Claude, Guideline Generation

A natural guess is that when Claude writes a brand voice guideline, it forms its own opinion about what your writing should sound like, the way an editor with good taste would. The video says that is not what happens. The guideline comes from a written specification that pulls patterns out of your own samples.

The wrong guess: an editor's ear

The assumption is that a guideline comes from good literary taste. The skill does not work that way. It is a specification: a written instruction set that tells Claude exactly how to read the source material and pull out the same kind of patterns every time, run the same way no matter whose writing it is. Because it is a procedure and not a sensibility, there is no step in it for deciding whether your voice is any good.

Four stops from source to document

The video describes what the skill does as four stops. It reads the source material, extracts the patterns the file defines, structures them into a guideline, and returns the document.

The worked example is ten blog posts handed over to produce a brand voice guideline. They are read, run through the same extraction steps, structured into sections, and returned as one guideline. The tone, word choices and examples in it are pulled straight from those ten posts.

Why the result repeats

The guideline holds steady because the extraction runs the same way every time. Feed the same ten posts through twice, and the tone notes, word choices and examples come back identical. That repeatability is the practical value of a written procedure over an improvised judgment.

Where it stops

The limit follows from the same design. Ask something outside extraction, such as whether the voice in those posts is actually a good one, and there is nothing tailored to reach for. The skill stops exactly where its extraction steps stop. The video's point is that a brand voice guideline from Claude is a fixed extraction run on your source material, not Claude's own taste in good writing.

What the video does not claim

The video presents this as a plain-language retelling for a general audience of an earlier skill showcase. It makes no claim about any specific company or brand, or about the literal contents of the skill's instruction file. It describes only the general mechanism, a written procedure that extracts from your material, and the two directions in which it can fail: reading it as taste, and expecting a judgment it was never written to give.

Key takeaways

• A brand voice guideline from Claude does not come from Claude's private taste in writing.
• The guideline-generation skill is a written specification that Claude reads before acting.
• It reads the source, extracts defined patterns, structures them, and returns one guideline document.
• Ten blog posts in gives a guideline of tone, word choices and examples drawn from those posts.
• The same samples run twice produce the same guideline.
• Asking whether the voice is good falls outside the steps, so the skill has nothing tailored for it.
• The skill stops the moment you ask it to judge rather than extract.

Try it yourself

Hand Claude a handful of your own writing samples and ask it to run the guideline-generation skill: read them, extract the patterns, and return a brand voice guideline. Then ask whether that voice is actually a good one. Watch whether Claude invents an opinion or tells you plainly that judging quality is outside what the skill covers. That reaction is the fastest way to see that the guideline comes from a procedure rather than taste.