What Should an AI Assistant Do When It Cannot Answer a Property Inquiry?

An AI assistant that cannot answer a property question has three acceptable moves: answer from verified listing data, ask one clarification question the prospect can actually resolve, or escalate to a human with useful context. Inventing a plausible answer is never one of them. The expensive failure in real estate is a confident wrong answer about price, availability or ownership that an agent then has to walk back with a live prospect; a refusal costs far less.

TL;DR: Approve three permitted outcomes for your assistant: answer only from verified listing data, ask one necessary clarification question, or escalate with the inquiry message, channel, listing reference and the specific gap it could not verify. It must never infer price, availability, ownership, exact location or legal status. This rule set is yours to approve and have configured at onboarding.

The facts an assistant must never guess

If the listing data does not contain the answer, the assistant must say so and either clarify or escalate. Put five facts on the never-infer list:

• Price or rent
• Availability and sale status
• Ownership and title position
• Exact location and boundaries
• Legal or compliance status

Two modelled scenarios show why. These are illustrative, not drawn from any client system. A prospect asks the price of a house whose listing has no price recorded; the assistant does not quote a figure from similar properties, it says the price is being confirmed and escalates. Two conflicting rent figures exist for the same flat; the assistant treats the conflict itself as the unanswerable fact and escalates with both figures flagged, instead of quietly picking one.

One model provider says the same thing in its own guidance. Anthropic's guidance on reducing hallucinations recommends explicitly permitting the model to say it does not know and grounding answers only in the documents it is given. An assistant that is never allowed to say "I don't know" will fill the gap with plausible fiction.

When one clarification question is enough

Ask a clarification question only when the prospect holds the missing information: which of two similar listings they mean, whether they want to rent or buy, or which currency their budget is in. One question, then answer or escalate. The approved qualification questions already cover budget, timeline and location; the assistant draws from that set rather than improvising new ones.

Escalate immediately, with no clarifying detour, when the problem sits in your data or your policy: conflicting or stale listing information, a complaint, urgency, anything touching ownership or legal status, or a request outside the assistant's approved scope. A clarification question cannot fix bad data.

What an escalation must carry

An escalation is useful when the receiving agent can act without reconstructing the conversation. Require four fields:

1. The prospect's message, unedited
2. The channel it arrived on: website form, WhatsApp or a connected listing platform
3. The listing reference in question
4. What the assistant could not verify, and why

Where the escalation goes next is assignment logic, and the full inquiry-to-agent workflow covers the surrounding steps. The fallback rules only need to hand over a complete parcel.

One legal note: inquiry messages contain personal data, and processing it in Zimbabwe falls under the Cyber and Data Protection Act (Act 5 of 2021), enforced by POTRAZ. The law firm Michalsons dates its commencement to March 2022. Escalations should carry the minimum personal data the agent needs, not the assistant's full transcript by reflex.

An unavailable fact is not an unauthorised request

Two different handovers look identical to a prospect unless the assistant labels them. "That detail is not in the listing data" covers missing prices and unconfirmed availability: the fact exists, the assistant cannot verify it. "That needs a person by policy" covers negotiation, committing to a viewing time, requests involving someone else's personal data, and complaints: the assistant could produce an answer here, and must not.

Have the assistant name which is happening in plain language. Modelled wording: "I don't have a confirmed price for this property, so I'm handing you to the agent who does." That treats the prospect like an adult. The cross-industry principle behind a clean transfer is covered in how to hand off leads from an AI assistant to a human; the rules above are its property-specific implementation.

What prospects can be told about timing

TechTribe's AI lead nurturing replies within seconds, day and night, so the assistant can promise its own speed honestly. An agent's response time is a different matter: unless your agency has committed to one, the assistant must not name a clock. It promises a next step with an owner instead. "The agent who handles this listing will contact you about the confirmed rent" is safe wording; "within two hours" is a promise someone else has to keep.

The approval checklist

TechTribe configures the assistant with your agency's details and preferred wording at onboarding, so these are decisions for you to approve, not defaults to inherit. Before the assistant goes live:

• Confirm the never-infer list: price, availability, ownership, location, legal status.
• Approve which ambiguities earn one clarification question.
• Approve the immediate-escalation triggers: data conflicts, complaints, urgency, legal or ownership questions, out-of-scope requests.
• Confirm the four escalation fields: message, channel, listing reference, unverified gap.
• Approve the exact wording for "not in the data" and "needs a person".
• Confirm timing language names a next step and an owner, never a clock.
• Decide who will check that the rules are followed once live.

The Real Estate Platform is where these rules get configured for your agency; the pricing page sets out how it differs from a standard website.