The fastest way to train a content creation agent on your voice is to give it three things: a written description of how you talk (your tone, your pet phrases, what you avoid), three to five examples of content you are proud of, and a description of your audience using the exact words they use to describe their own problems. With those three inputs in the agent’s context, the output will sound like you rather than like a generic AI article.
Why Generic AI Content Sounds Generic
When an AI produces content without voice context, it defaults to the average of everything it has read. That average is technically competent, grammatically correct, and completely unmemorable. It sounds like a thousand other pieces of content in your niche. Your students would not recognise it as yours.
Your voice is not just tone — it is the specific phrases you use, the analogies you reach for, the things you refuse to say, the level of directness you bring, and the way you signal respect for your reader’s intelligence. None of that is in the default AI output unless you put it there explicitly.
Building the Voice Context Document
Start by writing a voice guide — not for AI, but for yourself. Answer these questions: What three adjectives best describe how I want my content to feel? What do I always say that sounds like me? What phrases do I hear from other educators that I would never use? What is my typical sentence length and structure? Do I use contractions? Do I ask rhetorical questions? Do I tell stories?
Then add your audience language layer. List the exact phrases your students use when they describe their problems — copy them from community posts, emails, and intake forms. “I’m not tech-savvy” sounds different from “I’m not a digital native” sounds different from “I always feel like I’m behind on the tech stuff.” The specific phrasing your audience uses is the vocabulary your agent should write toward.
Finally, paste in three to five examples of content you are genuinely happy with — a blog post, an email, a community post, a section of a lesson. Tell Claude: “These examples represent my voice at its best. When in doubt, write more like this.” Examples are more powerful than descriptions. Showing is faster than telling.
Where to Put the Context
In Claude, this voice guide goes into the system prompt — the standing instructions that apply to every conversation. If you are using Claude in Cowork mode or a similar agent setup, it goes into a configuration file that loads at the start of each session. Either way, the agent reads it before generating anything, which means every output is calibrated to your voice before you see it.
Update the guide whenever you notice the agent drifting — adding a new example of something it did well, or a new instruction about something it keeps getting wrong. The guide improves over time as you refine it.
What This Means for Educators
A well-configured voice context is the difference between content you publish and content you rewrite before publishing. Educators who invest 90 minutes building a strong voice guide typically save that time back within the first two content production sessions — and then keep saving it every week after that.
The Simple Rule
Tone description, audience language, best examples — put all three in the agent’s context before asking it to write anything. The more specific the guide, the less editing you do on the output.