SBA · SBA-29 · record R3185

Office of Advocacy AI System-Prompt Suite

Small Business Administration · Advocacy: Office of Advocacy

Overview

Development stage: PilotSource: b) Pilot – The use case has been deployed in a limited test or pilot capacity.
High-impact designation: Not high-impactSource: c) Not high-impact
Impact justification: Not high impact: This AI system does not meet the criteria of any of the six pillars that make up High Impact AI in Memorandum M-25-21.
Start date: 2025-12-10 (day precision)Source: 12/10/2025 00:00:00
Withheld from public reporting: NoSource: a) No

Mission

Topic area: Information Technology
Operational functions: Document processing and extraction, Summarization Derived by keyword rules; see methodology

Problem the AI is intended to solve

Federal staff spend significant time on repetitive text-processing tasks such as rewriting, summarizing, drafting replies, checking grammar, and fact-checking. Traditional approaches require manual effort and lack consistency across analytic workflows. Unstructured AI outputs create ambiguity and are difficult to integrate into automated processes.

Expected benefits

Significantly increases frequency and quality of analytic work by reducing friction in text processing, improving consistency through structured JSON outputs, enabling automation at scale, making outputs audit-ready and transparent, and supporting reuse across government agencies. Published prompts advance Executive Order 14179 and OMB M-25-21 goals for responsible AI adoption and cross-government innovation.

System outputs

Suite of structured Large Language Model (LLM) system prompts for federal text analysis tasks. Provides automated AI capabilities including calendar event generation, text rewriting, reply drafting, text summarization, grammar/spelling checking, and fact-checking. Each prompt demands structured JSON output, transforming LLMs from conversation generators into predictable, machine-usable workflow components. Inputs: unstructured text via API (sentences, documents, emails). Outputs: structured JSON-formatted results (rewrites, summaries, replies, error reports, fact-checks).

Technology

AI classification: Generative AISource: Generative AI: AI that generates new or synthetic content (e.g., images, videos, audio, text, code).
System name(s): Office of Advocacy LLM System
Custom-developed code: Yes
Public source code: data.sba.gov ↗

Sourcing

How it was built: Developed in-houseSource: b) Developed in-house
Vendor (as reported): In-house with commercial LLM APIs
Vendors (standardized): None on the standard list

A vendor is the supplier named by the agency. It does not identify the underlying model or AI technology.

Data and privacy

Involves PII: No
Privacy Impact Assessment: Not publicly available
Authorization to Operate: Yes
Demographic features: k) None of the above
Training and evaluation data: Commercial LLM provider training data (OpenAI or similar)
Federal Data Catalog: advocacy.sba.gov ↗

Governance

8 of 8 minimum-practice questions answered. A blank answer means the agency reported nothing; it does not mean the practice is absent.

Pre-deployment testingReported in placeSource: a) Yes

AI impact assessmentReported in placeSource: a) Yes

Independent reviewReported in placeSource: a) Yes – by another appropriate agency office or reviewer not directly involved in the AI’s development

Ongoing monitoringReported in placeSource: a) Yes, sufficient monitoring protocols have been established

Operator trainingReported in placeSource: a) Yes, sufficient and periodic training has been established

Fail-safeReported in placeSource: a) Yes

Appeal processNot applicableSource: b) Not applicable

User and public consultationReported in placeSource: Conducted direct usability testing with Office of Advocacy staff, published system prompts and Python scripts publicly for transparency and feedback, established internal protocols for monitoring output quality, and incorporated user feedback into prompt refinement. Supports cross-government reuse and innovation through open sharing of AI resources.Flag: free-text answer

Potential impacts and how they were identified

Potential impacts include quality and consistency of text analysis outputs used in regulatory analysis and small business advocacy. Impacts identified through user testing, validation of structured JSON outputs, and transparency requirements under EO 14179 and OMB M-25-21. Mitigation includes human oversight of all outputs, audit trails via structured JSON, published prompts for transparency, and ongoing accuracy monitoring.