Plenty of small cannabis delivery operations in Quebec are testing AI tools for customer replies, product descriptions and order-status messages, and most quickly discover the same problem: the same question gets a different answer every time. The quickest fix is to start from prompt engineering templates that have already been written, tested and refined, rather than improvising a new instruction each time someone on the team needs a draft.
Why generic prompts fail in a regulated niche
Asking a chatbot to ‘write a cannabis product description’ sounds harmless, but the output often includes language that is too promotional, makes health claims, or speaks to an audience that is not allowed to be targeted. In Quebec, the rules around cannabis promotion are strict, and a single careless sentence in a newsletter or social post can create real problems for a licensed business.
A good prompt for this sector does three things at once. It defines the role of the assistant, it sets clear boundaries on what cannot be said, and it specifies the format the team needs. Without those elements, the model fills the gaps with whatever sounds persuasive, which is the opposite of what a compliance-minded operator wants.
What a useful prompt template looks like
Think of a template as a fill-in-the-blank recipe. The fixed part contains the instructions: tone, length, forbidden topics, and the required disclaimer. The variable part contains the details that change from one message to the next, such as the product name, the delivery window or the customer’s question.
- Role: ‘You are a customer support writer for a licensed cannabis delivery service in Quebec.’
- Boundaries: ‘Do not make medical claims. Do not mention effects as treatments. Do not address anyone under 21.’
- Format: ‘Reply in under 80 words, in plain French or English depending on the customer’s language, with one clear next step.’
- Variables: order number, postal area, product category, and the customer’s exact question.
Keeping the fixed part separate from the variable part makes it easy for a second staff member to reuse the same template and get a consistent voice, even if they have never written a customer message before.
Practical uses for a delivery team
Order updates and delay notices
When a driver is running late because of weather or traffic, the team needs a short, calm message that does not overpromise. A template with fields for the new estimated window and the reason for the delay produces replies that sound human without requiring a manager to write each one from scratch.
Product education pages
Describing terpene profiles, strains or consumption methods requires care. A template can instruct the model to stick to factual, label-based information, to avoid superlatives, and to direct readers toward the product packaging and official information for dosage guidance. The result is copy that is informative without drifting into claims the business cannot support.
Age verification and policy questions
Customers often ask how ID checks work, what happens if nobody is home at delivery, or whether they can change an address after ordering. A standard template for these questions protects the business by making sure every answer matches the written policy. Review the policy text yourself first, then build the template around it rather than asking the model to invent rules.
Internal training notes
New drivers and support staff benefit from short scenario-based explanations. A template that turns your existing procedures into a quiz or a checklist can speed up onboarding, provided a manager signs off on the final version. To go deeper, explore The marketplace for AI prompts that actually work.
How to test and improve a template
Treat each template like a small piece of software. Write down the expected output, run the template with five or six realistic inputs, including awkward ones, and check the results against your policies. Note every place where the output strays, then tighten the instructions. Keep a version number and a short changelog so you know which edit caused an improvement or a regression.
It also helps to collect real customer questions, anonymized, and use them as test cases. A template that handles a mildly confusing question about delivery fees is more valuable than one that only works with perfect input.
Keep humans in the loop
No template removes the need for judgment. Anything that touches legal age, health, pricing or complaints should be reviewed by a person before it is sent. AI drafts are best used as a starting point that a trained team member edits, not as an automatic reply system, especially in a category with this level of regulation.
Set a simple rule: if a message would be uncomfortable to explain to a regulator or a parent, it does not go out without review. That single guideline catches most of the risky outputs before they reach a customer.
Building a small prompt library
Once a few templates prove reliable, store them in a shared document with clear names, such as ‘delay notice – evening’ or ‘FAQ – changing delivery address’. Add a note on when to use each one and who approved it. Over a few months, the library becomes one of the most useful operational assets the business owns, because it captures how the team wants to communicate in a single, searchable place.
If you are starting from nothing, browse a curated prompt library to see how other teams structure their instructions, then adapt those structures to your own rules and voice rather than copying them word for word. Borrowing the skeleton is useful; the policy details have to come from your own business.
Final thoughts
AI can save real time for a cannabis delivery team, but only when the prompts are written with the same care as the policies they reflect. Start with a handful of templates for your most common messages, test them against realistic cases, keep a person in the approval loop, and build the library gradually. Consistency, accuracy and a calm tone will do more for customer trust than any clever wording ever could.
Leave a Reply