Most small delivery operators in Spokane have already tried an AI chatbot at least once. The first attempt usually goes something like this: someone types a vague request, gets a generic paragraph back, and decides the tool is not worth the time. The problem is rarely the software. It is the prompt. If you want to use gpt prompts for business that produce usable results on the first try, you need a library of tested, specific instructions rather than improvised requests typed at 9 p.m. during a rush.
Why delivery businesses need prompts, not just tools
A cannabis delivery company runs on a surprisingly repetitive set of tasks. You write product descriptions for a menu that changes weekly. You answer the same questions about delivery windows, ID checks, and minimum order sizes. You send order confirmations, handle reschedules, and try to keep a handful of regular customers engaged without crossing the lines that state regulators care about.
Repetitive work is exactly where prompts pay off. A good prompt sets the role, the audience, the constraints, and the format in one place. Once you have that prompt saved, the output stays consistent from one week to the next, and a new team member can produce work that sounds like the rest of your brand.
The three categories worth building first
Rather than trying to automate everything, start with three buckets where mistakes are cheap to catch and time savings are obvious.
1. Menu and product copy
Menu descriptions are tedious because they need to be accurate, short, and scannable on a phone screen. A useful prompt tells the model what information you already have, such as strain type, terpene profile from the supplier sheet, package size, and price tier, and then asks for a fixed-length description with no health claims.
The key instruction is restraint. Tell the model explicitly not to promise effects, medical outcomes, or anything the packaging itself does not state. Then review every output against the supplier documentation before it goes live.
2. Customer service replies
Delivery questions cluster around a few themes: where you deliver, how long it takes, what happens if nobody answers the door, and how to adjust an order after checkout. A prompt that includes your actual policies as reference text produces far better replies than a prompt that asks the model to ‘be helpful.’
- Paste the exact delivery zone description and cutoff times into the prompt.
- Ask for a reply under a set word count, with one clear next step for the customer.
- Require the model to say ‘I don’t know, let me check’ when the policy text does not cover the question.
3. Staff training notes
Driver onboarding is another good candidate. Instead of rewriting the same checklist every quarter, use a prompt that turns your current policy document into a one-page quiz with answers. Have a manager verify the output against the real policy before handing it to new drivers.
Writing prompts that survive contact with reality
Many prompts fail for the same reason: they leave out information the model needs. Before you save any prompt to your library, run it through a short checklist.
- State the job. ‘You are writing a text message to a customer whose order is out for delivery’ is far more useful than ‘write a message.’
- Supply the facts. Include the estimated window, the driver’s first name, and the customer’s order number placeholder, so the model is not guessing.
- Set hard limits. List the words or claims that are off limits, such as medical claims, comparisons to pharmaceuticals, or pricing promises you cannot keep.
- Define the output. Specify length, tone, and whether you want a subject line, bullet points, or plain sentences.
- Ask for a self-check. Add a final line asking the model to list any statements that might need legal or compliance review.
That last step is not a substitute for human review, but it surfaces risky phrasing quickly and makes your review faster.
Compliance is a human job
Cannabis advertising and communication rules vary by jurisdiction, and in Washington they are set and enforced by the state liquor and cannabis board. Rules can change, and enforcement details matter. Treat any AI-generated marketing copy as a draft that needs review against current state requirements and, where appropriate, your attorney’s guidance. An AI tool does not know your license conditions, your age-verification workflow, or which promotions your regulator has flagged in the past. To go deeper, explore The marketplace for AI prompts that actually work.
This is also why a prompt library is more valuable than a single clever prompt. When you save a prompt, save the compliance notes next to it: which claims it must avoid, which disclaimers it must include, and who signs off before publication. Over time, that record becomes an internal policy document that new staff can learn from.
Building a shared prompt library for your team
A library works only if people can find things. Organize prompts by task rather than by tool. A simple structure might look like this:
- Menu copy, with separate prompts for flower, edibles, and accessories
- Customer messages, grouped by order status: confirmed, out for delivery, delayed, and completed
- Driver training, including onboarding quizzes and incident-report summaries
- Social and email, with prompts that exclude any content directed at minors or that resembles a promotion your license prohibits
Give every prompt a name, an owner, a last-reviewed date, and a short note describing what a good output looks like. When a regulation changes or a policy updates, the owner revises the prompt and logs the change. That discipline keeps the library trustworthy.
Measuring whether prompts are helping
Do not judge prompts by how impressive the output sounds. Judge them by how much editing they require. Track a simple number for each prompt: how many minutes a staff member spends fixing the result before it goes out. If a prompt consistently needs heavy rewriting, the instructions are probably missing a constraint or a fact. If it needs almost no edits, you have found something worth keeping.
You can also compare response times before and after adopting a prompt for customer messages. Use your own order log as the baseline. Avoid claiming large efficiency gains until you have measured them yourself.
A practical first week
If you are starting from zero, here is a realistic plan that does not require new software or a budget line item:
- Pick one repetitive task, such as delivery-status texts.
- Write three prompts with different levels of detail and test them against real past orders.
- Keep the version that needs the fewest edits and has no compliance issues.
- Save it with the constraints, facts, and review owner included.
- Repeat the process for one new task the following week.
Within a month, you will have a small, tested library tailored to your own menu, your own policies, and your own customers in the Spokane area. That library is harder for a competitor to copy than any generic template, because it reflects how your business actually runs.
The bottom line
AI tools are only as useful as the instructions behind them. For a cannabis delivery business, the best results come from specific, constrained prompts that match your policies and get checked by a person before anything reaches a customer. Start small, write prompts that state the facts and the limits, keep compliance notes with every template, and measure the editing time rather than the wow factor. Do that consistently, and AI becomes a reliable part of your operations rather than another experiment that gets abandoned after a week.

Leave a Reply