If you run a cannabis delivery operation in Calgary, you already know how much writing fills the day: order confirmations, product descriptions, FAQ answers, driver updates, social captions, and the occasional apology when a route runs late in a cold snap. Many owners are turning to AI tools to carry some of that load, and the teams getting real value tend to start with custom ai prompts built around their own workflow instead of typing vague requests into a chat window and hoping for the best.
Why most AI prompts disappoint
Most disappointing AI output has the same root cause: the request was missing context. Asking a tool to ‘write a product description for a sativa pre-roll’ gives you something generic, often with claims you would never want to publish. A prompt that works specifies who the audience is, what the brand voice sounds like, what information must be included, what must be left out, and what format the output should take.
A useful prompt for a delivery business usually contains five parts:
- The role the assistant should play, such as a customer service writer for a licensed retailer.
- The exact task, with a clear input like a product name, a delivery window, or a customer question.
- Constraints, including word count, reading level, and banned phrases.
- Facts the assistant must use, pasted in directly so it does not invent details.
- An output format, such as three bullet points, a two-sentence reply, or a table with set columns.
When you test prompts this way, you stop judging them by how polished they sound and start judging them by whether a staff member can use the result with minimal editing.
Start with the compliance layer
Cannabis is a regulated product in Canada, and marketing rules are strict at both the federal and provincial level. In Alberta, retailers and delivery services operate under provincial oversight, and the rules on promotion, packaging, and how products can be described are not something an AI tool understands on its own. Any prompt you use for public-facing copy should include a compliance instruction, but that instruction is only a starting point. Your legal or compliance advisor should review the templates you adopt, and you should keep a record of which version went live.
In practice, that means writing prompts that explicitly tell the model to avoid health or therapeutic claims, avoid language that appeals to minors, avoid lifestyle imagery descriptions that would not be allowed in your jurisdiction, and stick to verified product information from your own catalogue. A simple rule helps: if a sentence would make a regulator or a worried parent uncomfortable, it does not belong in the prompt’s output.
Calgary-specific needs that generic prompts miss
Generic prompt collections were written for generic businesses. A Calgary delivery service has different pressures. Customers may be ordering during winter evenings when roads are slick, during long weekends, or during events that shift traffic patterns across the city. Your order-status messages need to be clear about delays without sounding alarming. Your neighbourhood coverage messages need to be accurate, which means your prompt should take the current zone list as input rather than relying on the model’s memory.
Build a small set of prompts around these recurring moments:
- A delay notice that gives a realistic window and an option to reschedule.
- A weather-related message for a sudden cold snap or heavy snowfall.
- A substitution notice when an item is out of stock, with the replacement listed explicitly.
- A courtesy message for customers who ordered at an unusual hour.
Each of these should be short enough to read on a phone and warm enough to feel human. Ask the prompt to produce three variants, then have a staff member pick the one that best fits the moment.
Product descriptions that stay accurate
Product copy is where AI errors do the most damage. A model can confidently describe terpene profiles, effects, or strain lineage that do not match your actual inventory. The safe approach is to treat the model as an editor rather than a source. Paste in the verified specification sheet from your supplier, including THC and CBD percentages, weight, format, and any lab-tested fields you are permitted to show, then ask the model to rewrite that information in your brand voice without adding anything new. To go deeper, explore The marketplace for AI prompts that actually work.
Good instructions for this task include a phrase like ‘use only facts present in the input; if a detail is missing, write nothing about it.’ That single line prevents a large share of the invented claims that cause problems. Review every output against the source sheet before it reaches your website or your delivery app.
Support replies that sound like your team
Customer support is the other place where prompts pay off, because the questions repeat. Refund requests, delivery address corrections, questions about age verification at the door, and confusion about order totals all arrive every week. A solid support prompt includes your actual policy text, the tone you want, and a rule that the assistant must escalate anything involving a complaint about safety, a suspected minor, or a legal question to a human.
Keep a library of approved replies and update it when policies change. The goal is consistency: a customer who writes in on Tuesday should receive the same answer about a missed delivery as one who writes on Saturday, whether the reply was drafted by a person or by a tool and then approved.
How to tell whether a prompt is working
Judging a prompt by first impressions is misleading. Test it on real examples from the past month, including messy ones. Ask a teammate who did not write the prompt to run it and report what they had to change. Track edit time as well as output quality: if staff spend nearly as long fixing a draft as writing from scratch, the prompt needs more constraints or better source material.
Versioning matters too. When you improve a prompt, save the old version and note what changed. This makes it easier to explain a decision later and to roll back if a new version introduces a problem.
Where a prompt marketplace fits
Writing a complete prompt library from nothing takes time, and many small operators do not have a spare person for it. Browsing a marketplace can speed up the early stage, because you can compare how different authors structure role, constraints, and format. The caution is that a prompt written for a software company or a restaurant will not automatically fit a regulated product, so every purchased or borrowed prompt should be adapted to your compliance rules and your catalogue before use.
Treat any shared prompt as a draft. Read it line by line, delete anything that promises outcomes you cannot deliver, and add the specific facts and policies that make it yours.
A simple starting checklist
- Write one prompt for each recurring message type, starting with delays and substitutions.
- Add a compliance instruction to every prompt that touches public copy.
- Require the model to use only facts you supply.
- Set an escalation rule for safety, age, and legal topics.
- Have a second person review outputs for the first thirty days.
- Save versions and record who approved each one.
- Revisit prompts whenever provincial rules, your menu, or your delivery zones change.
AI tools will not replace the judgment your team brings to a delivery route or a difficult customer call, but they can take the repetitive writing off your plate if the prompts behind them are specific, tested, and honest about their limits. Start small, measure edit time, and let the results decide which prompts earn a permanent place in your workflow.

Leave a Reply