Judd

Content production at MGMA had a structural bottleneck: a single copywriter controlled voice and tone, which created two compounding problems. First, output speed was tied to one person's availability, slowing backlog delivery. Second, because the standards lived in that person's head rather than in a documented system, consistency was hard to enforce and even harder to scale. When I audited the situation, I did not frame it as a people problem. I framed it as a system problem. The organization lacked a defined copy decision hierarchy, an output schema, and any mechanism for enforcing brand voice at scale. The copywriter was not the bottleneck. The absence of a system was.

Judd is a content QA and brand voice agent built in ChatGPT as a Custom GPT. The name and personality were intentional: I gave him a character because I knew adoption depended on teams trusting the tool, not just tolerating it. A named agent with a consistent voice and a feedback loop is something people engage with. A generic AI tool is something people route around. Judd operates on four layers: Knowledge layer: Healthcare copywriting frameworks, MGMA-specific briefs, SEO and keyword context, and real approved content as ground truth. Not just examples to mimic: source material to extract rules from. Reasoning layer: A copy decision hierarchy that tells Judd what wins when priorities conflict: audience clarity first, then MGMA authority, then conversion intent, then SEO. This prevents clever-but-useless output. Output layer: A strict schema every response follows: page intent summary, audience snapshot, core messaging, draft copy by section (hero, proof, CTA), SEO enhancements, and notes. Judd does not generate copy until all required inputs are present. Transparency layer: Every interaction closes with Judd reflecting back its understanding: what the user asked for, and how it approached the generation or review task. This is not a summary. It is a trust and alignment mechanism. The user sees Judd's reasoning, not just its output, which means misalignment gets caught before anyone acts on a wrong answer. An additional knowledge folder contains the names, departments, and areas of ownership for key internal contacts, so Judd can escalate appropriately when content requires legal, compliance, or subject matter review. This turns Judd from a content tool into a content and routing tool. The six governing rules built into Judd: Opening rule: Lead with a clear outcome tied to practice growth, career advancement, or operational improvement. Never lead with product features. Authority rule: Ground claims in MGMA expertise, data, or real scenarios within the first section. Quantifiable proof required. Tone rule: Calm, direct, executive-friendly. No hype. Prioritize clarity over cleverness. Benefit rule: Every bullet maps to revenue, efficiency, risk reduction, or career growth. No vague benefits. CTA rule: Advisory language only. Explore, Access, Learn, See how. No aggressive or sales-heavy phrasing. Flexibility rule: Adapt structure and tone based on page intent, audience, and conversion goal. Do not rigidly replicate examples.

New users enter through a wizard sequence before they reach the menu. The wizard introduces Judd by name and purpose, sets clear expectations about capability, states what not to share (member data, personally identifiable information, credit card numbers, login credentials, confidential financial data, content under legal review), and ends with a soft invitation rather than a demand. This design decision addresses the second problem underneath the first: even a well-built tool dropped on a team without onboarding becomes a source of anxiety rather than adoption.

After onboarding, every session opens with a structured menu. This removes the blank-prompt problem that stops non-technical users before they start. Create new content: Draft copy aligned to brand voice, audience, and conversion intent. Review existing content: Audit a draft against brand voice, tone, structure, and messaging hierarchy. SEO and AI search review: Evaluate content for search visibility, keyword alignment, and AI discoverability. Voice analysis: Analyze a piece of content and extract its tone, structure, and voice patterns. Ask a question: Not sure where to start? Tell me what you're working on and I'll point you in the right direction. Option 5 is the most important menu item for adoption. It gives uncertain or new users a soft landing that does not require them to already know which workflow they need. Reducing the activation cost of a tool is as important as building its capabilities.

Judd's architecture uses five documented prompt engineering techniques: Outline Expansion: Content briefs are expanded into full structural outlines before copy is generated, preventing AI output that is accurate but structurally wrong for the intended page type. Menu Actions: Structured menu options reduce the activation cost for non-technical users and ensure every interaction starts with a clearly defined task rather than an open-ended prompt that produces inconsistent results. Fact Check: Judd validates claims against the knowledge base before including them in output. Claims that cannot be grounded are flagged for human verification rather than generated speculatively. Tail Generation: Every interaction closes with a transparency reflection: Judd states what it understood the user to be asking, and how it approached the task. This catches misalignment before anyone acts on a wrong answer and builds user trust over time. Chain of Thought with Input Requests: For complex or ambiguous briefs, Judd works through its reasoning explicitly and asks for clarifying input before generating. This prevents the common failure mode of confident-sounding output built on a misunderstood brief.

Removed the single-contributor dependency from content production, decoupling backlog velocity from one person's availability. Standardized MGMA brand voice across content types and contributors, reducing revision cycles caused by inconsistent tone. Converted approved content into an operational system: rules, hierarchy, and schema, rather than a reference archive. Established a feedback loop that allowed Judd to evolve as priorities shifted, rather than calcifying at v1. Drove adoption through personality and trust, not mandate. Teams engaged with Judd because the experience was designed to earn buy-in. Built in escalation paths via the internal contact directory, turning Judd from a content tool into a content and routing tool. Designed onboarding and menu-driven UX to reduce activation anxiety and drive voluntary adoption rather than mandate compliance.