Running a cannabis delivery service in Seattle means writing a lot of copy under tight rules. Menu descriptions, order confirmations, driver texts, weekend promos, and FAQ pages all have to be clear, accurate, and careful about what they promise. When our small team started looking for help, many of us tried a chatgpt prompts for sale marketplace before writing our own. The experience taught us more about prompts than about marketing, and this post shares what we learned.
Why most AI prompts fail for regulated businesses
The first prompts we tried were generic. Something like “write a fun product description for a cannabis gummy” produced copy that sounded like every other shop online. Worse, it often included health language, dosage promises, or phrases like “relaxes your body” that we could not stand behind. In a regulated category, a clever sentence that breaks the rules is a liability, not a win.
A prompt that works for a regulated business usually does four things:
- It names the audience and the channel, such as a text message to an existing customer versus a homepage banner.
- It sets hard boundaries, such as no medical claims, no reference to effects, and no content aimed at minors.
- It asks for a specific format, such as 160 characters or three bullet points.
- It tells the model what to do when information is missing, such as asking a question instead of guessing a potency level.
Once we started building prompts with those four parts, the output became something we could actually edit and approve in a few minutes instead of rewriting from scratch.
A practical starting set for delivery operators
These are the categories where we found the most day-to-day value. Adapt the wording to your own license and the rules that apply to you.
Order confirmation and status texts
Customers in Seattle often order during rain, after work, or late at night, and they want to know three things: what is coming, when it will arrive, and who to call. A prompt that specifies those three items, a friendly tone, and a rule to never promise an exact arrival minute produces texts that cut down on “where is my order” messages.
Driver handoff notes
Drivers need short, plain instructions: building name, gate code note, whether the customer prefers a call at the door. We ask the model to turn messy order notes into a five-line summary, and we have it flag any field that looks incomplete. The flagging feature alone has saved us from several failed drops.
Menu descriptions without health claims
This is where caution matters most. We ask for descriptions that cover flavor, format, package size, and the product’s general category, and we explicitly forbid words about effects, healing, or comparisons to pharmaceuticals. Every draft still goes past a human reviewer who checks it against current state guidance before publishing.
Neighborhood-aware promotions
Seattle is not one market. Capitol Hill, Ballard, Georgetown, and South Lake Union have different rhythms, and our promotions reflect that. We give the model a short profile of each service area and ask for a promo idea tied to local context, such as a weekday lunch offer near office areas. We never ask it to invent sales numbers or testimonials, because those would be fabricated.
How to judge a prompt before you pay for it
If you are considering buying prompts rather than writing your own, look for these signals. A good listing shows the intended use, the inputs it expects, and sample outputs that are specific rather than flashy. It should say what the prompt does not do. Be wary of listings that promise guaranteed results, because no prompt can guarantee how a model will respond on a given day.
We also recommend testing any purchased prompt with your own real examples before it goes live. Run it five or six times with different inputs, including messy ones with typos or missing addresses. If the outputs stay consistent and within your rules, it is worth keeping. If they drift, the prompt needs tightening or the product is not a fit.
For operators who want a curated place to browse, a marketplace of tested, reviewed prompt templates can be a useful starting point, especially when you are comparing several options for the same task. Treat any template as a draft, not a finished policy.
Compliance is still your job
No prompt replaces legal review. Washington’s cannabis rules cover advertising, age verification, packaging, and delivery requirements, and they can change. Before you publish anything AI-assisted, check it against the current requirements from the Washington State Liquor and Cannabis Board and, where needed, a qualified attorney. We keep a short internal checklist that every piece of customer-facing copy must pass: no health claims, no appeal to minors, accurate product information, and clear licensing details where required.
We also log which prompts produced which published content. If a rule changes, we can find every affected asset in minutes instead of hunting through old drafts.
A workflow that fits a small team
Our process is simple. One person owns the prompt library and updates it when rules or products change. Another person runs drafts through the model and marks anything uncertain. A third person, usually the shift lead, approves final copy before it goes live. This keeps the speed benefit of AI without letting a single unchecked output reach a customer.
We store prompts in a shared document with a version number, the date last reviewed, and a note on known weak spots. When a prompt starts producing awkward phrasing, we record the failing input so we can fix the instructions rather than just editing the output.
What we would tell another delivery shop
Start with the least risky tasks, such as order confirmations and driver notes, before moving to anything public-facing. Write your own boundaries first, then build prompts around them. Expect to iterate several times. And remember that the goal is not clever copy; it is clear, accurate communication that helps a customer get the right order at the right door.
AI prompts are a tool, and like any tool they reward careful handling. If you approach them with the same discipline you bring to inventory and driver safety, they can save real hours each week. If you approach them as a shortcut around the rules, they will eventually cost you more than they save.
If you have built prompts for your own delivery operation, we would like to hear what worked and what failed. Share your experience with your local team, and keep testing every change before it reaches the people you serve.
Leave a Reply