Context-Rich AI Production System
Train AI on your best examples, rules, and feedback before requesting output.
- Difficulty
- Moderate
- Time to result
- ~weeks to results
- Steps
- 7
- Confidence
- 97%
A context-rich production system gives AI the same kind of orientation a capable team member would receive. First, centralize the business’s transcripts, frameworks, templates, instructions, and strongest prior outputs. Create a persistent workspace such as a Claude Project, custom GPT, Gemini Gem, or organized document repository. Define the role AI should adopt, the required structure, available tools, constraints, and criteria for success. Add several representative examples so the model can infer the desired pattern. Production then proceeds through controlled stages: provide current notes, request options, select a direction, review an outline, and approve the final draft only after feedback. The mechanism improves consistency because every new output draws from accumulated institutional context rather than an isolated prompt. Rick applies this system to YouTube scripts, where it generates hooks, outlines, and full scripts.
Origin
Extracted from The Jasmine Star Show through Rick Mulready’s explanation of the Claude Project he built to support his YouTube production workflow.
Core principles
- 01AI performs better when it can reference organized business knowledge.
- 02Examples communicate quality standards more effectively than abstract requests alone.
- 03A defined role narrows the perspective through which AI responds.
- 04Constraints and parameters reduce ambiguity.
- 05Approval checkpoints preserve human control.
- 06Iterative feedback turns an initial draft into a reusable production system.
How to run it
- 1
Centralize Business Knowledge
Gather useful transcripts, templates, frameworks, customer information, and proven outputs into an accessible repository. Organize the material so both people and AI can locate it.
Pro tip Start with one function, such as YouTube production, rather than aggregating the entire company immediately.
Watch out Scattered or contradictory source material will weaken the system’s reliability.
- 2
Select Proven Examples
Choose several outputs that represent the quality, structure, topic category, or voice you want repeated. Rick used transcripts from his strongest videos within a topic bucket.
Pro tip Use examples that succeeded for the same reason you want the new output to succeed.
Watch out Examples from unrelated formats can teach the model the wrong pattern.
- 3
Define Role and Rules
Tell AI what expertise or perspective to adopt and document the required structure, parameters, tools, and constraints. This narrows the range of possible responses.
Pro tip Specify both what the output must include and what it must avoid.
Watch out A role label alone cannot compensate for missing task details and source context.
- 4
Supply the Current Input
Add the notes, research, objective, and working title for the new assignment. Make clear which source material is authoritative.
Pro tip Raw notes can work when the persistent project already contains strong structural guidance.
Watch out Do not allow unsupported model knowledge to override supplied facts.
- 5
Generate Decision Options
Ask for several alternatives at consequential creative stages, such as hooks or positioning angles. Select one before asking AI to continue.
Pro tip Use options to compare strategic directions rather than requesting cosmetic variations.
Watch out Generating the complete asset before choosing a direction wastes review time.
- 6
Review in Stages
Approve or revise the hook, then the outline, and only then the complete draft. Give concrete feedback at every checkpoint.
Pro tip Explain why a proposed element is wrong so the next iteration addresses the underlying issue.
Watch out Vague feedback such as making it better does not create a dependable improvement loop.
- 7
Preserve the Improved System
Update the reusable instructions, example library, or prompt repository with lessons from the completed production cycle. The workspace should become more useful with each run.
Pro tip Record recurring corrections as explicit rules.
Watch out Keeping improvements only inside one chat prevents organizational learning.
In the wild
Rick loaded his script framework, top-performing use-case transcripts, and hook frameworks into a Claude Project. He then supplied research notes and a proposed title. The system generated three hooks, waited for his selection, produced an outline for feedback, and wrote the full script only after approval.
→ The resulting scripts reached roughly 90% of Rick’s desired quality before final editing.
Rick aggregated a client’s historical coaching transcripts so the client could later ask questions against prior discussions. The stored context allowed AI to retrieve patterns and guidance from a long coaching relationship rather than respond only from general knowledge.
→ The client retained a useful, searchable extension of past coaching after direct work ended.
Common mistakes
Relying on One-Shot Prompting
A standalone request lacks the accumulated context and examples needed for consistent, business-specific work.
Uploading Everything Without Structure
A repository is useful only when its contents are relevant, organized, and sufficiently consistent for AI to interpret correctly.
Skipping Approval Gates
Allowing AI to move directly from raw notes to a final asset removes opportunities to correct the strategic direction early.
Is it for you?
Best for
Creators and teams that repeatedly produce similar outputs and have existing examples, transcripts, templates, or standards to supply as context.
Not ideal for
Businesses without reliable source material or anyone expecting uploaded examples to eliminate the need for review and revision.
From the transcript
“So I've created the Cloud project, which is Claude's version of a GPT or a Gemini gem.”
“I grab all those transcripts. I put them into the cloud project as examples of the types of scripts that I want”
“And then once I choose the hook, it writes the outline.”
From the episode
What 7-Figure CEOs Know About AI That You Don’t with Rick Mulready
Rick Mulready