How to build an AI content system around your expertise

Give the system the context you use

In his content-system walkthrough, Nick describes a Claude Code setup built around a short direction file, an audience profile, voice guidance, accumulated lessons and separate skills for scripts, review and captions.

The useful principle is to write down the decisions you otherwise repeat in every prompt. Who needs the content? What do they already know? Which examples can you support? What language sounds like you?

• Direction: the purpose of the content and the rules every draft follows.
• Audience: the person, their problem and the decision you want to help them make.
• Voice: examples of your writing and the wording you avoid.
• Source material: the original idea, demonstration, transcript or notes.
• Review criteria: the checks a draft must pass before publishing.

Give writing and review different jobs

A practical way to implement that structure is to define an input and an output for each stage. A script brief contains the claim, evidence and audience. The writer produces a draft. The reviewer identifies specific gaps. The caption stage works from the approved idea.

Ask the reviewer to check whether the promised demonstration actually exists, whether a number has a source and whether the reader can understand the next step. ‘Make this better’ is much harder to evaluate than a named requirement.

Adapt one useful idea to each format

Keep the substance consistent as the format changes. A video can show a workflow on screen. A newsletter can explain why you chose it. A LinkedIn post can focus on the business decision. A short caption can point to the useful moment.

A simple repurposing brief helps: name the original source, the audience for the new piece, the one point it should carry and what the reader can do next. Different formats need different openings and levels of detail.

Check the draft against the original

Before publishing, compare the draft with the source. Keep caveats attached to claims, distinguish an example from a client result and remove details the source does not support. Give technical instructions a separate accuracy check when tools have changed.

Store corrections with the relevant writing or review instructions. The next draft should benefit from what you learned rather than require the same explanation again.

• Does the opening describe something the piece actually delivers?
• Can you trace every factual claim to a source or demonstration?
• Does the language sound like your existing work?
• Is the next action useful for this particular audience?

Start with one complete piece

Use a recent idea you understand well. Run it through the full process, keep the edits and improve the instructions before increasing volume. The first goal is a draft you can confidently use. More output becomes useful once that quality is repeatable.