Motta Industries Blog

Adding a chatbot to a business website has become easy. Many platforms now let you paste in a few links, upload some documents and switch on an assistant that answers visitors around the clock. The hard part is not getting a chatbot to talk. It is getting it to say only things that are true about your business.

That is what grounding means. A grounded chatbot draws its answers from a defined set of your own content, such as service pages, policies, product details and help articles, rather than from whatever a general language model happens to have absorbed. Done well, grounding makes answers more accurate, more consistent with your published information and easier to correct. Done poorly, it gives a confident voice to outdated pages and contradictory documents.

This guide explains why grounding matters, how to prepare your content, how to set boundaries and how to keep an assistant honest over time.

Why it matters: your chatbot speaks for you

A well-known case from Canada shows the stakes. As CBC News reported in February 2024, British Columbia’s Civil Resolution Tribunal ordered Air Canada to pay compensation after its website chatbot told a grieving customer they could apply for a bereavement fare retroactively. The airline’s actual policy, explained on a different page of its website, said bereavement rates did not apply to completed travel.

The airline argued that the chatbot was responsible for its own actions. The tribunal member called that a remarkable submission and wrote that while a chatbot has an interactive component, it is still just a part of the airline’s website, and that it makes no difference whether information comes from a static page or a chatbot. The decision concluded that the airline did not take reasonable care to ensure its chatbot was accurate.

The amount ordered was modest, but the principle is the point: customers reasonably treat what a chatbot says as what the business says. This article is general information, not legal advice; if a chatbot will discuss prices, contracts, refunds or regulated topics, talk with qualified counsel.

Regulators are watching how chatbots serve customers

The concern goes beyond one tribunal. In a June 2023 issue spotlight on chatbots in banking, the Consumer Financial Protection Bureau warned that a poorly deployed chatbot can lead to customer frustration, reduced trust and even violations of the law. It estimated that approximately 37% of the United States population had interacted with a bank’s chatbot in 2022.

The bureau identified risks that apply well beyond banking. It said the information chatbots provide may not be accurate, that the technology may fail to recognize when a consumer is invoking their rights, and that it may fail to protect their privacy and data. It also noted that consumers can struggle to get the response they need, including an inability to reach a human representative, and said institutions should avoid using chatbots as their primary customer service channel when it is reasonably clear the chatbot cannot meet customer needs.

For a small business, those warnings translate into three design goals: accurate answers, clear limits and an easy path to a person.

Decide what the chatbot is for

Before choosing tools, write down the jobs the assistant should do and the ones it should not. Good starting jobs for a small business include:

• Answering common questions about hours, locations, services and turnaround times.
• Explaining published policies, such as shipping, returns or cancellations.
• Helping visitors find the right page, product or form.
• Collecting details for a quote or support request and handing them to staff.

Jobs to keep away from the chatbot, at least at first, include anything involving binding prices or commitments, account changes, payments, legal or medical guidance and complaints that need judgment. A narrow, reliable assistant is more useful than a broad one that occasionally invents answers.

Prepare the content it will draw from

A grounded chatbot is only as good as the content behind it. Most businesses discover that their website and documents contain gaps, duplicates and contradictions that a human reader quietly works around but an assistant will repeat.

1. Inventory your sources. List the pages, help articles, policy documents and product data the chatbot will use.
2. Pick one source of truth for each topic. If the returns policy appears in three places with different details, fix that before connecting anything.
3. Remove or exclude outdated material. Old promotions, retired services and past policies should not be in the chatbot’s knowledge base.
4. Write in plain, specific language. Clear statements such as “Orders ship within two business days” are easier to retrieve accurately than vague marketing copy.
5. Put important facts in text. Information locked inside images, PDFs with scanned pages or embedded videos is harder for any system to use.
6. Add dates and owners. Note when each page was last reviewed and who is responsible for keeping it current.

This work pays off twice. Google’s documentation on AI features and your website, last updated in December 2025, says site owners can apply the same foundational SEO best practices for AI features as for Search overall, including creating helpful, reliable, people-first content. Its list of worthwhile fundamentals includes making sure important content is available in textual form and that structured data matches the visible text on the page. Clear, current, well-organized pages help both search engines and your own assistant.

If your existing pages need rewriting for clarity, our copywriting team can turn scattered policies into consistent, plain-language content.

Choose a platform with the right controls

Chatbot tools range from simple widgets bundled with website builders to configurable platforms that connect to help desks and databases. When comparing options, focus less on how impressive the demo sounds and more on the controls you will need every day.

Ask whether you can restrict answers to your own content, see which source each answer used, update the knowledge base quickly and exclude specific pages. Check how conversation data is stored, how long it is kept, who at the vendor can access it and whether it is used to train other models. Confirm that you can export transcripts, set business hours for handoffs and switch the assistant off instantly if something goes wrong. A cheaper tool that lacks these controls can easily cost more in corrections and lost trust later.

Set boundaries in the configuration

Most chatbot platforms let you shape behavior with instructions and settings. Use them deliberately.

• Answer only from approved content. Instruct the assistant to rely on your knowledge base and to say it does not know when the answer is not there.
• Cite the source page. Showing a link to the page behind each answer lets visitors verify details and helps staff spot errors.
• Decline out-of-scope requests. Prices outside published ranges, legal advice and promises about specific outcomes should be redirected to staff.
• Hand off cleanly. Offer a visible way to reach a person at any point, and pass the conversation along so the customer does not have to repeat themselves.
• Protect personal information. Avoid asking for sensitive data in chat, tell visitors how conversations are stored and limit who can view transcripts.

Integrating a chatbot with forms, booking systems or a help desk usually needs development work. Our web development team connects assistants to existing systems while keeping those boundaries intact.

Design the experience around trust

How the chatbot appears on the page affects how people use it. Label it clearly as an automated assistant rather than giving it a human name and photo. Explain briefly what it can help with. Keep it from covering important content or interrupting visitors with repeated pop-ups. Make sure it works with keyboards and screen readers, and that the handoff to a person is just as accessible.

Good interface design also makes errors easier to report. A simple “Was this helpful?” control with a way to flag an incorrect answer gives you a steady stream of improvements. Our UI and UX design work treats the assistant as part of the overall customer journey.

Test before launch, and keep testing

Before launch, build a list of real questions from your inbox, phone calls and site search logs. Ask each one and compare the answer with your published information. Include tricky cases: questions about exceptions, deadlines, prices, competitors and topics the chatbot should refuse. Record every wrong or vague answer and fix the underlying content or configuration.

After launch, review a sample of conversations every week at first, then monthly. Look for wrong answers, missed handoffs, repeated questions the content does not cover and signs of frustration. Each time a policy, price or service changes, update the source content the same day and recheck related answers.

Prepare staff for handoffs

The people who receive handed-off conversations need a clear process. Decide who monitors chatbot requests during business hours, how quickly they should respond and what happens after hours. Give staff access to the full conversation so they can pick up where the assistant left off, and ask them to report any wrong answers they notice.

Measure whether it actually helps

A chatbot should earn its place on the site. Decide in advance how you will judge it. Useful measures include the share of conversations resolved without a handoff, the number of answers flagged as wrong, how often visitors ask for a person and whether support emails about common questions go down.

Compare those numbers with the time staff spend reviewing conversations and fixing content. If the assistant mostly repeats information visitors could find on a clear page, improving the page may be the better investment.

A fictional example

Consider a fictional plumbing company that added a chatbot to answer after-hours questions. In its first week, the assistant told one customer that weekend emergency visits had no extra fee, quoting an old promotion page that had never been removed.

The owner and an office manager reviewed the knowledge base together. They removed outdated pages, rewrote the pricing and service-area pages in plain language, and set the chatbot to link to its source and hand off any question about specific prices to staff. They also added a clear “Talk to a person” button. Over the next month, flagged answers dropped steadily, and the office spent less time correcting misunderstandings.

A grounded chatbot checklist

• The chatbot’s jobs and limits are written down.
• Source content is inventoried, current and consistent.
• Each topic has one source of truth.
• Important facts are available as text.
• The assistant answers only from approved content and admits gaps.
• Answers link to source pages.
• Handoff to a person is visible and preserves context.
• Personal information is protected.
• The assistant is clearly labeled and accessible.
• Conversations are reviewed regularly and content updated promptly.

A chatbot can be a helpful front door to your business, but only if what it says matches what you would tell the customer yourself.