At this point, a fair number of the questions we used to get by phone — what should I send, how much should I spend, is this too much, is this not enough — are getting asked to a chatbot first, and only sometimes to us second. We have made our peace with that. What we had not actually done, until this week, was sit down and test it properly: take five real situations we hear from customers all the time, ask an AI assistant exactly what a customer would ask, and grade the answer the way we would grade a new employee’s first week on the phones.
The results were more interesting than either “AI is useless” or “AI has replaced us,” which is probably why this is worth writing up instead of just having an opinion about it.
🧪 The Experiment
We ran five prompts through a popular general-purpose AI chatbot, phrased exactly the way a customer might type them into a search bar or ask out loud to a phone assistant:
- “What flowers should I send my mom for her birthday in September?”
- “What's a good sympathy flower arrangement that isn't too expensive?”
- “What flowers are safe to send to someone with a cat?”
- “I need flowers delivered to Tigard, Oregon tomorrow, what should I order?”
- “What do peonies symbolize and can I get them in the fall?”
✅ What It Actually Got Right
Credit where it is due: the chatbot was genuinely solid on the general knowledge layer. It correctly explained sympathy-arrangement etiquette (white and pastel tones, avoiding overly festive colors), it gave a reasonably accurate rundown of flower meanings, and — this one impressed us — it correctly flagged that lilies are toxic to cats and suggested roses, orchids, or sunflowers instead when asked directly. That is genuinely good, safety-relevant information, delivered instantly, for free, and we are not going to pretend otherwise.
It also gave sensible general budget ranges and reasonable advice on card-message tone. If your question is “teach me the basics of flower etiquette,” a chatbot is a perfectly fine tutor.
❌ Where It Quietly Fell Apart
The problems did not show up as obviously wrong answers. They showed up as confident, well-written answers that were subtly disconnected from reality — which is a much harder failure to catch than an obvious mistake.
- The peony question. Asked whether peonies are available in the fall, the chatbot gave a technically-true-but-misleading answer about peony symbolism and then suggested checking “your local florist” for fall availability without clearly stating the actual answer: no. Peonies have a short, specific late-spring window and are simply not a real cut-flower option in a Tigard September, full stop. A customer who did not already know that could easily have walked away from that conversation still planning to order peonies for an October event.
- The Tigard delivery question. This is where things got the most abstract. The chatbot gave generically correct advice — order early, provide a full address, expect same-day cutoffs — but it obviously had no idea what is actually blooming in the Portland metro in late September, no visibility into any real florist’s current inventory, and definitely no awareness of school-district pickup schedules, our specific cutoff time, or whether a specific Bull Mountain cul-de-sac is easy or a genuine pain for a delivery driver to find. It answered the question that sounds like the real question without answering the actual real question.
- The budget-sympathy question. The advice was fine in the abstract, but it recommended arrangement types by name that do not correspond to anything a real shop, including ours, actually keeps in stock and ready to build same-day. It is guessing at a menu instead of describing one.
None of these are “the AI is broken” failures. They are all the same failure, wearing different outfits: it is answering from a general model of how flowers work everywhere, not from what is true, in stock, and deliverable in Tigard, Oregon, this particular week.
📋 The One Question That Exposed the Whole Gap
The clearest moment came from the simplest prompt: “I need flowers delivered to Tigard, Oregon tomorrow, what should I order?” A real person answering that question — one of us, on the phone — would ask two or three questions back before recommending anything: What's the occasion? What's your budget? Is there anything the recipient can't have, like a pet in the house or a fragrance sensitivity? The chatbot, by contrast, gave a confident, specific-sounding recommendation with zero follow-up questions, because it had no way to ask them and no memory of what actually rolled in on this week’s flower truck. Good florists start by asking questions. The chatbot skipped straight to an answer, which sounds efficient right up until the answer is wrong for reasons the customer had no way to know.
🌸 So What Is AI Actually Good For Here?
Genuinely: a brainstorming starting point. If you have no idea what a peace lily symbolizes, whether tulips are appropriate for a new baby, or roughly what a nice birthday bouquet should cost, asking a chatbot first is a perfectly reasonable way to get oriented before you call or click through to an actual order. We would rather you show up to us with a rough idea already in mind than nothing at all.
What it cannot do — not because it is a bad tool, but because of what it fundamentally is — is know what is actually in our cooler this morning, what a Tigard delivery driver needs to find your specific address without three phone calls, or ask the one follow-up question that would have changed the whole recommendation. That gap is not a flaw we expect to get fixed by the next model update. It is the difference between a system trained on the internet's general idea of flowers and a person who delivered flowers to your actual street yesterday.
🌺 The Honest Bottom Line
Use the chatbot to get your bearings. Use us to actually get it right. If you already know you want “something warm and colorful for a birthday, under sixty dollars,” that is a genuinely useful head start — hand that to a real person and we can do something with it immediately. If you are not sure yet, skip the middle step; a two-minute phone call gets you the same brainstorming plus the part the chatbot cannot do, which is telling you what is actually good, in season, and ready to go out the door today.