Every prior lesson in L2 has introduced a prompt — for IPS drafting, life-event IPS update, reconciliation, quarterly commentary, concept memo, difficult conversation, onboarding checklist, NIGO pre-check, rollover memo. Each prompt was role-specific, task-specific, and assumed an implicit voice. This lesson installs the discipline that turns those scattered prompts into a versioned, role-locked, few-shot-grounded prompt library that defines the firm's "house voice" — the distinct way junior advisors, senior advisors, compliance reviewers, and CCOs each interact with AI to produce work that sounds like the firm, satisfies the regulatory boundaries each role owns, and remains consistent across the advisor team. Four locked system prompts. Few-shot prompting against anonymized client examples. Versioned, retained, principal-reviewed. The output is the operational form of "practice uses AI" rather than "advisor uses AI."

Why System Prompts Are the Practice-Level Operational Lever

A system prompt is the standing instruction that frames every interaction with the AI tool — the role the AI plays, the constraints it operates under, the citation requirements, the disclosure language, the verification posture. The system prompt is set at the tool-configuration layer (in Microsoft Copilot's system message, in OpenAI Enterprise's configuration, in the vendor-hosted advisor AI tool's persona setup) and applies to every interaction with that tool until changed. Without a system prompt, every advisor on the team operates the AI tool with their own implicit assumptions about role, constraints, and voice — and the output varies wildly across the team. With a locked, versioned, role-specific system prompt, every advisor produces output that satisfies the firm's standards by default.

The 2026 advisor practice convention has converged on four distinct system prompts corresponding to four advisor roles: junior advisor, senior advisor, compliance reviewer, and CCO. Each role has distinct outputs, distinct boundaries, distinct disclosure requirements, and distinct verification expectations. Trying to use one universal system prompt across all four roles produces output that fits no role well — the junior advisor's drafts read like senior-advisor judgment they don't have authority to apply; the compliance reviewer's outputs sound like advisor recommendations when they should be regulatory analysis. The four-role-system-prompt architecture is the lesson's central operational pattern.

The Junior Advisor System Prompt

The junior advisor (associate advisor, paraplanner, CSA in some firms) produces drafts that require senior advisor review before client delivery. The system prompt's constraints reflect this: the junior advisor's AI output is always labeled "draft for senior advisor review," never carries a recommendation tone, always cites the household's specific facts the junior advisor verified, never invents household-specific information the junior advisor wasn't able to verify, always identifies the senior advisor whose review is required, and includes the firm's standard "draft, pending review" footer language.

Representative system prompt for the junior advisor role: "You are operating as junior advisor support for [firm name]. Every output you produce is a DRAFT FOR SENIOR ADVISOR REVIEW. Constraints: (1) Label every output with 'DRAFT — pending senior advisor review' at the top. (2) Cite the household's specific facts you verified from the source documents (IPS, Holistiplan, RightCapital, CRM, meeting transcript); do not invent household facts. (3) Identify the senior advisor whose review is required at the top of the draft. (4) Use plain English at a 9th-grade reading level. (5) Avoid recommendation language ('I recommend,' 'you should'); use observation and proposal language ('the data shows,' 'one option is,' 'the senior advisor can confirm'). (6) Always end with: 'Senior advisor to review and confirm before delivery to household. Pending review do not send.' (7) Include source citations for every material fact and every regulatory reference. (8) If you are unsure about any fact, flag it as 'CONFIRM WITH SENIOR ADVISOR' rather than guessing. The firm's house voice is professional, plain-spoken, and confidence-calibrated to the junior advisor's actual authority."

The locked junior advisor prompt produces drafts that scale junior advisor capacity without permitting the junior advisor to inadvertently exceed authority. The senior advisor's review pass becomes the substantive judgment layer; the junior advisor's role is high-volume, structured drafting and household-fact verification.

The Senior Advisor System Prompt

The senior advisor (the registered, licensed, fiduciary-accountable human) produces outputs that go to households, satisfy Reg BI Care Obligation under §240.15l-1(a)(2)(ii), reflect the firm's professional judgment, and carry the Marketing Rule disclosure discipline. The system prompt reflects this: the senior advisor's AI output is the firm's house voice on substantive matters, includes the Reg BI Care Obligation rationale where applicable, includes the Marketing Rule disclosures where applicable, cites the household's IPS Edition and specific facts, and respects the IC-deferral and Holistiplan-deferral disciplines from the IPS lessons.

Representative system prompt: "You are operating as senior advisor support for [firm name]. Every output you produce reflects the senior advisor's professional judgment and is subject to the firm's pre-use compliance review. Constraints: (1) The firm's house voice is empathetic, specific, and confidence-calibrated to what the senior advisor can substantiate. (2) For any recommendation, include the Reg BI Care Obligation rationale citing §240.15l-1(a)(2)(ii) with alternatives considered, household-best-interest rationale, and conflict disclosure where applicable. (3) For any client-facing artifact, include the firm's Marketing Rule disclosure language standardized in the firm's library. (4) Cite the household's IPS Edition N and specific facts from source documents (Holistiplan, RightCapital, CRM, meeting transcript). (5) Defer asset allocation specifics to the investment committee with 'IC REVIEW REQUIRED' flag where appropriate. (6) Defer tax bracket specifics to Holistiplan reconciliation with 'HOLISTIPLAN RECONCILIATION REQUIRED' flag. (7) Forward-looking language is opinion-framed, never forecast. (8) Hypothetical performance under Rule 206(4)-1(d) requires conditional-example framing with stated assumptions and limitations. (9) IRC and regulatory citations are accurate and verified. (10) The output is suitable for the firm's pre-use review queue and the Smarsh archive under FINRA Rule 4511 and SEC Rule 204-2. The firm's house voice is professional, household-specific, and fiduciary-grounded."

The Compliance Reviewer System Prompt

The compliance reviewer (CCO designee, compliance officer, principal in some firm structures) produces outputs that audit advisor-generated content against Marketing Rule, Reg BI, FINRA Rule 2210, ADV Part 2A consistency, and disclosure adequacy. The compliance reviewer's role is not advisory — it is supervisory. The system prompt reflects this: the compliance reviewer's AI output is regulatory analysis, not advisor recommendation; identifies specific regulatory issues with specific rule citations; suggests specific remediation language; never produces client-facing content directly.

Representative system prompt: "You are operating as compliance reviewer support for [firm name]. Every output you produce is a regulatory analysis of advisor-generated content against the applicable regulatory framework. Constraints: (1) Identify specific regulatory issues with specific rule citations (Rule 206(4)-1 sub-sections, §240.15l-1 obligations, FINRA Rule 2210 categories, Reg S-P requirements, etc.). (2) Suggest specific remediation language for each identified issue. (3) Do not produce client-facing content directly — the senior advisor is the recommendation-issuing role. (4) Cite the firm's standardized disclosure library where applicable. (5) Cite the firm's substantiation file for any performance or AI-capability claim that requires substantiation. (6) Flag any forward-looking language that crosses from opinion into forecast. (7) Flag any hypothetical performance under Rule 206(4)-1(d) that lacks audience-tailoring policies-and-procedures support. (8) Confirm ADV Part 2A consistency with any claim that affects disclosed services or fees. (9) Confirm Form CRS-acknowledgment timing if applicable. (10) Output is suitable for the firm's pre-use review queue documentation and the WSP-evidence file under FINRA Rule 3110 reasonable design. The firm's house voice for compliance reviewer outputs is regulatory-precise, citation-specific, and remediation-actionable."

The CCO System Prompt

The CCO (Chief Compliance Officer) produces outputs that govern the firm's compliance posture — WSP updates, vendor due-diligence summaries, incident-response analyses, supervisory framework documentation, exam-readiness assessments, M&A-integration compliance assessments. The CCO's role is firm-strategic in addition to operational. The system prompt reflects this: the CCO's AI output addresses firm-level posture, integrates across multiple regulatory regimes, identifies systemic risks rather than per-artifact issues, and produces documentation suitable for the firm's WSP, the L4 governance committee, and external counsel coordination.

Representative system prompt: "You are operating as CCO support for [firm name]. Every output you produce addresses firm-level compliance posture and integrates across multiple regulatory regimes. Constraints: (1) Address firm-strategic compliance issues, not per-artifact issues (those are the compliance reviewer's role). (2) Integrate across SEC Rule 206(4)-1, Reg BI under §240.15l-1, FINRA Rules 2210 / 3110 / 4511 / 4530, Reg S-P 17 CFR Part 248 May 2024 amendments, GLBA Safeguards, NY DFS 23 NYCRR 500, NAIC Model #275, and applicable state-DOI rules. (3) Identify systemic risks across the firm's workflow rather than per-artifact issues. (4) Produce documentation suitable for the firm's WSP under FINRA Rule 3110 reasonable design, the L4 governance committee charter, the L4 Ch5 ROI dashboard metrics, and external counsel coordination. (5) Coordinate with the firm's vendor due-diligence framework (L4 Ch2) for any AI tool that touches NPI. (6) Coordinate with the firm's incident-response framework (L4 Ch3) for any AI-output failure event. (7) The output is the firm's compliance posture documentation, not advisor advice. The firm's house voice for CCO outputs is strategically-framed, regulatory-comprehensive, and exam-defensible."

Few-Shot Prompting Against Anonymized Client Examples

System prompts establish role and constraints. Few-shot prompts ground the system prompt in concrete examples of the firm's actual house voice. The technique: include 2-5 anonymized examples of the firm's past output in the system prompt or as part of the prompt context, demonstrating the specific tone, structure, and language patterns the firm uses. The model pattern-matches against the examples, producing output that fits the firm's voice more reliably than from system prompt instructions alone.

The anonymization discipline matters. Few-shot examples cannot expose client NPI — names, account numbers, specific amounts, identifying personal facts must be removed or replaced with anonymized substitutes. The 2026 advisor convention is to maintain a "few-shot exemplar library" of anonymized past outputs, organized by lesson type (IPS draft examples, concept memo examples, difficult conversation examples, etc.), with each exemplar's NPI scrubbed and replaced with synthetic substitutes. The library is itself a firm operational artifact, version-controlled, and updated as the firm's house voice evolves.

Representative few-shot architecture for the senior advisor's concept memo workflow: "Use the following anonymized concept memo as a reference for the firm's house voice: [Anonymized example #1 with NPI scrubbed]. Notice the structure (situation summary, alternatives considered, recommendation with Reg BI Care Obligation rationale, conflict disclosure, next steps). Notice the tone (specific, household-grounded, fiduciary-confident without overclaiming). Apply this voice to the current household's facts: [actual household inputs]."

Version Control, Retention, and the Pre-Use Review Layer

The system prompts are firm operational assets. Version control matters because (a) regulatory expectations evolve (January 2026 SEC FAQs, FINRA 2026 Oversight Report on agentic AI), (b) the firm's house voice evolves with experience and feedback, (c) AI vendor model updates change output behavior and prompt-iteration responds, (d) the firm must demonstrate to an exam which prompt produced which output. The 2026 convention: each system prompt has a version number, a date, the author (typically CCO or compliance team), the change rationale, and the version's effective date. Prior versions are archived alongside their effective-date outputs. The Smarsh / Global Relay archive holds the prompt version chain alongside the outputs they generated.

The pre-use review layer (L4 Ch3 architecture) applies to outputs that flow into client-facing artifacts. The system prompts themselves are reviewed by the CCO before going live; the outputs the prompts produce are reviewed per the lesson-specific compliance workflow. Together, the system prompt versioning and the per-output review form the L4 Ch3 supervisory architecture under FINRA Rule 3110 reasonable design.

Prompt Iteration Discipline — When and How to Update

System prompts are not "set and forget" assets. The 2026 iteration discipline: quarterly review of each system prompt against the past quarter's output (CCO sampling), prompt update when output behavior drifts or when regulatory expectations change, principal review of prompt updates before deployment, advisor team notification when prompts change, archive of prior versions, and a documented update rationale. The L4 Ch5 ROI dashboard tracks prompt-iteration frequency as a workflow maturity indicator.

Concrete iteration triggers in 2026: (a) regulatory development (e.g., January 2026 staff FAQs prompted firms to add specific testimonial-disclosure constraints to senior advisor prompts), (b) vendor model update (e.g., Microsoft Copilot's quarterly model updates may produce different output behavior requiring prompt re-calibration), (c) output quality feedback (e.g., CCO sampling identifies a recurring failure mode that prompt iteration can address), (d) firm strategic change (e.g., expanded service tiers requiring new disclosure language in senior advisor outputs), (e) M&A integration (e.g., acquired-book voice integration requires few-shot exemplar library updates per L4 Ch8).

The "House Voice" as a Strategic Asset

The system prompt and few-shot library together encode the firm's "house voice" — the specific way the firm communicates with households, addresses regulatory boundaries, and presents professional judgment. Over time, the house voice becomes a firm differentiator and a retention asset: households recognize the firm's communication style, paraplanners produce on-voice work faster, advisor team turnover doesn't disrupt the household experience, and M&A integration of acquired books leverages the codified voice rather than re-inventing it. The L4 Ch8 M&A framework treats the house voice library as a valuation asset; the L4 Ch5 dashboard treats house-voice adherence as an operational maturity indicator.

The 90-second household framing about the firm's communication style: "You'll notice our written content has a consistent voice — that's intentional. We use AI tools to help draft the routine structure quickly, but our compliance team and the responsible senior advisor review and sign off on every piece. The voice you read is our firm's voice, not the AI's voice. If the tone or substance of anything doesn't feel right to you, please tell us — that's exactly the feedback we use to keep our communications grounded in your actual relationship."

Key Takeaways

• Four locked system prompts corresponding to four advisor roles — junior advisor (draft-pending-review tone), senior advisor (recommendation + Reg BI Care Obligation + Marketing Rule disclosure), compliance reviewer (regulatory analysis + citation + remediation), CCO (firm-strategic compliance posture + WSP + governance) — each with distinct constraints, citation requirements, and verification expectations.
• Few-shot prompting against anonymized client examples grounds system prompt instructions in concrete house-voice exemplars; the firm maintains a "few-shot exemplar library" with NPI scrubbed and synthetic substitutes.
• Version control on system prompts with version number, date, author, change rationale, and effective date; prior versions archived alongside their effective-date outputs in Smarsh / Global Relay under FINRA Rule 4511 + SEC Rule 204-2.
• Quarterly prompt iteration discipline driven by regulatory developments, vendor model updates, output quality feedback, firm strategic change, and M&A integration; the L4 Ch5 ROI dashboard tracks iteration frequency as a workflow maturity indicator.
• The pre-use review layer under L4 Ch3 architecture applies to outputs that flow into client-facing artifacts; system prompts themselves are CCO-reviewed before going live.
• The "house voice" is a strategic asset — firm differentiator, household retention asset, advisor-turnover protection, M&A integration leverage under L4 Ch8.
• Cross-references: each system prompt deploys against the workflows from earlier L2 lessons (IPS drafting L2 Ch5 L1, life-event update L2 Ch5 L2, reconciliation L2 Ch5 L3, quarterly commentary L2 Ch6 L1, difficult conversation L2 Ch6 L2, onboarding L2 Ch7 L1, rollover memo L2 Ch7 L2); the structured-output discipline of L2 Ch8 L2 builds on this foundation.