Albuquerque. Two AI assistants told customers to confirm it's still operating, because an unclaimed Leafly profile (one the business hasn't taken control of) says "this business may be permanently closed," and a directory says the same. We can't prove it's open either. Its Google listing has no hours, and its own web address now redirects to a gambling-app download page. ChatGPT gave that address as the dispensary's website.
A Las Vegas dispensary. Claude led with "may have closed permanently." Nevada still licenses the shop, its Google listing has no hours or phone number, Leafly shows it closed, and its own website lists only its other location. We honestly can't tell if it's open. If we can't, an AI can't either, and that's the point.
Perplexity's "best dispensary" answers usually open with a row of shops pulled from Google Maps, each with a star rating. In 17 of its 40 answers, covering 9 of the 20 cities, at least one shop looked like a spam listing: letters swapped for lookalike characters from other alphabets, misspellings like "Premium Suply," or names like "Greencare Source Licensed Plasma Dispensary." It also showed one in a Boston cheapest-weed answer.
Google Maps listing, Boston
Cаnnаbis De𝗅ivery Dispensary
We checked suspicious listings in five cities on Google Maps:
We didn't open the others on Maps, so for those, "likely spam" is as far as we'll go. We can't tell from the outside who runs these listings or where an order placed through one would go. That's what worries me. A weed shopper who taps the first shop in an AI answer has no reason to suspect it.
My guess is the lookalike letters are there to make the listings harder to catch, since searching the name as it looks won't find them. I can't prove that. None of the other AI assistants showed these listings in the answers we saved.
Locals Cannabis, Cincinnati. ChatGPT and AI Mode said the doors lock at 9:50 PM, and Gemini gave a 9:45 PM closing time. ChatGPT called it "published policy." The dispensary's site says it's open 9 to 10 daily. The only "9:50" anywhere on its site is inside a customer review it displays on its own location page: "Since when does the door lock at 9:50?" ChatGPT also pulled criticism from an app made for a similarly named cannabis business in Washington state. That criticism had nothing to do with the Cincinnati dispensary.
Next to some answers, ChatGPT shows a small map with a star rating for each dispensary. Those ratings showed 2.9 stars for a Las Vegas dispensary with 4.9 on about 29,000 Google reviews, 3.3 for a Columbia, Missouri dispensary at 4.9 on about 1,800, and 3.9 from 18 reviews for a San Jose dispensary at 4.9 on 537. They come from map and listing sources other than Google. AI Mode said the same Las Vegas shop had "abysmal 2-to-3-star averages on Yelp and Google."
Lots of good reviews can even count against you. Perplexity said it "wouldn't choose it solely because of its unusually high review score" about that 4.9-star Las Vegas dispensary. Claude said a Chicago dispensary's 7,000-plus reviews were "likely driven by review-solicitation programs," with no evidence offered.
AI Mode built a Cincinnati dispensary's complaint answer from reviews of "Local Cannabis Company," which may be a different business
Perplexity listed complaints about The Mint in Reno that Claude traced to Mynt, a different Reno pot shop
AI Mode blamed High Profile in Springfield, Illinois for a closure story about a High Profile dispensary in Springfield, Missouri, and for a review that belongs to Maribis, a competitor across town
Asked if a customer should go to Therapy Cannabis in Cleveland, AI Mode showed only a panel for the 988 Suicide and Crisis Lifeline, and Perplexity suggested follow-up questions about cannabis-friendly therapists
Perplexity treated BOUTIQ in East Boston as a boutique hotel, and suggested bars and seafood restaurants for a Maribis dispensary in Springfield, Illinois
Several AI assistants pulled answers about Bloom Wellness in Peoria from medspas and counseling practices with the same name in Georgia, Michigan, and Arizona
The Mint, Reno. Claude said it couldn't find a dispensary called Mint Cannabis in Reno and gave the address, hours and rating of a different shop, Mynt / The Dispensary. The Mint is at 1085 S Virginia St and has 242 Leafly reviews. Its state license is under a different name (Jade Cannabis Co.), which is likely part of why AI couldn't keep its identity straight. AI Mode called it "formerly Jade" in one answer and "formerly Mynt" in another.
A Fresno dispensary. Claude said it couldn't find the dispensary, offered a shop with the same name in Michigan instead, and suggested a Fresno weed store with a similar name. The dispensary holds an active California license. Its website lists the Fresno location only as a menu link, with no address, hours, or phone number.
A Las Cruces dispensary: the homepage shows its other dispensary in Ruidoso. Claude couldn't find the Las Cruces address and suggested a competitor. Perplexity gave the Ruidoso shop's 9-to-9 hours for a Las Cruces dispensary that Google lists as open 24 hours
A San Jose dispensary: the homepage has no San Jose details. Claude pulled the address from state license records, couldn't confirm rec or medical, and suggested two rivals as the dispensaries "that have posted hours"
A Reno dispensary: the Reno page has no address, phone number, or hours. Claude couldn't give a closing time and suggested two competitors
A Springfield, Missouri dispensary: the site was last updated in 2022 and still talks only about medical patients, so ChatGPT wasn't sure it sells to adults 21 and over
A Tucson dispensary: the hours only appear after the page finishes loading. Perplexity listed the dispensary's own site as a source and still couldn't find them
These are some of the easiest problems in this report to fix, and none of them costs much. We can't promise every AI assistant picks up the fix, which is why we'd ask the same questions again a few weeks later.
A directory site that copies Google reviews froze five reviews from one weekend in March 2025, and 19 months later it was the cited source for complaints about dispensaries in Ohio and Missouri. Nobody at those shops can reply on that site
An unclaimed, half-updated Leafly profile, still describing a Tucson dispensary's old location with the wrong opening hour, supplied its complaint list
A 27-review Yelp page at 1.7 stars led Perplexity's warnings about a Phoenix dispensary with about 12,000 Google reviews at 4.9
Perplexity and AI Mode used Indeed and Glassdoor reviews written by employees as if they were customer complaints
Most wrong facts weren't made up. They came from two or more published sources that disagree, and the AI picked one.
A Peoria dispensary: Perplexity found an "older location page" saying the shop is closed Monday and Thursday. Its live page and Google both show seven days open
Hours differed between the dispensary's own site, Leafly, Weedmaps, and directories for seven of our 40 dispensaries, in Baltimore, Columbia (two), Las Cruces, Cincinnati, San Jose, and Phoenix. One Springfield, Massachusetts dispensary's own pages disagreed with each other
Rec or medical: BOUTIQ in East Boston holds an adult-use license only, and Claude still said it serves both. Cookies Fresno is licensed for both, and one AI assistant called it recreational only. Shangri-La in Columbia, Missouri says on its own site that it serves medical patients too, and Perplexity called it recreational only. We didn't grade this for one Boston dispensary, whose state record is mixed, or for the Illinois dispensaries, because Illinois added medical sales at 37 adult-use dispensaries in September 2026 without a public list we could find
Phone numbers: one Cincinnati dispensary has one number on its site and another on the state's map, and ChatGPT gave out the state's
Dispensaries with more than one location got blended. Perplexity silently picked one of Dispensary33's two Chicago shops, answers split HiKei across three San Diego-area dispensaries, and ratings for Maribis couldn't be tied to its Lindbergh location in Springfield, Illinois
Listings were mislabeled. A directory filed one Tucson dispensary as a restaurant, High Profile in Springfield, Illinois was listed under Springfield, Missouri, a Google Maps panel called Purple Lotus in San Jose a clothing store, and one dispensary's own review page carries a review of a restaurant meal
ChatGPT changed its top pick in 7 of the 20 cities: Baltimore, Chicago, Columbia, Reno, San Diego, San Jose, and St. Louis. In St. Louis the number one pick flipped between two runs five minutes apart. In Columbia, the city's most-reviewed 4.9-star dispensary was the top pick in the first run and missing from the second. In Las Vegas, the city's most-reviewed dispensary was third in one run and gone in the next
Gemini changed the first dispensary it named in 10 of 20 cities: Albuquerque, Chicago, Cincinnati, Cleveland, Peoria, San Diego, San Jose, Springfield (Illinois), Springfield (Massachusetts), and St. Louis
Google AI Mode changed it in 5 of 20: Cincinnati, Reno, San Jose, Springfield (Massachusetts), and Tucson. It also changed its facts. In Tucson, one run said D2 Dispensary "has repeatedly been voted 'Best Dispensary'" in the Tucson Weekly awards, and the next run gave that title to Earth's Healing
Claude changed it in 5 of 20, with its runs about five hours apart: Albuquerque, Fresno, Las Cruces, Peoria, and St. Louis. In Phoenix it led with JARS both times, but a different JARS location
Perplexity answered in one run and declined in the other in 9 cities. Where it named a shop both times, the pick held in every city but Chicago
San Jose was the only city where the first pick changed on all three of Gemini, ChatGPT, and AI Mode. If you're checking your own city, one answer tells you very little.
Smaller observations about how individual AI assistants behaved. They didn't make the main list, but they're worth knowing.
Claude twice used phrases that sounded like it remembered earlier chats, in fresh chats with no history, such as "the weakest of the shops I've looked at for you" and "the other Tucson stores we've looked at"
Claude added ratings and complaints to answers that only asked for hours and location, for Story Cannabis in Cleveland, Trulieve Roosevelt Row in Phoenix, HiKei in San Diego, and Shangri-La in Columbia, Missouri. It also suggested two competitors to people asking where Bloom Wellness is
Claude took High Profile's hours from a job posting
Perplexity's suggested follow-up questions pointed people toward "nearby dispensaries with higher ratings" for Trulieve Roosevelt Row
Claude's free plan ran out after roughly 10 to 30 answers every five hours, which is part of why Claude's numbers came in last
Claude's Springfield, Massachusetts cheapest-weed answer said the city "has five dispensaries" and listed them. Enlite, one of the two Springfield dispensaries in our study, wasn't among them
Claude's Peoria price answer leaned on reviews from 2019 and 2020. To its credit, it said they might not reflect current prices
AI Mode cited ranking pages from Dispenza, a cannabis marketing agency, in at least three answers
Claude flagged a Cincinnati Magazine article about a dispensary as probably sponsored, because the byline read "CM Sales." None of the others questioned their sources that way
Not reliably. Every assistant we tested gave at least one cheapest-weed or best-deals answer with an old price, a shop in another state, or a wrong address or tax rate. Perplexity wouldn't name a cheapest dispensary in 18 of 20 cities. If price is how you win customers, keep a dated deals page on your own site so AI has something current to quote.
About 1 in 15 times in our sample. When we asked where a dispensary is, when it's open, and if it's rec or medical, 13 of 200 answers got at least one of those facts wrong against the state license list, the dispensary's own site, or its Google listing. Mix-ups with other businesses and competitor suggestions are on top of that.
Google AI Mode and ChatGPT. Each gave correct details with a link to the dispensary's own site or listing in 38 of 40 of those answers. Claude had the most trouble, with 5 wrong facts and a competitor suggested in 8 of 40. Gemini was hard to check, because it rarely showed where its answers came from. Being accurate didn't make AI Mode gentle, though. It was also the hardest on top-rated dispensaries when asked about complaints.
Not in our sample. Fewer than half of the "should I go?" answers about each city's best-rated dispensary said yes without hedging. In 7 cities, one or more assistants leaned toward the weaker shop instead.
Sometimes. We tested three dispensaries that had closed for good. ChatGPT and Claude caught all three. Gemini described two of them as open, and Google AI Mode one. Two AI assistants also said an open Peoria dispensary might be closed.
Start with what you control. Keep a dated deals page, and make sure your own site spells out your hours, address, and rec or medical status for every location. Then claim your Leafly, Weedmaps, and Yelp profiles, and report fake Google Maps listings near you. The full list is in How to Fix What AI Gets Wrong About Your Dispensary.
ColaDigital, "When AI Gets Your Dispensary Wrong, and How to Fix It," November 2026, by Vee Popat. Link: https://coladigital.ca/ai-dispensary-answers-study/. Writers and researchers are welcome to use our charts and numbers, as long as they credit ColaDigital and link back to the study.
Book a 30-minute call and Vee will go through what the five AI assistants tell shoppers about your dispensary, and which pages and listings to fix first. It's part of the dispensary SEO work we do for cannabis retailers. You can see what that work looked like for one dispensary in our dispensary SEO case study.
The AI assistants. ChatGPT (OpenAI), Google AI Mode (the AI answer tab inside Google Search), Gemini (Google's standalone assistant), Perplexity (an AI search tool that shows its sources), and Claude (Anthropic). Our original plan named three. We added Gemini and Claude during testing because cannabis consumers use them too.
The cities and dispensaries. The 20 cities come from our earlier study, Who Owns Dispensary Search?, which looked at who holds Google's first page for "dispensary [city]" searches. In each city we picked two dispensaries from Google Maps before asking any AI anything:
The 40 dispensaries range from 3.8 to 5.0 stars on Google. We checked every one against its state license list. New Mexico's public lookup couldn't be searched, so the four New Mexico dispensaries are unverified on that point.
The questions. For each dispensary we asked four questions, in a fresh chat each time, on all five AI assistants:
That's 800 answers, and we graded every one. For each city we also asked "What's the best dispensary in [city]?" twice (about half an hour apart, or about five hours apart for Claude), plus "Which dispensary has the cheapest weed in [city]?" and "Which dispensary has the best deals in [city]?" And we asked the first and fourth questions about three dispensaries that had closed for good. Those city-wide and closed-dispensary answers, 430 in all, sit outside the 800 graded ones. For them we counted things we could check, such as if an answer named the city's top-rated dispensary, and we checked addresses, tax rates, dates, and the prices we could against official sources and the dispensaries' own websites.
We planned to test five dispensaries that had closed for good in 2025 or 2026, confirmed by the state and one other dated source. We couldn't find any in our cities, so we tested three earlier closures we could confirm, one from 2023 and two from 2024. We checked the suspicious Perplexity listings by opening them on Google Maps in five cities.
How we asked. ChatGPT, Google AI Mode, and Gemini were used logged out. Perplexity and Claude were used on free accounts with no chat history. Testing ran November 5 to 10, 2026, through a US internet connection. Each question started a fresh chat with no follow-ups.
How we graded. We read every answer and marked anything that could win the sale (the dispensary recommended by name, praised, linked correctly, or its rating cited favorably) and anything that could lose it (a wrong fact, a competitor suggested, a complaint repeated, a damaging claim about honesty, safety, or legality, a different business mixed in). When the things that could lose the sale outnumber the things that could win it, or there's nothing on the winning side, we count the answer as doing more harm than good. A fact only counts as wrong when an official record contradicts it: the state license list, the dispensary's own website, or its Google listing. Anything we couldn't check isn't counted either way.
A second grader checked our grading. ChatGPT re-graded a random 80 of the 800 answers, 16 per AI assistant, without seeing our grades. We gave the same overall grade to 61 of 80 (76%; a statistic called Cohen's kappa, which adjusts for agreement by chance, came out at 0.61, usually read as substantial agreement). We settled the 19 disagreements against our written rules. Most came down to four wording questions, which we answered and then applied to all 800 answers: a conditional "go there if" counts as a recommendation, a regulatory notice or recall counts as a claim about safety even when it's real, naming a competitor only to compare ratings isn't sending the customer there, and a link to the dispensary's homepage counts as its own site. That changed the grade of 29 answers. Every figure in this report uses the corrected grades.
The download has the numbers behind every chart and table in this report as spreadsheet files (CSV), the four charts as images, and a short note on how we collected the answers. It has totals only and doesn't name the dispensaries in the negative findings. You're welcome to use any of it, as long as you credit ColaDigital and link back to this study.
One snapshot per question per AI assistant, between November 5 and 10, 2026. Answers change
Logged-out or fresh-account answers can differ from what a signed-in shopper sees
40 dispensaries in 20 cities is a sample. Results for any one city are examples and shouldn't be read as city rankings
The complaint question invites negative answers. It measures how AI responds when asked, not how often people ask
We measured statements that could change a shopper's decision. We didn't observe sales
Ratings and review counts show associations, not causes
Where an AI cited a source we couldn't open, we didn't count the claim as right or wrong
The "best dispensary" question was asked twice per city, a small test of consistency. Two runs can't tell you how often a pick changes over a week or a month
Some of our saved Gemini answers to the "best dispensary" question are cut short, so its count of top-rated dispensaries named is a minimum
Perplexity's row of shops lists many dispensaries, so its "named" counts include shops it showed but didn't recommend
We checked the quoted prices we could against dispensaries' own websites. The rest are reported as the AI gave them
The closed-dispensary test covers three shops, too few to give a rate
We checked every source below on November 9, 2026. Any other date in an entry belongs to the source itself.
SOCi 2026 Local Discovery Index: soci.ai/insights/local-discovery-index (figures as reported by WSOC-TV, November 4, 2026, and MarTech)
5W Cannabis AI Visibility Index 2026: 5wpr.com/ai-visibility-index/cannabis-ai-visibility-index-2026
Sweed, The 2025 Cannabis Shopper Experience Report: online survey of 955 US adults 21+, October 6 to 10, 2025, conducted by Wired Research (info.sweedpos.com)
Google Ads policy on recreational drugs: support.google.com/adspolicy/answer/16489299
Nevada Cannabis Compliance Board list of licensees: ccb.nv.gov/list-of-licensees
Illinois IDFPR adult-use dispensary list: idfpr.illinois.gov
California DCC license records: real.cannabis.ca.gov
Massachusetts CCC retail license records
The dispensaries' own websites and location pages, checked November 9
Sacred Garden closure: KVIA, September 4, 2024 (kvia.com); Google listing "Out of business" (Nov 9); Leafly "may be permanently closed"
MedMen San Jose closure: San José Spotlight, February 5, 2024 (sanjosespotlight.com); state license at 1075 N 10th St now The Cake House San Jose (C10-0001576); City of San José registered cannabis businesses list, as of February 25, 2025 (sjpd.org)
Paradise Exotics Distro license revocation: New Mexico Regulation and Licensing Department, announced July 13, 2023 (Albuquerque Journal, KOAT)
Illinois purchase limits: IDFPR SB 3222 fact sheet, effective June 12, 2026 (idfpr.illinois.gov)
Olswell Cannabis Co. locations: JointCommerce listings for Adrian, Kalkaska, Traverse City and Grand Rapids, Michigan
Greenlight Dispensary Springfield, Missouri (218 N National Ave): Leafly and JointCommerce listings
Prices checked on the dispensaries' own sites: planet13.com, cookiesdowntownstlouis.com, luxuryleafstl.com
Urbn Leaf San Diego (C10-0000246) and Airfield Supply Co. San Jose (C10-0000514): licenses shown as active
Massachusetts marijuana excise tax of 10.75%: mass.gov, General Laws chapter 64N, section 2
Google Maps listings, checked November 9: "Cаnnаbis De𝗅ivery Dispensary" (200 Clarendon St, Boston), "Ｗeed And Cаnnаbis De𝗅ivery Dispensary" (N Central Ave, Phoenix), "Greencare Source Licensed Plasma Dispensary" (3008 N Stone Ave, Tucson), three "GreenLight Source Licensed Plasma Dispensary" listings (Las Vegas Strip), "Evergreen Leaf Plasma Dispensary" (Baltimore)
Addresses for High Profile Springfield, Illinois (4211 Conestoga Dr, per directory listings and High Profile's own opening announcement) and 6 Brick's Springfield, Massachusetts (1860 Main St, per 6bricks.com)
Google listings for the dispensaries in this report, checked November 9
Leafly profiles for the dispensaries in this report
App Store listing id6446124261 (Locals Canna House, Washington Fresh LLC)
CityLab/Bloomberg, January 2019, on Las Vegas driver referral payments
