Chatbot UX Design for Error Moments: How to Write the Replies for When Things Go Wrong
Chatbot UX design usually focuses on the happy path, where the visitor asks a clear question and gets a perfect answer. Real conversations spend much of their time off that path: misunderstood messages, questions the bot can’t answer, upset customers, and nobody available to help. This guide is about designing for those moments.
The bot checks its guess instead of guessing silently.
β‘ In short
- Plan replies for seven error moments, not just for questions you expect.
- When unsure, the bot should check its guess (‘Do you meanβ¦?’) rather than answer the wrong question.
- Every ‘I can’t’ needs an alternative: a person, a page or a callback.
- Write error replies in your normal voice; error moments are where tone matters most.

Why errors matter
People remember how a bot behaves when it fails
Think about the last frustrating automated system you used. You probably don’t remember the parts that worked. You remember being stuck in a loop, being told “I didn’t understand” three times, or getting an answer to a question you didn’t ask. Failures are memorable; smooth successes are not.
That’s why error moments deserve deliberate design. Conversation designers call this “repair”: the ways a conversation recovers when understanding breaks down. Humans do it naturally (“Sorry, do you mean the blue one?”). Bots do it only if someone designs it.
The good news is that error moments are predictable. Almost every failure in a website chat falls into one of seven types. Write a planned reply for each, and your bot will handle the bad moments better than most. This is the most valuable piece of chatbot UX design a small business can do, and it fits in an afternoon.
The seven error moments
Every failure fits one of these types
Click each to see the moment, the bad reply, a better reply, and the principle behind it.
The moment: the bot genuinely can’t tell what the visitor wants.
Bad: “I’m sorry, I didn’t understand that. Please rephrase.”
Better: “I didn’t quite catch that. Is it about an order, delivery, or something else? Or I can get a person for you.”
Principle: narrow the options instead of asking the visitor to start again, and always offer a person.
The moment: the bot has a likely interpretation but isn’t certain.
Bad: answering the guessed question confidently, and getting it wrong.
Better: “Do you mean changing the address on an order you’ve placed? I can help with that.”
Principle: check a guess before acting on it. One extra question beats a wrong answer.
The moment: the question is clear, but the bot doesn’t have the information.
Bad: inventing an answer, or “I can’t help with that.”
Better: “I don’t have that detail. Our team can confirm it; they usually reply within an hour. Want me to pass it on?”
Principle: never invent. Say what you don’t know, then offer the path to someone who does.
The moment: the visitor asks something the bot shouldn’t answer: medical, legal, financial advice, or a disputed refund.
Bad: a cheerful answer to keep the conversation going.
Better: “That’s best answered by your doctor, as it depends on your situation. I can share the product details, or have our team call you.”
Principle: decline clearly, explain briefly why, and offer something useful you *can* do.
The moment: the visitor is annoyed, angry or repeating a complaint.
Bad: “I understand your frustration. Please select from the options below.”
Better: “I’m sorry, that’s not okay. I’m getting a person from our team now. Could you share your order number so they can start straight away?”
Principle: acknowledge once, act quickly, and hand over with the details collected.
The moment: the visitor needs a person, and it’s outside opening hours.
Bad: “An agent will be with you shortly” (when nobody is coming).
Better: “Our team is offline until 9 am tomorrow. Leave your email or number and they’ll reply first thing.”
Principle: tell the truth about timing, and turn the request into something that will be followed up.
The moment: the visitor has asked the same thing twice and the bot keeps missing.
Bad: the same unhelpful reply a third time.
Better: “Sorry, I’m clearly not getting this right. Let me pass you to a person who can help.”
Principle: after two misses, stop trying and hand over. A loop is the most damaging moment in any chat.
Writing principles
Six rules for writing replies in error moments
The tabs above show the replies. These rules explain how to write your own for situations specific to your business.
1. Own the problem, briefly
“I didn’t catch that” is better than “Your message was unclear”. The bot takes responsibility. One short apology is enough; three in a row sound insincere.
2. Always give a next step
An error reply that ends with nothing to do is a dead end. Offer options, a person, a page or a callback.
3. Narrow, don’t restart
“Please rephrase” asks the visitor to do all the work again. Offering two or three likely topics lets them continue with a tap.
4. Keep the visitor’s words
If you’re checking a guess, use their words: “Do you mean changing your address?” They’ll recognise their own question immediately.
5. Match the emotion, then move on
A frustrated visitor needs acknowledgement before information. An unsure visitor needs reassurance. A neutral visitor needs neither; just fix the issue.
6. Stay in your voice
Error replies are often written in a stiffer, more formal tone than the rest of the bot, because they feel like “system messages”. Keep them in your normal voice. The guides on positive scripting and sounding human have many examples.
Script bank
Error replies you can copy
Adapt the details, keep the shape.
Designing the process
A five-step chatbot UX design sprint for error moments
You don’t need a design team to do chatbot UX design well. One person who knows your customers can work through these steps in an afternoon, with a short follow-up after two weeks.
- Collect failures. Read the last fifty bot conversations and highlight every moment where the conversation went wrong. Sort each into one of the seven types.
- Count. Which types happen most? For most small businesses, “not understood” and “no answer known” top the list.
- Write replies. For each type, write a reply using the six principles above. Specific to your business: your team’s name, your opening hours, your most common topics.
- Set up. Add the replies to your bot: fallback answers, handover rules, away messages and opening hours. In LiveChatBot, opening hours and the away form handle moment 6; handover rules and the AI Agent‘s instructions handle the rest.
- Test and review. Deliberately trigger each error moment from your phone. Then, two weeks later, read fifty new conversations and see whether failures are rarer and recoveries smoother.
Fix the cause too
Error replies make failures less painful; they don’t stop them happening. While collecting failures, note the knowledge gaps behind them and fill those as well. The guide on training your AI chatbot covers that side.
Before and after
One conversation, designed twice
The same visitor, the same question, with and without designed error moments.
Wrong answer, then a loop, then a lost customer.
Recognised as a problem, handed over with details, clear promise.
Measuring
How to know your error design is working
Two numbers tell you most of what you need. The first is how often visitors leave right after a bot reply; better repair should reduce it. The second is how often a visitor asks the same thing twice; good guess-checking and handover after two misses should make loops rare. Our chatbot UX metrics guide explains how to count both in a few minutes a week.
Beyond numbers, read the handed-over chats. If your team keeps having to ask visitors to repeat information, the bot isn’t collecting the right details before handing over. If visitors arrive at the handover already angry, the bot is trying too long before giving up.
Where to start today
If you only have an hour, write replies for moments 1, 3 and 6: not understood, no answer known, and nobody available. Those three cover most failures on small-business sites. Set up opening hours so moment 6 is honest, and everything else can follow. Every piece of this runs on LiveChatBot’s free plan; start on livechatbot.in.
Frequently asked questions
What is repair in chatbot UX design?
Repair is how a conversation recovers when understanding breaks down: checking a guess, narrowing options, admitting a gap or handing over to a person.
How many times should a bot try before handing over?
Two misses is a good rule. After the second failed attempt on the same question, hand over to a person or collect contact details.
Should the bot apologise when it fails?
Briefly, once. Then offer a next step. Repeated apologies without action annoy people more than a short, practical reply.
Is it okay for the bot to say ‘I don’t know’?
Yes, and it’s much better than inventing an answer. Pair it with a way to get the answer from your team.
Do these replies need AI?
No. Fallback messages, handover rules and away messages work with or without AI. AI helps the bot make better guesses in the first place.
Keep reading
Related guides
Chatbot UX Metrics: 8 Numbers That Tell You Whether Your Bot Is Helping or Annoying People
Chatbot User Interface on Phones: Designing for Thumbs, Keyboards and Slow Networks
Chatbot User Experience, Mapped: The 7 Moments a Visitor Goes Through and How Each One Feels
Design your seven error moments this week
Read fifty chats, write seven replies, and set them up free on livechatbot.in.