If you run a cannabis delivery operation in Vancouver, you have probably tried an AI writing tool for product descriptions, a delivery-window announcement, or a reply to a customer asking where their order is, and got back something bland and vaguely off-brand. The problem is rarely the tool. It is usually the prompt. Browsing an ai prompt marketplace is one way to see how other people structure their requests, but the real skill is learning to adapt those structures to a regulated, fast-moving business like ours.
Why generic prompts fall flat for delivery businesses
A prompt like ‘write a product description for a sativa pre-roll’ will produce copy that could belong to any retailer on earth. It will not reflect your product range, your customer base, or the rules that govern what you are allowed to say. Cannabis is a sector where wording matters, and the vague, enthusiastic tone many AI tools default to can create real problems if it drifts into health claims or promotional language that regulators do not permit.
Good prompts for this industry do three things at once. They give the model a specific role, they supply the constraints it must respect, and they describe the audience in enough detail that the output sounds like a human who knows the neighbourhood wrote it.
Build your prompts around four inputs
Before you write any prompt, collect the information a skilled copywriter would ask for. In practice that means:
- The product facts. Strain or product type, format, THC and CBD levels as listed on your packaging, and the size. Paste these exactly as they appear on the label rather than letting the model guess.
- The audience. Are you writing for first-time buyers, experienced customers looking for a specific effect profile, or people ordering for a special occasion? Each group reads differently.
- The channel. A product page, an SMS, a social post, and an in-app notification have different length limits and different expectations.
- The guardrails. List the words and claims you will not use. This is where most of your compliance thinking belongs.
When you put those four inputs into one prompt, the output stops being generic. You can then ask the model to produce three variations so you can compare tone rather than accepting the first draft.
A sample structure for product descriptions
Here is the kind of structure that tends to work well, described in plain language rather than as a fixed script, because your exact wording should reflect your own licence conditions and brand voice:
- Set the role: you are a copywriter for a licensed retailer offering delivery in British Columbia.
- Provide the verified product facts and tell the model to use only those facts.
- State the audience and the channel, including the character limit.
- Prohibit health, medical, or therapeutic claims, and prohibit any language that targets people under the legal age.
- Ask for a plain, factual tone with one sentence on how the product is typically chosen by customers, not how it will make them feel.
- Request a short list of the claims the model used, so you can check each one against your packaging.
That last step matters more than it looks. Asking the model to show its claims gives you a quick audit trail, and it makes review faster for whoever on your team signs off on copy.
Prompts for customer communication
Delivery businesses live or die on messages: order confirmations, driver arrival notices, delays caused by weather on the Lower Mainland, and apologies when a product is out of stock. These are good candidates for reusable prompts because the situations repeat.
For a delay message, a useful prompt gives the model the expected window, the reason in one sentence, the next step the customer can take, and a tone limit such as ‘apologetic but brief, no more than three sentences.’ For a stock-out notice, ask for two alternatives drawn from a list you supply, so the model does not invent products. Keep a short library of these prompts in a shared document and update them after each busy weekend, when you will notice which messages generated complaints. To go deeper, explore The marketplace for AI prompts that actually work.
Keep a human in the loop
AI output should never go live unreviewed, especially anything that describes a product or makes a promise to a customer. Assign one person to check copy against your current product data and your regulatory obligations. Check the provincial rules yourself and confirm them with your licensing contacts, because advertising restrictions can change and the details are not something to delegate to a language model.
Review also protects your brand voice. Over time you will notice patterns in what the model gets wrong for your business, such as overusing a particular adjective or misnaming a strain. Write those corrections into your prompts so the errors do not come back.
Testing whether a prompt actually works
A prompt that works is one that produces usable copy with minimal editing, consistently, across different products. To test a prompt, run it on five products of different types and time how long each edit takes. If one prompt needs heavy rewriting every time, the constraints are probably too loose or the product facts are incomplete. Adjust one variable at a time so you know what changed the result.
Track the results in a simple sheet: the prompt version, the product, the edit time, and whether anything had to be removed for compliance. After a month you will have a clear picture of which prompts earn a place in your workflow and which ones should be retired.
Where to start this week
Pick one recurring task, such as product descriptions for your new arrivals or delay notices for delivery drivers. Write a prompt using the four-input structure above, run it three times, and review the outputs with a colleague. Save the version that performs best, then move on to the next task. Within a few weeks you will have a small, tested library that fits your business rather than a collection of generic templates.
The goal is not to let software write your brand. It is to give your team a faster starting point so that people spend their time on the judgment calls that matter: accuracy, compliance, and the tone that makes a regular customer trust your next order.
Leave a Reply