A fallback message is what a chatbot says when it cannot match what the visitor typed to anything it was built to handle. Every chatbot has one, whether or not anyone wrote it on purpose, and it appears at the moment a visitor is most likely to give up. This guide covers when fallbacks happen, what a good one says, examples you can adapt, and exactly what an Alma chatbot does when nothing matches.
The fallback is one small part of planning, writing and testing a chatbot, and it is worth writing on purpose rather than leaving to a default.
When a chatbot falls back
A fallback is not one event. It happens in several different situations, and each one needs a slightly different reply.
- The visitor types instead of tapping a button. The chatbot offered three choices, and the visitor wrote a question of their own.
- An answer fails a check. The chatbot asked for an email address and got "tomorrow", or asked for a number and got a sentence.
- The question is outside what the chatbot covers. A plumbing company's chatbot is asked about roofing, or about a job the business does not take.
- An AI chatbot does not have the answer. The question is reasonable, but the information the chatbot was given does not cover it.
- The conversation has already ended. The visitor finished a flow, then typed something new.
Before you write a single fallback message, write down which of these can happen in your chatbot. A chatbot built mostly from buttons meets the first and the last far more often than the others.
What a good fallback message does
Two ways an Alma chatbot answers
Every plan
Flows you write
- You write each question, button and answer in the flow builder.
- The bot says only what you wrote, in the order you set.
- Best for the questions you can predict: hours, services, booking and contact details.
Optional paid monthly add-on
AI answering
- You give our team your business information as text, up to 50,000 characters, and we load it into the bot.
- The bot answers typed questions from that information alone.
- It does not read uploaded files or search your website.
A good fallback does four things in two or three short sentences.
- It admits the problem plainly. "I did not understand that" is honest. Pretending to understand and answering a different question is worse than saying nothing.
- It says what the chatbot can do. Visitors do not know the chatbot's limits. Name the two or three things it handles, such as booking, prices and opening hours.
- It offers a next step. Buttons work better than "please rephrase", because a visitor who did not know what to type the first time still does not know.
- It gives a real way to reach a person. That can be a phone number, an email address, or a question that collects the visitor's contact details so someone can reply.
Tone matters as much as content. The message should never suggest that the visitor did something wrong. "That is not a valid option" reads as a correction, while "Here is what I can help with" reads as an offer. If you have not yet settled how your chatbot should sound, getting your chatbot's tone right covers it, and the fallback is the best place to check that the tone holds up.
Fallback message examples
These examples are written for a made-up home services company. Change the details to match your business, and keep each one short enough to read at a glance in a small chat window.
When the visitor types instead of tapping
I can help fastest if you choose one of the options below. If your question is about something else, choose "Something else" and I will take your details for our office.
When an answer fails a check
That does not look like an email address. Please check it and type it again, in the form name@example.com.
When the question is outside what the chatbot covers
I am not able to answer that here. I can book a service visit, give price ranges for common jobs, or take your details so someone from our office can call you back.
When an AI chatbot does not have the answer
I do not have that information. You can reach our office at 630-555-0100 on weekdays from 8 am to 5 pm, or leave your details here and we will call you.
When the conversation has ended
Is there anything else I can help with? Choose an option below to start again.
Each one says what the chatbot can do and ends with something the visitor can tap or type. None of them asks the visitor to guess what the chatbot wants to hear.
Never promise a person who is not there
Where a lead goes after the chat
- A visitor opens the chatThe widget appears on every page where you pasted the snippet.
- The bot asks your questionsOne at a time, with buttons for the common answers.
- Alma saves the leadIn your Leads inbox, with tags from the visitor's answers.
- You get an emailAs soon as the visitor leaves an email address or phone number.
Where else the lead can go
- Every plan
- Email notification
- Growth and up
- Zapier, and CoreCNX once your account is linked
- Agency and up
- Webhooks to your own systems
The most damaging fallback is one that promises help that does not arrive. A message such as "Connecting you to an agent" followed by silence is worse than having no chatbot at all, because the visitor waits instead of calling.
If nobody watches the chat in real time, say so, and say what happens instead: who will reply, how and when. If your chatbot collects contact details for a callback, tell the visitor when to expect it in terms of your real working hours. If a phone number is the quickest route, give the number.
The same applies to AI chatbots. An AI that answers from your business information should be told what to do when the answer is not there: say that it does not know, and give your contact details. Without that instruction, it may try to answer anyway.
Change course after the second miss
One fallback is normal. Two in a row usually means the chatbot cannot help with what the visitor wants, and showing the same message a third time is likely to lose them.
Plan a different second message. The first fallback offers the options again. The second skips the options and asks for contact details, or gives the phone number outright. A visitor who has been misunderstood twice usually wants a person, and the most useful thing the chatbot can do is get them to one.
How Alma handles a message it does not expect
If you build your chatbot in Alma, this is what happens in each situation, so you can write for it.
- Typing at a Buttons or Quick Replies step. Alma shows the same question and its buttons again. It does not try to match the typed words to a button. Give every set of buttons a choice such as "Something else" that leads to a Collect Input step, so a visitor with a different question has somewhere to go.
- An answer that fails a check. A Collect Input step can check that an answer is an email address, a phone number or a number. When an answer fails, Alma shows the step's Error Message and asks again. Write your own Error Message in the step's settings, or Alma uses a short default such as "Please enter a valid email address:".
- A message after the conversation has ended. A chatbot created from a template starts its opening flow again from the first step, so the visitor sees the opening options rather than an answer to what they typed. A chatbot started from scratch runs its default flow when the chat opens, and a message typed after that flow ends gets no reply from the flows. In both cases the simplest fix is to end every path with a Buttons step, such as "Anything else?" with "Start over" and "Contact us", so the conversation never reaches a dead end.
- AI answering. AI answering is an optional add-on for $25 a month that the Alma team sets up. It answers typed questions from the business information you provide, up to 50,000 characters, so the most useful line in that text is the one that says what to do when the answer is not there. The team can place it as a step inside a flow, or set it to answer typed messages that arrive when no conversation is running on a chatbot that has no opening flow to restart. If AI answering cannot reply, for example after the month's AI answers are used up, the chatbot says it was not able to answer, suggests rephrasing or using the contact details on your website, and leaves the AI step, so the visitor's next message is handled the same way as a message after the conversation has ended.
Alma has no live chat, so a person cannot take over the conversation. When a visitor needs your team, the chatbot should ask for a name and an email address or phone number in Collect Input steps whose variables are named email or phone. Alma then saves the lead and, once you have saved a notification address under Settings, Integrations, emails it to that address so someone can follow up.
Test every fallback before launch
Fallbacks are the part of a chatbot that rarely gets tested, because the person building it knows which buttons to press. Before you launch, try each of these on your own website:
- At every set of buttons, type a question instead of tapping, and read what comes back.
- At every question with a check, type a wrong answer on purpose.
- Finish each flow, then type something new.
- Ask one question your chatbot was never built for, such as a service you do not offer.
After launch, read the conversations each week for the first month. The messages that led to a fallback show you which buttons are missing and which answers to add. Looking at chatbot examples by industry also helps, because a flow that already offers the common questions as buttons needs its fallback far less often.
Common questions
What is a fallback message in a chatbot?
It is the reply a chatbot gives when it cannot match the visitor's message to anything it was built to handle. It is also called a default reply or a fallback response.
What should a chatbot fallback message say?
It should say plainly that the chatbot did not understand, name what it can help with, offer buttons or another clear next step, and give a real way to reach a person.
How many times should a chatbot fall back before it offers a person?
Twice at most. After a second miss, stop offering the same options and give the visitor your phone number, or ask for their contact details so your team can reply.
Can an AI chatbot avoid fallbacks altogether?
No. An AI chatbot can answer more varied questions, but it still meets questions its information does not cover. Its fallback is the instruction to say that it does not know and to point the visitor to your team.
To start from flows whose questions and buttons are already written, browse the chatbot templates and add a "Something else" choice to each set of buttons. When you are ready, start the 14-day free trial.
