Every week our team faces the same problem: a new strain arrives, the menu needs updated descriptions, a loyalty email is overdue, and the social calendar is empty. Some of us have started testing ready-made instructions for AI writing and image tools, including collections of midjourney prompts that are written and checked by their authors before they are sold. The biggest lesson so far is that the quality of the instruction matters more than the tool you pick, and a well-built prompt saves far more time than a clever one-liner.
Why most AI marketing copy sounds generic
When you ask an AI tool to ‘write a product description for a cannabis gummy,’ you get something that could belong to any shop in any state. It is polite, vague, and often full of phrases that no real customer uses. The problem is rarely the software. The problem is that the request gives the model no audience, no tone, no format, and no boundaries.
A prompt that works usually includes five things:
- Audience: who is reading, such as a first-time visitor or a regular who orders every Thursday.
- Context: what the business is, where it operates, and what the piece is for.
- Format: word count, headline style, bullet points, or a fixed structure.
- Voice: plain, friendly, local, or formal. Local references like neighborhoods or island weather can make copy feel specific.
- Limits: what must not appear, such as health claims or language aimed at minors.
Once we started writing prompts with those five parts, the first drafts needed far less rewriting.
Compliance comes first
Cannabis marketing is regulated, and the rules are set at the state level. Hawaii’s medical cannabis program has its own requirements for dispensaries and delivery, and those rules change over time. Before using any AI-generated copy, check it against the current guidance from the Hawaii Department of Health and your licensing conditions. Do not rely on a tool or a template to know the law.
In practice, we build a standing instruction into every prompt that covers the basics:
- No claims that a product treats, cures, or prevents any medical condition.
- No imagery, slang, or themes that appeal to people under the legal age.
- No promises about effects, timing, or potency beyond what the lab testing record states.
- A reminder to include the required disclaimers in the footer of every post.
Treat AI output as a first draft that a human reviews. That review step is not optional for a licensed business.
A practical workflow for menu descriptions
Menu descriptions are the easiest place to start because the facts are fixed: the product type, the lab-tested potency, the ingredients, and the packaging size. Here is the workflow we use:
- Paste the verified facts first. Copy the lab sheet numbers and ingredient list directly from your inventory system. Never ask the model to estimate potency.
- Specify the length and structure. For example, two sentences for the menu card and a longer paragraph for the website.
- Name the voice. Ask for a calm, informative tone rather than hype.
- Add the compliance rules from the list above.
- Edit for accuracy. Check every number and remove any sentence that sounds like a promise.
This process turns a vague writing task into something closer to a checklist, and it keeps the facts under your control.
Where a prompt library fits in
Building a good prompt from scratch takes practice, so many teams begin by studying examples. A marketplace where prompts are listed with their intended use, the tool they were written for, and sample outputs makes it easier to see what a strong instruction looks like. If you are comparing options, the PromptMart prompt marketplace is one place to browse listings by category and read how each prompt is described before you buy or adapt it. Treat any purchased prompt the same way you would treat a template from a vendor: adapt it to your own facts, your state rules, and your brand voice rather than copying it word for word.
Social posts that stay on the right side of the rules
Social platforms add another layer of restriction, and cannabis content is often limited or removed. We use AI to draft captions around non-product themes, such as local events, community cleanups, delivery-day logistics, and customer service tips. Product posts are reviewed more carefully and kept factual.
A useful prompt for this purpose asks the model to produce three caption options, each under a set character count, with no product claims and no emojis that could be read as targeting younger audiences. We then pick the most neutral option and rewrite the closing line in our own words.
Email and loyalty messages
Email is often where a delivery business builds repeat customers. Here AI helps most with structure: subject line options, a greeting that uses the customer’s first name, a short explanation of a seasonal menu change, and a clear unsubscribe line. Ask the model for two subject lines with different lengths so you can test them, and keep the body short. Customers who opted in for updates usually want practical information, not long essays.
Common mistakes to avoid
- Trusting invented details. AI tools sometimes produce specific-sounding numbers or studies. If a figure is not in your own records, remove it.
- Using one prompt for everything. A menu card, a blog post, and a text message need different instructions.
- Skipping the audit. Save a log of what was published and who approved it. This is useful if a regulator or platform asks questions.
- Letting the draft set the voice. If every post reads the same, rewrite the opening sentence yourself. Your local knowledge is the part a generic model cannot supply.
Keeping the human touch
For a Honolulu delivery service, the most valuable content is specific: which neighborhoods get same-day drops, how the driver handles a rainy Kailua afternoon, what the team recommends for a first visit. AI can help organize those ideas, but the details should come from the people who run the business. A prompt is a tool for getting to a clean draft faster, not a substitute for knowing your customers.
Getting started this week
If your team is new to structured prompts, pick one recurring task, such as weekly menu updates, and write a single prompt with the five elements described above. Run it three times, track how much editing each draft needs, and refine the instruction until the edits get smaller. Then add the compliance block and make it a standard part of your process. Within a few weeks you will have a small library of instructions that fit your shop, your state rules, and your voice.
Leave a Reply