If you run a cannabis delivery service in Michigan, you probably have a growing pile of writing to do: menu descriptions for dozens of flower strains, edible listings, email updates about weekend specials, and social posts that have to follow state advertising rules. Some operators have started looking at AI writing tools to speed this up, and a common shortcut is to buy midjourney prompts for image concepts or to pick up tested text prompts for the writing side. The catch is that a prompt is only worth something if it reliably produces output you can publish without rewriting it from scratch.
Why most AI prompts fail for cannabis copy
Generic prompts like “write a fun description of our sour gummies” tend to produce copy that sounds like every other product page on the internet. Worse, they often drift into territory you cannot use: health claims, dosage promises, or language that sounds aimed at younger customers. A prompt that works for a clothing store will not work for a regulated product sold to adults in a licensed market.
A prompt that works for your business does three things well. It sets the context clearly, it defines the boundaries, and it specifies the output format. When any of those three pieces is missing, you end up editing heavily, and the time you saved disappears.
A prompt structure that holds up
The most useful prompts we have seen for delivery menus follow a repeatable pattern. You do not need anything fancy, just consistency. A workable structure looks like this:
- Role: Tell the tool what it is writing as, such as a copywriter for a licensed adult-use delivery service in Michigan.
- Source facts: Paste the verified details from your product label, including THC and CBD percentages, weight, strain type, and ingredients. Instruct the tool to use only these facts.
- Boundaries: List what is off limits, such as medical claims, references to curing anything, appeals to children or youth culture, and exaggerated potency language.
- Voice: Give two or three sentences describing your brand tone, plus one sample of copy you already like.
- Output format: Specify the character count, whether you want a headline and body, and whether you want bullet points.
Keep this structure in a shared document so every team member uses the same version. Small changes to the boundaries section can make a big difference in what comes back.
Building a testing habit
A prompt that worked once is not necessarily reliable. Before you trust any prompt for live menu copy, run it several times with different product inputs and read every result. Look for three problems in particular: invented facts, wording that crosses your compliance line, and tone that does not match your brand.
It also helps to keep a simple log. Record the prompt version, the product it was used on, and whether the output needed major edits. Over a few weeks, patterns will show you which prompts are dependable and which need work. This log is also useful if you ever need to show a regulator or a partner how your content was produced and reviewed.
Where prompt marketplaces fit in
Some teams prefer not to build every prompt from scratch. Marketplaces that list prompts from independent creators can be a reasonable place to look for a starting point, especially for visual concepts like packaging mockups or lifestyle imagery that you would later adapt. Browsing a marketplace such as PromptMart’s listings of tested prompts lets you compare approaches side by side, but treat anything you find as a draft. Read it against your own boundaries, run it on your real product data, and adjust it before it touches a customer-facing page.
The same caution applies to any imagery. AI-generated images of cannabis products can accidentally look like they depict minors, or they can show packaging details that do not match what you actually sell. Review every visual before publishing, and never use generated images that imply a product is something it is not.
Compliance review is not optional
Michigan’s cannabis rules cover advertising, packaging, and the content businesses can present to customers. The details change over time, so check the current guidance from the Michigan Cannabis Regulatory Agency and, if you have one, your compliance counsel. AI tools do not know your license terms, and they will not flag a violation for you. That responsibility stays with the people who publish the content.
A simple review step helps. Assign one person to check every new piece of copy against a short checklist: Are all product facts sourced from the label? Are there any health or therapeutic claims? Does the language avoid appealing to anyone under the legal age? Is the age-gate notice present where required? If the answer to any question is uncertain, do not publish until it is resolved.
A realistic workflow for a delivery menu
Here is how a small operation might handle a weekly menu update without adding chaos:
- Pull the verified product data from your inventory system or label records.
- Run the approved description prompt for each new item, using only that data.
- Have a staff member edit each draft for accuracy and brand voice, not just grammar.
- Run the compliance checklist on every item before it goes live.
- Log any output that needed heavy correction, and update the prompt if the same problem keeps appearing.
This process is slower than pasting a prompt and hitting publish, but it is far faster than writing each description by hand, and it keeps your menu consistent. Over time, the edits get shorter because the prompts get better.
What to expect from AI in this niche
AI tools are useful for drafting, organizing, and generating variations. They are not a substitute for product knowledge, legal judgment, or a clear sense of what your customers actually want. The businesses that get the most value tend to be the ones that treat AI as a fast first draft and keep humans responsible for the final word.
If you are just starting, pick one repetitive task, such as product descriptions for a single category, and build a tested prompt for that alone. Once it is dependable, move on to the next task. A few well-tested prompts will do more for your workflow than a long folder of untested ones.
Final thoughts
The useful question is not whether AI can write cannabis copy. It can. The useful question is whether your prompts, your boundaries, and your review process are good enough that the output is accurate, compliant, and on brand every time. Build those pieces carefully, test them with real product data, and keep a person in the loop, and AI becomes a practical tool for a delivery business that has more writing to do than hours in the day.

Leave a Reply