What do I need to include in my prompt to get a well-built page?
A well-built page prompt has five layers: page identity, section list, design reference, content specifics, and hard constraints. Each layer adds specificity and reduces the gap between what you expected and what AI builds. According to Anthropic's prompt engineering documentation, structured, layered prompts consistently outperform broad single-paragraph instructions.[1]
Layer 1: Page identity
Name the page, its URL, and its purpose. "Build the homepage at index.html. Its purpose is to direct visitors to the free AI Visibility Scan."
Layer 2: Section list
List every section in order. AI fills sections more consistently when the full list is established before writing begins.
Layer 3: Design reference
Specify which existing files to match. Referencing actual code is always more powerful than describing styles in words.
Layer 4: Content specifics
Provide actual copy for key sections. Or specify the voice and constraints. "Write in a calm, intelligent, direct voice. No hype words."
Layer 5: Hard constraints
State what AI must never do. "All content must be in static HTML. No JavaScript-injected text. Schema JSON-LD must be in the head."
How do I describe my full site structure to AI?
The most reliable method is a plain text folder tree. Write your full architecture as a directory listing and ask AI to confirm the structure before building. A 2025 analysis by Vercel found that AI code generators produce 43% fewer structural errors when given explicit hierarchy maps upfront.[2]
vibecodeyourleads.com/
|-- index.html
|-- about.html
|-- /pillar-1/
| |-- index.html
| |-- /cluster-1a/
| | |-- index.html
| | |-- node-1.html
| | +-- node-2.html
| +-- /cluster-1b/
+-- /pillar-2/
Present this tree to AI and say: "This is the complete site architecture. Every page we build fits into this structure. When I ask you to build a page, I will tell you where it lives in this hierarchy. Confirm you understand before proceeding."
The confirmation step matters. Asking AI to confirm the architecture ensures AI has processed the hierarchy before generating files. And gives you a chance to catch misunderstandings before they cascade across ten pages.
How do I keep my site consistent when building across multiple sessions?
Use a persistent project brief. A single document that AI reads at the start of every session. For Claude Code, the project brief is the CLAUDE.md file in the project root. For Cursor, the brief is .cursorrules. For any AI tool, the brief can be a document pasted at the start of each conversation.
The project brief should contain:
- Site identity: domain, purpose, primary CTA, target audience
- Design system: hex color codes, font families and weights, spacing scale, component patterns (cards, buttons, CTAs, nav, footer)
- Architecture map: the folder tree for the entire site
- Hard constraints: no JS content injection, no inline styles, schema always in the head, all internal links must resolve or have TODO comments
- Reference files: "The nav and footer are established in index.html. Match them on every page."
With this brief active, each new prompt can be much shorter. Because the context that would otherwise need repeating is already present. The brief is an investment that pays compounding returns across every session for the life of the project.
What kinds of prompts give me the worst results?
Prompts that describe what AI should build without specifying what the output should match produce the worst results. GitHub's 2025 Copilot research found that prompts lacking reference files or design constraints generated outputs requiring 2.4× more revision cycles than prompts with explicit references.[3]
Examples of prompts that produce poor results:
- “Build me a contact page.”. No reference to the site’s existing design. Result: a contact page that may look nothing like the rest of the site.
- “Make the homepage more professional.”. No specific description of what professional means in this context. Result: AI makes changes according to its own interpretation of “professional.”
- “Add a FAQ section.”. No specification of the HTML structure (details/summary vs. JavaScript accordion), the design, or the questions to include. Result: inconsistent structure.
The fix in every case is the same: add specificity. Describe what the output should match, what it should contain, and what it must never do. Ambiguity in prompts is not AI's problem to solve. It is yours to eliminate.
How do I get AI to include schema markup on every page?
Include schema requirements directly in every page prompt. Schema markup is a required component, not an afterthought. Pages with properly implemented schema achieve a 41% citation rate in AI search compared to 15% without structured markup, according to research by Frase.[4]
Specify the exact schema types: "This page requires BlogPosting, FAQPage, and BreadcrumbList schema in a single application/ld+json block in the head. No schema may be injected via JavaScript."
For Authority Directory builds. Where every node carries the same schema stack. Include the schema template in the project brief. A schema template in the brief ensures schema is never forgotten and always sits in the static HTML source where AI crawlers can read it.
The VCYL Perspective
The CLAUDE.md file for this site is over 2,000 words. It describes the complete architecture, the design system with exact hex codes, the component patterns for every repeated element, the schema stack for every page type, and a list of things AI must never do. Every session, Claude reads all of it before writing a single character of code.
That investment is what makes every subsequent prompt short and precise. When I ask Claude to build a new node page, the prompt is often just a dozen lines. The global rules are already established. The brief does 80% of the work.
What I notice when people are frustrated by AI-generated websites is that the brief is almost always missing. They re-explain their colors, fonts, and constraints in every session. A two-hour brief investment saves hours of correction work and improves output quality from the very first page. The site built from a clear brief is a fundamentally different site than one assembled prompt by prompt without one.
