When our dispatch lead first asked an AI tool to write a product description, the result read like a pharmacy leaflet written by a robot. Nothing about it sounded like us, and nothing about it respected the rules we operate under. That experience is what pushed our team to look more carefully at how prompts are written, tested, and shared, and it is also why we started paying attention to resources like ai art prompts and other tested instruction sets. The lesson we took from it was simple: a prompt is a tool, and like any tool it only works when it is specific, reviewed, and used for the right job.
Why Most Prompts Fail Small Delivery Businesses
Most of the prompts floating around online were written for general audiences. They ask for things like ‘write a fun blog post about edibles’ or ‘make this sound better.’ For a cannabis delivery service, that kind of request creates three problems right away. The output tends to be generic, it often drifts into claims that could create compliance trouble, and it rarely reflects the tone you have built with local customers over time.
We noticed that the prompts that worked best for us shared a few traits:
- They named the audience, such as adult customers who already know the product category.
- They stated the format clearly, such as a 40-word menu blurb or a three-question FAQ.
- They listed what to avoid, including health claims, dosage promises, and anything that implies the product is safe for minors.
- They included a sample of our brand voice so the model had something real to imitate.
None of this is complicated, but it is easy to forget when you are busy running a shift.
Building a Prompt Library Around Compliance
The biggest shift for us was treating prompts like standard operating procedures. Before anything goes live, a prompt gets written down, tested against three or four real scenarios, and reviewed by whoever owns that part of the business. Dispatch owns prompts for order status messages. Our product manager owns menu descriptions. Our compliance contact reviews anything customer-facing that mentions effects, strength, or purchase limits.
We also keep a simple log with four columns: the prompt itself, the date it was last reviewed, the person responsible, and a note on the last change. When a state rule or local guidance shifts, we know exactly which prompts need attention. That log has saved us from reusing old wording that no longer fit current requirements.
One caution: prompts do not replace legal advice. Cannabis advertising and marketing rules are set by state and sometimes by local authorities, and they change. Anything you publish should be checked against current requirements with someone qualified to read them. Think of an AI draft as a starting point for a human reviewer, never as the final word.
Where Prompts Earned Their Keep
After a few months of trial and error, we found several places where a well-built prompt saved real time without creating risk.
Menu Descriptions
We feed the model the product name, category, growing or production notes we have approved, and a list of banned phrases. The prompt asks for a short, plain-language description with no effect claims. A reviewer then edits the result. What used to take twenty minutes per item now takes about five, and the copy is more consistent across the menu.
Customer FAQ Drafts
Our most common questions involve delivery windows, ID checks at the door, and how to change an order. We wrote a prompt that instructs the model to answer in two sentences, to avoid speculating about exceptions, and to direct anything unusual to a human. Staff review the drafts before they go into our help center.
Staff Training Scenarios
New drivers need practice handling tricky situations, like a customer who appears intoxicated or a recipient who is not on the order. We use prompts to generate role-play scenarios, then have experienced staff rewrite the responses to match our actual policy. The AI produces the variety; the policy stays human-authored.
How to Evaluate a Prompt Before You Use It
Whether you write prompts yourself or buy them from a marketplace, the same checks apply. Here is the process we use:
- Read the prompt as if you were the model. Would the instructions leave room for something you would never say?
- Run it three times with different inputs. Consistency matters more than a single impressive output.
- Check every output for claims, guarantees, and language that could appeal to minors.
- Confirm the prompt states its intended use. A prompt written for a blog post should not quietly end up on a product label.
- Record who approved it and when.
This sounds like a lot for a few sentences of text, but the review step is what separates a useful tool from a liability.
A Practical Way to Start
If your delivery business is thinking about adopting AI writing tools, start small. Pick one repetitive task, such as order status texts, and write a prompt for only that. Test it for two weeks. Have someone outside the project read the outputs. Only expand once the first prompt is boring in the best sense: predictable, on-brand, and compliant every time.
Many teams also find it useful to look at how other people structure their instructions. Reading a range of tested prompts, including those in image generation and creative writing, helps you notice patterns like explicit constraints, example outputs, and clear output formats. You can see how one curated marketplace handles this by visiting PromptMart, where prompts are shared with context about what they are meant to do. Borrow the structure, not the wording, and always adapt it to your own rules and voice.
What We Still Do by Hand
It is worth being honest about the limits. We still write our own policy language, our own compliance statements, and any message that involves a customer’s health or legal situation. AI drafts never go directly to a customer who has a complaint or a question about eligibility. Those conversations belong to people who know our business and the law.
The prompts have made our small team faster and more consistent, but they have not made judgment optional. The best results we have seen come from pairing a clear prompt with a careful reviewer who knows exactly what the business is allowed to say.
Final Thoughts
AI prompts that actually work are specific, reviewed, and tied to a real job. For a cannabis delivery business, that means building them around compliance from the start, logging changes, and keeping humans in the approval chain. If you do that, a prompt becomes less of a novelty and more of a dependable part of how your team operates.
Leave a Reply