A good AI chatbot setup takes four steps: give the bot your real business content, set rules for what it may and may not say, test it with real customer questions, and review what it gets wrong every week after launch. Skip any of these and the bot will either guess or give up.
This checklist is for small business owners who are deciding whether a chatbot is worth it and want to know what the work involves before they start.
When a chatbot helps, and when it does not
A chatbot is worth setting up when the same questions arrive again and again. Opening hours, delivery times, return rules, prices, how to book. If your inbox or phone is full of these, a bot can answer them at any hour and pass the rest to a person.
A chatbot does not help much when:
- Most questions need a judgment call, like custom quotes or complaints.
- Your information changes daily and nobody will keep the bot updated.
- You get only a handful of questions a week. A good FAQ page may be enough.
- Customers mostly want to talk to a specific person they already know.
If your questions are repetitive and your answers are stable, read on.
Step 1: collect the right content
The bot can only be as accurate as the content you give it. Gather these first:
- FAQ pages and any saved replies your team already uses.
- Policies: shipping, returns, cancellations, privacy, warranty.
- Product or service data: names, descriptions, prices, sizes, availability.
- Booking and contact details: hours, locations, links to forms or calendars.
- Past questions: a sample of real customer emails or chat logs, with personal details removed.
Leave these out:
- Drafts and old versions of policies. The bot cannot tell which one is current.
- Internal notes, staff names and anything you would not say to a customer.
- Customer personal data. The bot does not need it to answer general questions.
- Marketing copy full of claims you cannot back up. The bot will repeat them as fact.
One clean source is better than five overlapping ones. If two documents disagree, fix that before setup.
Step 2: set the rules
Rules shape how the bot behaves when the answer is not obvious. Write them down in plain sentences:
- Tone. Friendly and short? Formal? Give two or three sample replies you like.
- Topics it must not answer. Medical, legal or financial advice, pricing for custom work, anything about other customers.
- What to say when it does not know. “I’m not sure about that. Let me connect you with the team” is better than a guess.
- Handover to a person. When does the chat go to a human? Angry customers, refund requests, or any question it fails twice. Where does it go: email, a help desk, or live chat?
- Data to collect. If the bot hands over, should it ask for a name and email first?
These rules matter more than which AI model sits underneath. A bot without rules will sound confident about things it should not touch.
Step 3: test with real questions
Before launch, build a test set of 20 questions. Take most of them from real customer messages. Include:
- The ten questions you get most often.
- Five questions phrased badly, with typos or slang.
- Three questions the bot should refuse or hand over.
- Two questions about things you do not offer, to see if it invents an answer.
Run every question and read every answer. An engineer checks each one against your source content, looking for:
- Made-up facts. A price, date or policy that is not in your content.
- Outdated facts. An answer drawn from an old document.
- Missed handovers. The bot tries to handle something it should pass on.
- Tone problems. Too long, too stiff, or too casual for your brand.
Fix the content or the rules, then run the whole set again. Repeat until every answer is right or correctly handed over. Keep this test set: you will reuse it every time you change the bot.
Step 4: launch and watch
Launch on one page first, such as your contact or help page, before you add the bot site-wide. Then set a weekly routine:
- Read the list of questions the bot could not answer.
- Add missing answers to your content, or decide they should always go to a person.
- Check a sample of answered chats for accuracy.
- Update the bot when prices, hours or policies change.
- Rerun the 20-question test after any big change.
Fifteen minutes a week keeps a bot useful. Without this routine, answers drift out of date and customers lose trust in it.
What an AI chatbot setup costs
Our chatbot setup starts at $399 and takes about 5 days. AI drafts the bot’s knowledge and rules from your content, and an engineer tests it against real questions and checks every answer for made-up facts before you see it. The price is fixed before work starts. Details are on our AI chatbots page, and you can compare it with other tasks on pricing.
If your team also needs an internal assistant that answers from company documents, see our AI knowledge base service.
FAQ
How long does an AI chatbot setup take?
About 5 days for a standard setup: content gathering, rules, testing and launch. Larger bots with many integrations are quoted on the brief.
Will the chatbot make things up?
Any AI model can, which is why testing matters. Limiting it to your own content, giving it a clear “I don’t know” rule and checking answers by hand keeps this under control.
Is my business data used to train AI models?
No. We use business AI services that do not train on client data. Your files are used only for the task and deleted on request, and we sign an NDA if you ask.
Can the chatbot hand a conversation to my team?
Yes. You decide when it hands over and where the chat goes, such as email or your help desk.
Ready to set up a chatbot that answers from your real content? Start a Task and get a fixed price within 6 business hours.
