Base guardrails sit above every campaign configuration. They enforce behaviour that must be consistent regardless of how a campaign is set up — character limits, disclosure rules, tone standards, and off-topic handling.
The system prompt is built in two layers: base guardrails first, campaign second. The guardrail layer defines role, output rules, and compliance scope. The campaign layer adds persona, questions, and product context. The AI receives both as a single system prompt — guardrails always come first.
Each guardrail addresses a specific failure mode from real AI deployments. They're not aspirational guidelines — they're tested constraints with defined behaviour.
The AI uses campaign fields in a defined priority order: product_or_service, benefits, background, objection_handling, cross_up_sell. If no usable information is present, it deflects to a specialist rather than fabricating an answer.
How it works: Tested via golden conversation suite — AI must deflect to specialist when objection_handling field is empty.
When a contact directly asks whether they are talking to a human or a bot, the AI identifies itself as an AI assistant and offers to pass the conversation to a specialist. It never claims to be human under any phrasing of the question.
How it works: Enforced via base guardrails — applies to every campaign without configuration. Verifiable via golden conversation test.
When a contact sends a message outside campaign scope — a creative writing request, personal conversation, or unrelated question — the AI declines briefly, includes the next campaign question, and returns PAUSE. The conversation continues.
How it works: Off-topic golden test confirms directive must be PAUSE — not END — for a first off-topic message. END is reserved for cumulative disengagement.
The AI does not return END in response to ambiguous signals, momentary hesitation, or off-topic replies. A single negative answer to a campaign question does not end the conversation. Premature termination is a campaign quality failure.
How it works: Golden conversation suite validates termination only occurs at the correct moments — after two or more cumulative rejection signals.
The Campaign Tester lets you throw edge cases at the AI before any contact is messaged.