Dispensary Menu Crawlability Index 2026

Each menu URL got one measurement from each of three methods:

• Raw HTML requested with the GPTBot user-agent string, which is what a crawler that does not run JavaScript receives.
• Raw HTML requested with a desktop browser user agent, which separates a block aimed at GPTBot from a page that sends little text to anyone.
• The rendered page, loaded in headless Chromium after its scripts ran.

For every fetch we counted the readable words in the page body, after removing scripts and styles, and the JSON-LD nodes typed Product, Offer or AggregateOffer, duplicates included. Word counts include header links and footer text, so they measure how much text a page sends, not how many products it lists. A menu inside an iframe is a separate document, so its text is not in the host page's count.

Every reading is a median across the URLs in one platform group, never an average. Raw and rendered medians are taken separately, so within one group they can come from different URLs: read them as two readings, not as one page before and after.

Operators are anonymous. We name the platforms because the platforms are what the index measures, but no dispensary or store is named, and several menus from one business share one operator label.

The sample has limits:

• It is small and was chosen by web search and homepage inspection, not drawn at random.
• Requests used the GPTBot user-agent string from a non-OpenAI IP address, so a 403 shows how a site treats that user agent from an unverified source and may not match what OpenAI's crawler receives.
• No Dispense (AIQ Ecommerce) storefront was identified in the sample, and Treez ecommerce reached 2 operators, below the 3-operator minimum the index uses for a platform reading. Neither platform has a reading in the table.
• We did not evaluate robots.txt, check search engine indexing, or test whether any AI product cites these pages.
• A single run reflects one moment. Menus, age gates and store pickers change what a page returns.

Every figure here comes from public page fetches. Client accounts feed a different study, our state of dispensary search report, which reads Search Console and AI answer data from the client book.

The Dutchie rows differ by setup more than by vendor.

• The dutchie.com marketplace page refused the browser user agent as well, so it shows nothing about how Dutchie treats AI crawlers in particular.
• The embedded menu on a store's own site is an iframe, and Dutchie's support documentation says “Your Embedded menu will be partially indexed by google.” Our build method describes an embedded Dutchie iframe as not same-origin: the menu text belongs to Dutchie's document, not to your page.
• Dutchie Plus and headless builds carried no JSON-LD Product, Offer or AggregateOffer nodes at the median, raw or rendered. A median of zero does not rule out markup on an individual URL.
• Dutchie-hosted storefronts served from a store domain sent the least raw text of any own-domain group, and they were the only own-domain group with a 403 to the GPTBot user agent.

That last result matches an earlier check of our own: when we fetched nine live Dutchie storefronts as GPTBot on October 12, 2026, the median page body carried two words of text and none had an H1.

None of this settles whether ChatGPT can read Dutchie menus. OpenAI's crawler overview states that “OAI-SearchBot is used to surface websites in search results in ChatGPT’s search features.” The index sent the GPTBot string from a non-OpenAI IP address, so it shows what a crawler announcing itself as GPTBot, and running no JavaScript, was sent.

Disclosure: Bud Authority is a Dutchie Certified Partner. That status did not change how the Dutchie rows were fetched or counted, and the index ranks no platform.

Leafly was the only platform group whose raw HTML carried JSON-LD Product, Offer or AggregateOffer nodes at the median, from 3 URLs and 3 operators.

Weedmaps hosted pages showed the opposite pattern in this sample: a median of 35 raw words with a GPTBot user agent and 1,525 after rendering.

The gap between raw and rendered matters because crawlers differ. Vercel's study with MERJ, The rise of the AI crawler, reported as of its publication: “The results consistently show that none of the major AI crawlers currently render JavaScript.” The same study named an exception: “AppleBot renders JavaScript through a browser-based crawler, similar to Googlebot.” Google does render, and its JavaScript SEO basics page sets the limit: “If the content isn't visible in the rendered HTML, Google won't be able to index it.”

The anonymised dataset is published as dispensary-menu-crawlability-index-2026-10-27.csv: 37 rows, one per measured URL, with operators as OP labels and no URLs. Its columns are operator, platform_group, checked, gptbot_raw_status, gptbot_raw_words, gptbot_raw_h1, gptbot_raw_jsonld_product_offer_nodes, browser_raw_status, browser_raw_words, rendered_words, rendered_h1, rendered_jsonld_product_offer_nodes and age_gate_flag. The two jsonld columns count JSON-LD Product, Offer or AggregateOffer nodes, duplicates included.

Cite this as: Mark Wallace, Bud Authority, Dispensary Menu Crawlability Index, measured October 27, 2026, budauthority.com/research/dispensary-menu-crawlability-index. When you quote a reading, give the platform group with its URL and operator counts, and keep the non-OpenAI IP caveat beside any 403.

Take the menu out of the iframe. Dutchie documents a custom domain as the alternative to the iframe, which it calls “a custom URL, often referred to as a subdomain.” In this index, Dutchie-hosted storefronts on store domains, by custom domain or proxy, still sent almost no raw text before JavaScript ran, so check the new URL's raw response as well.

One cause OpenAI documents is a robots.txt opt-out: “Sites that are opted out of OAI-SearchBot will not be shown in ChatGPT search answers, though can still appear as navigational links.” The Vercel and MERJ study above adds a second check: the major AI crawlers it measured did not render JavaScript, so menu text that appears only after scripts run was out of their reach.