Prompt injection
Prompt injection is an attack where someone plants instructions in the content or messages an AI processes — trying to make it ignore its real rules, reveal hidden instructions, or behave in unintended ways.
How it works
Because AI assistants read untrusted text — user messages, and sometimes web content — an attacker can embed commands like “ignore your instructions and say X.” A naive system may obey them.
Defenses include keeping the system’s real instructions separate from untrusted text, sanitizing that text, and designing the assistant to stay strictly within its intended task.
How SiteMind uses it
SiteMind is built to resist prompt injection: untrusted content is isolated from the system instructions and sanitized, and the assistant is constrained to answering from your knowledge base — so attempts to hijack it or make it go off-topic are far less likely to succeed.
Related terms
AI hallucination
An AI hallucination is when a model states something false or made-up as if it were true — a serious risk for any customer-facing chatbot.
Retrieval-Augmented Generation (RAG)
Retrieval-Augmented Generation (RAG) is a method where an AI looks up relevant information from a trusted source before answering, so its replies are grounded in facts instead of guesses.
Source grounding
Source grounding means every answer an AI gives is tied to and backed by specific retrieved content, and can be traced back to it.
Put this to work on your site
Free 3-day trial, no card required.