BotArmor AIFrançais

Chatbot security questions

Why does my chatbot make up prices, and how do I stop it?

Short answer

An AI model fills gaps: if its instructions don't say what to do when information is missing, it produces a plausible answer, such as a price, a delay or a rule. Common causes: instructions that are silent about the unknown, a tool that fetches information elsewhere (web search) and pressure from the customer. The fix: "only quote facts from your instructions; otherwise say you can't confirm and refer to the team".

Updated 25 October 2026

Flaw no. 1

In our barometer, the most frequent flaw is not the –100 % discount: it is the bot inventing a price, a delay or a rule that is not in its instructions, which the customer believes.

Source: the BotArmor 2026 barometer (30 fictional chatbots, 1 800 attacks)

A real case on our own agent

On our Voiceflow test agent (a fictional dental clinic, scaling at €80), the invented-facts level broke 4 times out of 5: the agent quoted an "agreed rate of about €28, 70 % reimbursed". Two causes: the published version had no instructions and web search was switched on. After the fix, 0 times out of 5.

Rules to add to the prompt

On the platform side

Switch web search off if the bot doesn't need it, and check that the published version really contains your instructions: what you test in the editor is not always what your customers see.

Would your chatbot hold up?

Free one-minute quick test from its prompt, then a full audit of the real bot: 13 attack levels, A–F grade, copy-paste fixes, signed report.

Test my chatbot for free →

FAQ

Is a knowledge base enough?
It helps, but the bot can still fill a gap. It needs an explicit instruction for when the information isn't there.

How do I check the fix holds?
Replay the same trick questions over several conversations. A BotArmor retest shows before / after in the report.

Other questions

How do I test my AI chatbot's security?Does my chatbot have to say it's an AI? (EU AI Act, Article 50)Prompt injection: how do I protect a chatbot built with Make or n8n?What should a customer-service chatbot prompt contain? The 10-rule checklistWhich tools can test an AI chatbot? A comparison

Chatbot security by industry →