You can tell within two messages whether a chatbot was designed by someone who thought about the person using it. The good ones feel like a helpful front-desk person. The bad ones feel like a locked door with a cheerful sign. The difference is rarely the underlying technology. It's the design choices around it.
Most of those choices are small and easy to get right once you know what to watch for. Here are the ones that actually move the needle, framed as the do's worth copying and the don'ts worth avoiding.
Do be honest about what it is
Start every conversation by making clear the user is talking to a bot. A simple line does it: "Hi, I'm the automated assistant for Maple Street Clinic. Ask me anything and I'll bring in a human if I get stuck."
This isn't just polite, it's strategic. Setting the expectation up front means users judge the bot as a bot, which is a bar it can clear. Pretend it's a human, and every limitation becomes a lie the user catches. Honesty buys you patience. Deception spends it.
Don't fake a personality it can't back up
A quirky name and a wink are fine, right up until the bot can't answer a basic question. Then the cutesy tone reads as mockery. Personality should be a light seasoning on top of competence, never a substitute for it.
Match the voice to your brand and your customer's mood. A bot handling billing disputes should sound calm and clear, not bubbly. One for a skate shop can be looser. But whatever the voice, keep it out of the way of the answer. Nobody in a hurry wants three sentences of charm before the shipping date they asked for.
Do lead with the answer
Structure every reply so the useful part comes first. The customer asked a question. Give them the answer, then the context, then the next step. Buried answers waste the one thing chat is supposed to save: time.
Compare these two replies to "what's your return window?"
Weak: Thanks so much for reaching out! We really value your business and want to make sure you have a great experience. Our team has worked hard on a customer-friendly policy, and I'm happy to share that you can return items within 30 days.
Strong: You've got 30 days from delivery to return it, unused with tags. Want the link to start one?
Same information. The second respects the reader. Front-load the answer and let everything else be optional.
Don't trap people with no way out
The fastest way to make someone hate your chatbot is to corner them. They ask something it can't handle, it doesn't understand, and there's no button, no email, no human, no exit. Just a loop of "I didn't quite get that."
Always build a clear path to a person or a next step. Here's a checklist for an escape hatch that works:
- A visible way to reach a human, not buried three menus deep
- The bot admits when it's stuck instead of guessing or looping
- It captures the question and contact info so nothing's lost in the handoff
- A realistic follow-up time, like "someone will reply by noon"
- The full conversation reaches the human, so the customer doesn't repeat themselves
In SpideyChat you'd wire this handoff so an unanswered question becomes a ticket with the transcript attached, instead of a dead end. The user leaves feeling helped, even when the bot itself couldn't finish the job.
Do design for the messy way people type
Real users don't type like a help doc. They use typos, slang, fragments, and two questions crammed into one sentence. A bot designed only for clean input falls apart on contact with actual humans.
Test with the mess on purpose. "do u ship 2 canada" should work as well as "Do you offer international shipping?" If your bot only handles the polished version, it'll fail most of the time, because most people don't write polished. Design and test for how people actually talk, not how you wish they would.
A quick way to pressure-test this is to hand the bot to someone who didn't build it and watch them use it without coaching. You'll see them type in ways you never anticipated, abandon a sentence halfway, ask two things at once. Every stumble they hit is one a real customer would hit too, except the customer won't tell you about it, they'll just leave. Ten minutes of watching a fresh person poke at your bot surfaces more design problems than an hour of reading your own tidy test scripts.
Don't overwhelm with buttons and menus
There's a temptation to guide every conversation with a wall of quick-reply buttons. A few can help, especially for common paths. Too many turn a conversation back into the phone tree you were trying to escape.
Use buttons where a fixed choice genuinely helps: pick a date, choose a department, confirm yes or no. Let open questions stay open. If someone can just type "when does my order arrive," don't force them to tap through Orders, then Status, then Recent. The whole point of conversational design is that people can ask directly.
A quick reference
Here's the short version to keep near your bot's first draft:
| Do | Don't |
|---|---|
| Say it's a bot up front | Pretend to be a human |
| Lead with the direct answer | Bury it under pleasantries |
| Give a clear path to a human | Trap users in a loop |
| Test with typos and slang | Only handle clean input |
| Use a few buttons where useful | Rebuild the phone tree |
| Match tone to the situation | Force charm into every reply |
Where a small business lands this
Take Ridgeway Outfitters, a fictional gear shop. Their first bot opened with a paragraph of brand voice and no way to reach a human, and customers bounced. They rebuilt it around these rules: a one-line honest intro, answers first, and a "talk to a person" option always visible. Same underlying bot, far better reception. People stayed because the design respected their time and never left them stranded.
If there's a thread running through all of these, it's respect for the person on the other end. They came with a goal, usually a small one, and every design choice either moves them toward it or gets in the way. Honesty, answers first, a clear exit, testing with real mess: each is just a different way of saying you value the customer's time more than your own cleverness.
Good conversational design isn't about clever tricks. It's about being honest, fast, and easy to escape when needed. Draft your bot's first few replies, read them out loud as if you were a rushed customer, and cut anything that gets between the person and their answer. That single habit will fix most of what goes wrong.