System Prompt
The initial set of instructions provided to an AI model that establishes its role, behavior, constraints, and context before a user interaction begins. The system prompt is typically invisible to the end user but shapes how the model responds throughout the conversation. In commercial real estate, a well-designed system prompt might instruct the AI to act as a senior acquisitions analyst, follow specific underwriting conventions, cite data sources, and flag assumptions, ensuring that every response is calibrated to the professional context.
Putting System Prompt in Context
A CRE investment firm deploying an AI assistant for deal screening configures a system prompt that instructs the model to evaluate opportunities against the firm’s target return thresholds, apply its standard underwriting conventions for cap rate selection and vacancy assumptions, and always disclose when a figure is estimated rather than sourced from verified data, so that every analyst using the tool receives consistent, firm-specific output regardless of how they phrase their questions.
Frequently Asked Questions about System Prompt
What should a CRE team include in a system prompt to get useful AI output?
An effective system prompt for CRE use typically defines the model’s role, the specific asset class or transaction type being addressed, the firm’s preferred conventions for key metrics like cap rates or DSCR thresholds, and any formatting or citation requirements. Including context about the audience, such as whether outputs will be reviewed by junior analysts or presented to investment committee, also helps calibrate the depth and tone of responses. The more specific the instructions, the less the model has to guess about what a useful answer looks like.
How is a system prompt different from the instructions a user types into the chat?
A system prompt is set before the conversation begins and persists across the entire session, establishing baseline behavior that the user’s messages operate within. User messages, by contrast, direct individual responses and can ask the model to shift focus, but they generally cannot override the constraints established in the system prompt. In practice, this means a CRE firm can enforce consistent behavior, such as always flagging speculative assumptions, without relying on individual users to remember to ask for it each time.
Can a poorly written system prompt cause problems in a CRE workflow?
Yes. A vague or contradictory system prompt can produce responses that are technically accurate but misaligned with how a firm actually underwrites or communicates. For example, if the system prompt instructs the model to “be conservative” without defining what conservative means in context, the model may apply assumptions that differ from the firm’s actual standards. Overly long or conflicting instructions can also cause the model to deprioritize important constraints, particularly toward the end of longer conversations.
Is the system prompt visible to the people using the AI tool?
In most commercial deployments, the system prompt is intentionally hidden from end users, which allows firms to embed proprietary logic, preferred methodologies, or compliance guardrails without exposing them to every analyst or client interacting with the tool. Some platforms allow administrators to reveal the system prompt to certain users, but the default in enterprise tools is to keep it confidential. This separation is part of what makes the system prompt a useful configuration layer rather than just another chat instruction.
How does the system prompt relate to other AI configuration concepts like fine-tuning or RAG?
The system prompt operates at the session level, shaping behavior through natural language instructions each time the model runs. Fine-tuning, by contrast, modifies the model’s underlying weights through additional training on firm-specific data, which is a more permanent and resource-intensive process. Retrieval-augmented generation, or RAG, connects the model to an external knowledge base so it can pull in relevant documents at query time. A well-architected CRE AI tool often combines all three: a system prompt to set role and conventions, RAG to surface deal-specific documents, and optionally fine-tuning to internalize firm-specific terminology and patterns.
Click here to get this CRE Glossary in an eBook (PDF) format.