AI Output Disclaimer

Responses from a JestBot widget are generated by a large language model, grounded using retrieval from the bot owner’s uploaded documents, URLs, and pasted text. They are composed automatically, in real time, in response to each visitor message — they are not written or reviewed by a human before being shown to a visitor, unless a human agent has taken over that specific conversation via handover.

When a visitor asks a question, JestBot searches the bot’s knowledge base using embedding similarity (cosine similarity across roughly 1500-character chunks of the uploaded content) and passes the top-matching chunks — up to five, above a minimum relevance score — to the language model as context for its answer. This is what distinguishes JestBot from a generic chatbot: answers are meant to be based on the bot owner’s actual content rather than the model’s general training data.

Retrieval grounding meaningfully reduces, but does not eliminate, the chance of an inaccurate or incomplete answer. A bot can still:

• Retrieve a less relevant chunk than the one that actually answers the question, especially for ambiguously worded queries
• Answer confidently from a chunk that is itself outdated, if the bot owner hasn’t refreshed their source documents
• Miss context that exists in the knowledge base but wasn’t among the top-matching chunks for that specific phrasing
• Blend information from multiple retrieved chunks in a way that changes their original meaning
• Answer a question the knowledge base doesn’t actually cover, if similarity scoring still surfaces a loosely related chunk

JestBot makes no guarantee, express or implied, that any bot response is accurate, complete, current, or fit for any particular purpose.

Even with relevant source material provided, the underlying language model can still generate text that sounds plausible but isn’t actually supported by the retrieved content — a well-documented behavior of large language models generally, sometimes called “hallucination.” Retrieval grounding lowers the rate at which this happens compared to an ungrounded chatbot, but does not reduce it to zero, and no vendor (JestBot included) can currently offer a chatbot with a guaranteed zero hallucination rate.

Bot owners can choose between the default self-hosted model or their own OpenAI, Anthropic, Gemini, Groq, or Mistral key. Different providers and models have different accuracy characteristics, tones, and failure modes. JestBot does not control or guarantee the output quality of any third-party model provider, and switching providers on the same bot with the same knowledge base can produce meaningfully different answers to the same question.

The account holder who creates and configures a bot is responsible for: the accuracy and currency of the content they feed into it, reviewing how the bot represents their product, service, or business, testing bot behavior (via Test Chat) before embedding it publicly, and how the bot’s answers are presented to their own end users or customers. JestBot provides the infrastructure that powers the bot; it does not write, curate, or endorse the content of any individual bot’s knowledge base or the answers it produces.

If you’re chatting with a JestBot-powered widget as a visitor: treat responses as a helpful starting point rather than a guaranteed-accurate final answer, particularly for anything involving money, deadlines, legal terms, or safety. Where a response seems wrong, incomplete, or contradicts something you know to be true from the business directly, use the handover option (where available) to reach a human, or contact the business through another channel to confirm.

Unless a bot owner has explicitly built and staffed a bot for that purpose with qualified professionals reviewing its content and behavior, JestBot-powered bots are not a substitute for professional legal, medical, financial, tax, or safety advice. Visitors should not rely on bot output for consequential decisions in these areas without independent verification from a qualified professional.

Where a bot owner has enabled human handover, a visitor can request to speak with a real agent after a configurable number of messages. Once a conversation’s handover status is marked assigned, messages sent from that point by the agent are human-authored, not AI-generated, and this disclaimer no longer applies to them — though the AI-generated messages earlier in the same conversation remain subject to it.

If you’re a bot owner and a bot gave a wrong or misleading answer, use Test Chat to see exactly which chunks were retrieved and their similarity scores — this is usually the fastest way to diagnose whether the issue is a knowledge-base gap, an outdated document, or a retrieval mismatch. For platform-level issues you believe are a bug rather than a knowledge-base problem, reach out via the contact page.