Most website chatbots are glorified FAQ buttons. They don’t know what your business does, who your team is, or how to tell a serious prospect from a tire-kicker. So visitors get generic responses, lose interest, and leave.
EmpireVault’s AI chat agent is different. It reads a set of context files that you control — plain-text documents that teach the agent everything it needs to know about your business. The result is an agent that can hold a real conversation, answer specific questions about your services, and qualify leads using your actual criteria.
This post explains what context files are, how they work under the hood, and how to get the most out of them.
What Are Context Files?
Context files are Markdown documents that define your AI agent’s knowledge and behavior. Every time a visitor sends a message, the agent reads all your enabled context files and uses them to craft its response. Think of them as the agent’s training manual — except you can edit it anytime and changes take effect immediately.
When you register a site with EmpireVault, four default context files are created automatically:
- Business Overview — What you do, the problems you solve, and who you serve
- Services & Team — Your services, expertise, and the people behind the business
- Qualification Guide — How the agent should assess and qualify leads
- Conversation Rules — Tone of voice, hard rules, and how to handle edge cases
The first two are about what your business is. The last two are about how the agent should behave. Together, they give the agent enough context to represent your business authentically.
The Two Types of Context Files
Crawl-Populated Files (Business Knowledge)
Your Business Overview and Services & Team files are marked as crawl targets. This means EmpireVault can automatically populate them by crawling your website.
When you click “Start Crawl” from your site dashboard, EmpireVault:
- Visits up to 20 pages on your site, following links up to 2 levels deep
- Extracts the main content from each page (stripping navigation, footers, and boilerplate)
- Sends the combined content to AI, which distills it into structured markdown for each crawl-target file
- Analyzes what it found and suggests new context files you might want — like FAQ, Pricing, or Case Studies
The crawl is a starting point, not the final word. You can (and should) review the generated content and refine it. The agent will only ever say what’s in your context files — it won’t make things up.
Manual Files (Behavior Rules)
Your Qualification Guide and Conversation Rules are not crawl targets — they’re about behavior, not business facts. These come pre-populated with sensible defaults that work for most businesses.
The Qualification Guide tells the agent how to assess visitors:
- What information to gather naturally during conversation (name, company, needs, timeline)
- How to score lead quality — strong leads have clear needs and decision-making authority; weak leads are vague and early-stage
- How to close the conversation with a clear next step (book a call, follow up by email)
The Conversation Rules set boundaries:
- Tone: Professional and warm, adapting to the visitor — more formal with executives, more relaxed with founders
- Hard rules: Never invent facts, never quote prices, never discuss competitors by name
- When stuck: Honestly say “I’ll have someone follow up” and ask for their email
- Off-topic: Politely redirect to your services
These defaults are battle-tested, but you should customize them for your business. A law firm might add “Never provide legal advice” to the hard rules. A SaaS company might want the agent to discuss pricing tiers. A recruitment agency might want different qualification criteria entirely.
Create Your Own Context Files
The four defaults are just the starting point. You can create any number of custom context files for your specific needs:
- FAQ — Common questions and answers, so the agent doesn’t have to improvise
- Pricing — Plan details, what’s included, how billing works
- Case Studies — Real examples of client work the agent can reference
- Company History — Background story, founding mission, milestones
- Product Features — Technical details for visitors evaluating your product
- Compliance — Industry-specific rules the agent must follow
Each file can be marked as a crawl target, meaning the site crawl will automatically populate it with relevant content from your website. Or leave it as manual-only and write the content yourself.
You can also disable a context file without deleting it. Disabled files are excluded from the agent’s knowledge — useful for seasonal content or A/B testing different approaches.
Smart Crawl Suggestions
After running a site crawl, EmpireVault doesn’t just update your existing files — it analyzes what it found and suggests new ones. If your site has a pricing page, the crawl might suggest a “Pricing” context file with draft content already filled in. If it finds testimonials, it might suggest a “Client Testimonials” file.
You review each suggestion, preview the draft content, and accept or dismiss with one click. Accepted suggestions become new context files that the agent can immediately use in conversations.
How It All Comes Together
Here’s what happens every time a visitor sends a message to your AI agent:
- The agent loads all enabled context files, ordered by position
- A safety preamble is prepended — rules that prevent prompt injection, secret leaking, and other attacks
- The context files are assembled into a system prompt: your business knowledge + your behavior rules
- The visitor’s message (plus conversation history) is sent to AI alongside this system prompt
- The agent responds with knowledge of your specific business, following your specific rules
The more detailed and accurate your context files are, the better your agent performs. A business with well-written context files gets an agent that sounds like a knowledgeable team member. A business with placeholder content gets generic responses.
Best Practices
1. Run the crawl first, then refine
The site crawl gives you a solid starting point. Review the generated content, fix any inaccuracies, and add details the crawl couldn’t pick up (like pricing, internal processes, or team bios that aren’t on your website).
2. Be specific in your Business Overview
Don’t say “we help businesses grow.” Say “we build custom Shopify stores for DTC brands doing $1M-$10M in annual revenue, specializing in subscription commerce and post-purchase upsells.” The more specific you are, the better the agent qualifies leads.
3. Customize your Qualification Guide
The default guide works for most B2B services, but your business has specific criteria. What makes someone a great client for you? What are the red flags? What’s the ideal next step — a call, a demo, a free trial? Tell the agent.
4. Add hard rules that matter to your industry
Healthcare? Add HIPAA language. Financial services? Add compliance rules. Legal? “Never provide legal advice.” These rules are non-negotiable — the agent follows them even if a visitor pushes back.
5. Create an FAQ file
If you find the agent struggling with certain questions, create an FAQ context file with the exact answers. This is the fastest way to improve response quality for specific topics.
6. Review and iterate
Read your conversation logs. When the agent gives a wrong or weak answer, check which context file should have covered it — and update it. Context files are living documents, not set-and-forget.
Getting Started
Context files are available on every EmpireVault plan. To set them up:
- Go to Engage > Sites and select your site
- Click Start Crawl to auto-populate your business knowledge files
- Click Edit Context Files to review and customize
- Use the editing guides (shown above each file) for tips on what to include
- Create custom files for FAQ, pricing, or anything else your agent should know
The whole process takes about 15 minutes, and the improvement in conversation quality is immediate.