Blog · · 5 min read

Will an AI Chatbot Make Up Prices? Keeping Answers Accurate

A 3D clay-style miniature shop counter with a small friendly robot holding a magnifying glass over a neat stack of price cards, a tiny shop owner figure beside it giving a thumbs up, and a pinned noticeboard behind them.

A salon owner tries a free chatbot on her website. A visitor asks how much a full head of highlights costs. The bot replies "around £120", which sounds reasonable. The problem is that she charges £145, and the visitor arrives expecting £120.

This is the fear most small business owners have about AI chat: that it will confidently tell customers something that isn't true. It is a fair worry. AI chatbot hallucination, where the bot makes up an answer, is a real issue with generic tools. The good news is that it is largely a setup problem, and there are clear ways to keep answers accurate.

Why generic chatbots make things up

AI language models are trained on huge amounts of text from the internet. They are very good at producing answers that sound right. When you ask one a question it doesn't have the facts for, it doesn't naturally stop. It produces the most likely sounding answer instead.

For a small business, that means:

  • Prices get filled in with what similar businesses seem to charge.
  • Policies like cancellations, deposits and refunds get filled in with what is "usual".
  • Areas and hours get guessed from the town name or general patterns.
  • Totals get calculated, sometimes wrongly, by adding up figures or applying discounts that don't exist.

None of this is the bot being sneaky. It is doing what it was built to do. The fix is to change what it is allowed to answer from, and what it does when it doesn't know.

How to keep answers accurate

1. Answer only from approved facts

The most important step is to limit the chatbot to your own information: your price list, your service pages, your policies and answers you have written. Not the internet, and not general knowledge about "typical" businesses.

Better still, you should be able to see each fact it uses and approve, edit or remove it. If your call-out fee changes from £60 to £65, you change one fact and every future answer uses the new figure.

2. Copy prices exactly

Prices should be copied word for word from your information, not rounded or estimated. "From £45" should stay "from £45", not become "about £40 to £50".

3. Show where each answer came from

When a chatbot shows its source, such as "From Prices" or "From your Services page", two things happen. Visitors trust the answer more, and you can check it quickly. If an answer is wrong, you know exactly which fact or page to fix.

4. Never calculate totals

Totals are a common trap. A visitor asks "How much for a cut, colour and blow dry?" and a bot adds three prices together. But maybe you offer a package price, or colour depends on hair length. A careful chatbot gives the individual prices you have listed and leaves the final figure to you.

For example, it might reply: "A cut and blow dry is from £40 and colour is from £65. For an exact price for both, the salon will confirm."

5. Hand over when unsure

The safest answer to an unknown question is "I don't have that information, but I can pass your question on." A good chatbot then:

  • asks for the visitor's name and phone number or email,
  • tells you about the question,
  • lets you step into the chat if you are free.

A visitor who gets an honest "let me check" and a quick reply later is far better off than one who gets a confident wrong answer.

6. Review unanswered questions

Every question the bot couldn't answer is a gap in your information. Look at them regularly and answer them once. If the tool remembers your answer, the next visitor who asks gets it straight away. Over a few weeks the gaps shrink.

7. Test it like a customer

Before you put any chatbot live, ask it the questions your customers ask. Try the awkward ones: "Can you do it cheaper?", "Do you work Sundays?", "What if I cancel?" Fix anything that is wrong or vague.

What this looks like in practice

Take a Croydon plumber. His approved facts include "Call-out fee £65, includes first 30 minutes" and "Areas: Croydon, Purley, Sutton". A visitor asks about a job in Brighton. A careful chatbot won't say yes because Brighton is "nearby enough". It will say Brighton isn't in the listed areas, offer to take their details and pass the question to him. He replies, and adds a written answer about longer trips for next time.

How KipDesk handles this

KipDesk was built around these rules. It answers only from facts you approve, answers you write, files and text you add, and your website pages. It copies prices, times and areas exactly as you gave them and never estimates or invents them. Each answer shows where it came from. When it can't find an answer, it says so, asks for the visitor's details, asks you to join the chat and adds the question to your to-do list. When you answer, it uses that answer for every future visitor who asks. You can test it on the Test page and tap "Fix this answer" under any reply.

It isn't perfect, and no AI tool is. But because every answer comes from a fact you can see and change, mistakes are easy to spot and fix. More at /for/salons and /for/trades, and pricing at /#pricing.

Accurate by design, checked by you

AI chatbots make things up when they are left to fill gaps on their own. Limit them to your approved facts, have them copy prices exactly, show sources, avoid totals and hand over when unsure. Then check the unanswered questions each week. Do that, and a chatbot becomes a reliable first reply rather than a risk. For a wider checklist, see choosing an AI chatbot for your small business website, and if you run a software product, customer support for small SaaS teams.