Core Idea
The creation request is not where AI output quality is decided — the context loaded before the creation request is. Most coaches who get generic output from Claude are not using a weak model or writing weak prompts; they are asking the model to generate something about a client it has never met. The page reads like a stock template because, to the model, the client is a stock template.
The fix is sequencing, not prompting. By front-loading the Ideal Client Handbook and then talking with the model about the client before asking for any deliverable, you force the model to build a specific mental representation of who the page is for. Every subsequent generation is then anchored to that representation rather than to the generic “coach client” the model would otherwise default to.
What breaks when you skip this: the model fills in the blanks with population-average assumptions about your client (cost-sensitive, outcome-obsessed, short attention span, standard objections). Those assumptions are not wrong — they are just not yours. And because they are invisible, you cannot spot them in the output. You just feel that the page is “fine” but not alive. By the time you notice, the generic framing has already set the tone for every edit you make on top of it.
The Insight
Before asking Claude (or any AI) to create a website page, design a product, or write content, load your Ideal Client Handbook (ICH) into the project or conversation first — then have a substantive conversation about the client’s situation, needs, and context before making any creation request.
Donald Kihenja described this directly in response to Don Back’s question about how to use Claude to update an old website with new positioning:
“Make sure you have your ICH in your project or chat then start chatting in it, then go from there. Chat for a while, let it get context, ask it recommendations on design etc then when you ask it to create the page it will be super smart for you.”
The Framework
Step 1 — Load context: Add your ICH (Ideal Client Handbook) to the Claude project or paste it into the conversation. This grounds every subsequent response in your specific client psychology.
Step 2 — Chat before creating: Don’t jump straight to “write me a page.” Instead, have a real conversation about your client’s situation, your positioning, what the page needs to accomplish. Let the model build a mental model of your context.
Step 3 — Ask for design recommendations: Before creating, ask the AI what it thinks would work well given what it now knows. This surfaces the model’s reasoning before it starts generating.
Step 4 — Create from context: Now make the creation request. The output will reflect a richer understanding of your ideal client’s psychology and your positioning because the model has been contextually loaded before generating.
Why This Matters
Most AI-generated website or content output feels generic because the model is working from generic context. The ICH-first protocol is a systematic way to eliminate that genericity before the creation phase begins — not by giving better instructions, but by giving more relevant context.
The underlying principle: AI output quality scales with context quality. A single well-loaded conversation produces better results than a dozen bare prompts.
Kasimir’s Extension
Kasimir Hedstrom demonstrated a related pattern during the same session: using Claude’s co-work (Claude extension on a browser page) to directly build website content on a live platform, letting Claude scan the page and generate or update content in-place. His recommendation was to use system.io as a platform for this, and to use co-work rather than Comet Assistant, which he found slower.