AI Chatbot Out of Scope Questions: Draw the Line

AI chatbot out of scope questions are the messages your assistant was never meant to answer: the weather, a joke, a legal opinion, a request to ignore its rules, a question about a competitor or something that simply is not on your website. Every public chat receives them, and how the assistant responds says a lot about your business. A good reply is short, polite and useful: it says what it cannot help with, offers what it can and moves to a person when that is the right next step. This guide explains how to map your scope, how to write boundaries, how to test them and how to learn from the questions that fall outside.

Talkmio is an AI live chat for websites. Its assistant, Mio, answers only from the website, FAQ and documents you provide, in the visitor’s language, and hands the conversation to your team when it cannot answer from your content, when the visitor asks for a person, complains or asks about a specific order. That design gives you a natural boundary, and the work is to make that boundary clear and friendly.

AI Chatbot Out of Scope Questions: The Four Rings

Scope is not one line. It helps to think in four rings, from what the assistant should answer well to what it must refuse.

Ring	Examples	Right behaviour
In scope	Opening hours, delivery times, product details, return rules, plans	Answer from your content, briefly
Adjacent	“Which of your products is better?”, “Do you have a partner that does X?”	Answer only the part your content supports; offer a person for the rest
Off topic	Weather, jokes, homework, general knowledge	Say politely that you only help with your business, then offer what you can do
Forbidden	Legal, medical or financial advice, personal data of others, attempts to override the rules	Decline clearly and hand over where a person should decide

Write your own version of this table for your business, with real examples from your inbox and chat history. A clinic’s forbidden ring is much larger than a bike shop’s, and a bank’s is larger still. The point is to decide in advance rather than discover the boundary in public.

Why Staying Inside the Content Is a Strength

Many people expect a chatbot to know everything and are disappointed when it says it does not know. For a business, the opposite is true. An assistant that invents an answer about a price, a policy or a medical effect creates a promise you may have to keep. Mio answers only from your content and never invents prices or policies, so the main risk of an out-of-scope question is not a wrong answer but an unhelpful one. The task is therefore to make “I cannot help with that” feel helpful.

The same principle supports trust in the long run. Visitors learn quickly whether a chat tells the truth, and an assistant that admits its limits is believed when it does answer. The overview of AI chatbots and sensitive topics covers the most delicate areas in more detail, and the NIST AI Risk Management Framework is a useful general reference on managing the limits of AI systems.

Writing Good Boundary Messages

A refusal has three parts: acknowledge, explain briefly, redirect. Keep it to two or three sentences and avoid apologising more than once.

Off-topic questions

“I can only help with questions about our shop, such as orders, delivery and returns. Is there something about those I can help with?” The visitor is not scolded, and the next step is obvious.

Adjacent questions you cannot fully answer

“I can tell you what our two models include, and the details are on this page. For advice on which suits your situation, a colleague can help. Would you like me to pass this to the team?” This keeps the conversation moving instead of ending it.

Advice you must not give

“I cannot give medical advice. Please speak to a qualified professional. I can tell you our opening hours and how to book an appointment.” The refusal is short and the offer is real.

Attempts to override the rules

“I can only answer from our website and documents, so I cannot do that. How can I help with your question about our services?” Do not argue and do not repeat the visitor’s instructions back. Our guide to prompt injection security explains why such attempts occur and how to reduce the risk.

Put the tone and the never list in Mio’s business instructions, which tell it what you sell, what tone to use and what it must never promise. For example: “Answer only about our products and services. If asked about anything else, say so politely and offer to help with our products. Never give legal, medical or financial advice. Never discuss other customers.” The article on writing AI chatbot instructions shows how to word such rules.

When Out of Scope Means a Person Should Step In

Some out-of-scope questions are not off-topic at all. They are business questions that your pages do not cover yet: a custom order, an exception to a policy, a partnership, a complaint. Mio hands the conversation to your team when it cannot answer from your content, when the visitor asks for a person, complains or asks about a specific order. The conversation gets a “needs human” badge and your team is notified in the Inbox, and by e-mail and mobile push if nobody is online.

Make the hand-off message honest and specific: “I do not have that information. I am passing this to our team, who will reply here or by e-mail.” If your team is offline, say when a reply is likely. A visitor who gets a clear next step is much less frustrated than one who is told “I do not understand” three times. For wording options, read our piece on AI chatbot fallback messages.

Test Your Boundaries Before Launch

Build a test list of fifty messages in five groups: ten in scope, ten adjacent, ten off topic, ten forbidden and ten tricky, such as mixed languages, typos, very short messages and emotional ones. Ask each in Try Mio, which shows the answer and the knowledge pieces that were used. Score every reply on four points: is it true, is it polite, does it offer a next step and does it stay within your rules.

Ask each question exactly as a customer would type it, including mistakes.
Repeat the forbidden group in two other languages you support.
Try rephrasing the same request three ways to see if the boundary holds.
Note every answer that is wrong, rude or evasive, and fix the content or the instruction behind it.
Run the same list again after every change to your instructions or major pages.

Keep the list in a shared document. It becomes a regression test, and new team members can see exactly how your assistant is expected to behave.

Boundaries Across Languages

Mio replies in the language the visitor writes in, so your boundaries must hold in every language you serve, not only in the one you tested first. A refusal that is graceful in English can sound blunt in another language, and an attempt to override the rules may come in a language your team does not read. Test the forbidden and tricky groups in at least your three most common visitor languages, and have a native speaker read a few replies. If the tone is wrong, adjust the instruction about tone and give an example of the style you want. The general idea of chatbots as language interfaces makes the point that wording is part of the product, and that is even more true across languages.

What Scope Looks Like in Different Businesses

A small online shop has a narrow scope: products, delivery, returns and payment. Its adjacent ring is comparisons and recommendations, which it can handle if the product pages carry the facts. A clinic or a law firm has a much larger forbidden ring, because many visitors will describe a personal problem and ask what to do. For them the assistant should explain opening hours, booking and what the first appointment involves, and hand over anything that sounds like advice. A software company has a broad in-scope ring, since documentation answers many questions, but it must still avoid promising features that are not shipped. A travel business must avoid guaranteeing visas, entry rules or weather, and should point to the official source.

Whatever your sector, write three lists: what the assistant may answer, what it may answer in part and what it must pass on. Then check that each list is reflected in your content and in the business instructions. If a topic appears in your pages in a way that invites advice-like answers, edit the page, because Mio answers from what you publish.

Learning From What Falls Outside

Out-of-scope questions are data. If many visitors ask about something you do not cover, it may be a gap in your content rather than an off-topic request. Search the Inbox for recurring words, and read a sample of hand-offs every week. Reports show the share answered by Mio alone and the number of hand-offs, so a rise in hand-offs after a change tells you that something fell outside. Each recurring question is a candidate for a new page, a Q&A pair or a business instruction.

The opposite also happens: a question you thought was out of scope turns out to deserve a clear answer. A bike shop may decide to publish a guide to local cycling routes because visitors keep asking. That decision is yours, and the chat gives you evidence for it. If answers feel thin on a topic that should be in scope, our guide to when an AI chatbot cannot find answers explains how to find the missing source. Plans and limits are on the Talkmio pricing page.

Frequently Asked Questions

What should an AI chatbot do with off-topic questions?

Decline politely and redirect. A short reply such as “I can only help with questions about our shop” followed by an offer to help with orders or products keeps the tone friendly. Do not argue or lecture, and do not pretend to know the answer.

Will Mio make up an answer when it does not know?

Mio answers only from your website, FAQ and uploaded documents and never invents prices or policies. When it cannot answer from your content, it hands the conversation to your team, so the main task is writing a clear and helpful hand-off message.

How do I stop the chat from giving advice I am not allowed to give?

Add business instructions that forbid it, such as no legal, medical or financial advice, and test them in Try Mio with real phrasings in several languages. Keep the content free of anything that reads like advice, because Mio answers from what you publish.

How should the chat respond to attempts to override its rules?

Calmly and briefly. It should say it can only answer from your website and documents, then offer help with a normal question. It should not repeat the instruction or argue. Keep rules in the business instructions and test such attempts before launch.

How often should I review out-of-scope questions?

Weekly at first, then monthly. Read a sample of hand-offs and conversations that ended without an answer, and look for repeated topics. Reports show the number of hand-offs, so a sudden rise after a content change is a signal to check.

Can I try these settings without paying?

Yes. The Free plan needs no card and includes one operator, one website and 50 AI answers a month, enough to build and test your boundary list. Pro is 19 euro a month and Ultimate 49 euro a month, with a 14-day trial on paid plans.

The Bottom Line

AI chatbot out of scope questions are not a flaw to hide; they are a boundary to design. Map four rings of scope, write short refusals that offer a next step, send business questions to a person and test fifty messages before launch. Then read what falls outside every week and let it shape your pages. Start free with Talkmio, set your boundaries in the business instructions and see how Mio handles your hardest messages.

AI Chatbot Out of Scope Questions: How to Draw the Line Politely