Most clinics that say “we tried a chatbot and it didn’t work” didn’t get a broken product. They got one that worked exactly as designed — and that design doesn’t match how patients actually type.
If you have ever put a chatbot on your Facebook page and ended up switching it off, this will sound familiar. A patient writes: "For the clear aligner promo you're running now, how many installments can I pay in? And do I need a cleaning first?" The bot replies: "Please choose a topic: 1. Promotions 2. Branch locations 3. Contact admin"
That one question holds three things — which promo is still valid, the installment terms, and what has to happen first — but the system only sees a message that doesn't match any button.
The limits built into button bots
1. It doesn't read. It matches.
Traditional bots run on keywords and buttons set up in advance. Whoever builds it has to guess what customers will type, then write a path for each guess. The problem is that people ask the same thing a dozen different ways — "How much is a thread lift?" "Price for fishbone threads?" "I want some lifting, is 20,000 baht enough?" — and every phrasing nobody guessed is a dead end.
2. The conversation doesn't connect
A patient asks the price of a health check-up package, gets an answer, then types "What about the one for men?" A button bot has no idea what "the one" refers to, because each message is treated on its own. A person reads it and knows straight away the patient is still on the same topic.
3. The more you add, the more it breaks
The only way to make the bot answer more is to add more buttons and conditions. As the flowchart grows, whoever maintains it can no longer remember which path ends where. Eventually nobody dares touch it, and the promo that changes every month stays stuck on last month's.
4. It doesn't know what it doesn't know
For a clinic this is the most dangerous one. A keyword bot always picks the "closest" answer. A customer asks about an allergic reaction and gets a promotion back. That isn't funny — it's your clinic's credibility.
So how does an AI Bot work differently?
The real difference isn't that it's "smarter". It's how it gets to an answer. An AI Bot reads what the customer actually typed, checks it against the clinic knowledge you've given it, and writes a reply in plain language. It doesn't walk a flowchart drawn in advance.
What that means in practice:
- It understands however people ask — it can answer questions nobody set up beforehand, as long as the answer is in the clinic's information
- It answers several things in one message — price + preparation + a slot this Saturday, all in one reply, without making the patient ask one question at a time
- It follows the conversation — it remembers which program you're talking about
- You add knowledge without rebuilding a flowchart — want it to answer something new? Add the information. No new buttons.
What an AI Bot doesn't fix on its own
This part matters, because plenty of vendors talk as if adding AI solves everything. The reality:
- Poor information in, poor answers out — if the clinic has never written down its prices or conditions anywhere, the system has nothing to answer from. The real setup work is gathering the clinic's knowledge completely, not flipping a switch.
- Without guardrails, it will overreach — a system that writes its own answers can also say too much. For a clinic, the lines to set on day one are: no diagnosis, no promised results, no recommending drugs or dosages, no discounting on its own.
- Some cases need a person, full stop — price negotiation, allergic reactions, unhappy customers. A good system knows when it has reached the point to stop and bring in your team, and hands over everything it has collected.
Side by side in one table
| Topic | Menu-button chatbot | AI Bot |
|---|---|---|
| Customer asks off-script | Dead end, loops back to the menu | Answers if the information exists |
| Several questions in one message | Catches only one | Answers all of them in one reply |
| Continues the conversation | No | Yes |
| Adding new topics | Add buttons / extend the flowchart | Add information |
| Updating this month's promo | Edit the flowchart | Tell it in plain language |
| Topics with no information | Gives the closest match | Says it doesn't have the information, then hands off to a person |
| Risk of answering out of scope | Low — it can only say what you wrote | Real, unless clear guardrails are set during setup |
In short: a button bot is safe because it can only say what you wrote — but it can also only answer what you guessed in advance. An AI Bot answers far more, in exchange for you drawing clear lines on what it can say and what goes to a person. For a healthcare provider, those lines aren't an extra. They're part of the setup work, done from day one.
AIB Clinic builds AI Bot chat replies for clinics and healthcare providers of every specialty, with guardrails set from day one.
Talk to our team