The AI Chatbot Knowledge Base Template (Build It in Under an Hour, Maintain It in 15 Minutes/Month)

The quality of a local-business AI chatbot is almost entirely determined by the quality of its knowledge base. A great LLM (Claude 4.7, GPT-5) with a weak knowledge base produces vague, generic answers that drive customers away. A mediocre LLM with a strong knowledge base produces specific, helpful answers that capture leads.

The knowledge base is the leverage point. This article is the structure, the extraction prompt, and the maintenance cadence that keep it durable.

The Six-Section Structure

Every local-business knowledge base should have these six sections in this order:

1. Business Overview. Two sentences. "Acme Family Dental is a general and cosmetic dental practice in Tampa, FL, serving patients of all ages. Dr. Patel and Dr. Chen lead a team of 8 professionals."

2. Services. Bulleted list. Each service is a one-line description. Include price where publicly stated. "Routine cleaning: $129. Includes exam, X-rays as needed, polish."

3. Hours & Holidays. Standard hours per day. Holiday exceptions for current year. "Mon-Thu 7am-6pm. Friday 7am-3pm. Closed weekends. Closed Dec 24-Jan 1."

4. Insurance & Payment. Specific insurance plans accepted (named, not "most major plans"). Payment methods. Financing options.

5. Location & Contact. Address, parking instructions, primary phone, after-hours contact (if any), email for non-urgent inquiries.

6. Common Questions. 15-20 actual questions the owner answers most often, with specific answers in the owner's voice.

Total length: 2-3 pages in markdown. Anything longer creates LLM context-window strain and maintenance impracticality.

The 30-Minute Capture Workflow

The capture move is interviewing the owner once and letting Claude do the structuring. Faster than typing the knowledge base yourself.

Step 1 — Schedule a 30-minute call with the owner. Tell them: "I'll ask about the 20 most common customer questions and your standard answers. We'll record so I can build your knowledge base from it."

Step 2 — During the call, ask in this order:

1. "Tell me the 5 services your customers most often ask about, and what you usually tell them."
2. "Walk me through your hours, including any holiday closures coming up."
3. "Which insurance plans do you accept? Be specific — name the providers."
4. "What are the most common questions you get by phone that aren't about booking? Tell me how you usually answer each."
5. "Are there any topics you DON'T want the chatbot to handle — anything that should escalate to you personally?"

30 minutes covers all five. Record it. Most phones and laptops do this natively.

Step 3 — Transcribe. Most recording tools do this automatically (Fathom, Otter, Granola). If yours doesn't, paste audio into OpenAI Whisper.

Step 4 — Run the Claude extraction prompt. See below.

Step 5 — 15-minute owner review. Send the draft knowledge base to the owner. Ask: "Any answers feel off?" Most owners flag 2-3 small items. Fix and finalize.

Total operator time: ~45 minutes. Total owner time: ~45 minutes (30 interview + 15 review).

The Claude Extraction Prompt

Paste this into Claude with the interview transcript:

You are a Senior Knowledge Engineer building a structured knowledge base for a local-business AI chatbot. Convert the owner interview transcript into a 6-section knowledge base in markdown.

Sections (in this order):

1. Business Overview (2 sentences)
2. Services (bulleted list, prices included where stated, one line each)
3. Hours & Holidays (standard week + current-year holiday exceptions)
4. Insurance & Payment (specific plans named; payment methods; financing options)
5. Location & Contact (address, parking, phone, email, after-hours)
6. Common Questions (15-20 Q&A pairs in the owner's own phrasing)

Style rules:

• Use the owner's exact phrasing in Common Questions answers
• Be specific (named insurance plans, specific dollar amounts, specific times)
• No marketing fluff
• Total length under 3 pages
• If the interview is missing information for any section, mark [TBD: ] and ask one consolidated clarifying question at the end

For Common Questions, prioritize questions in this order: (1) operational (hours, location, services), (2) commercial (pricing, insurance, booking), (3) qualifying (new patient questions, specific service questions).

Claude returns a structured knowledge base in 2-3 minutes of model time.

The Common Questions Section (Where Most Knowledge Bases Fail)

The Common Questions section is the differentiator. Most installers paste 10 generic FAQs from the business's website and call it done. The website FAQ is almost never the questions customers actually ask.

The right approach: ask the owner what questions they get on the phone. Those are the real questions. Specific. Often weirder than website FAQ. Always more useful.

Example (dental office):

• Q: "Do you take Delta Dental?"
• A: "Yes, we're in-network with Delta Dental PPO and Premier. We're not in-network with Delta DHMO. We can verify your specific plan if you call us with your member ID."

vs. the generic website version:

• Q: "Do you accept insurance?"
• A: "We accept most major dental insurance providers. Please contact us to verify your specific plan."

The first answer captures the customer. The second loses them.

What NOT To Include

A few things that bloat the knowledge base without adding value:

Owner biography. Customers asking the chatbot rarely care about the owner's background. If they do, they'll check the About page.

History of the business. Same.

Mission/vision statements. Marketing copy. Skip.

Generic "we care about quality" language. Adds words, no information.

Detailed staff bios. A name + role line is enough. "Dr. Patel — Lead Dentist" not three paragraphs.

Long-form policies. Cancellation policies, payment terms, etc. Summarize in one line each. Link to the full version if needed.

If the knowledge base feels longer than 3 pages, audit for these patterns first.

The Monthly Maintenance Cadence

Operators running the $99/mo maintenance retainer should do this monthly:

Minute 1-10: Check in with the owner. Email or text: "Any updates this month? New services, hours changes, new staff?"

Minute 11-20: Update the knowledge base with anything they flagged.

Minute 21-30: Run a "stress test" — paste 5 common questions into the chatbot, check the answers are still accurate.

Minute 31-40: Review the captured-leads log for the month. Look for questions the chatbot couldn't answer confidently. Those signal knowledge base gaps.

Minute 41-45: Send the monthly report email to the owner (covered in the chatbot install failures article).

Total monthly maintenance: ~45 minutes per client. At $99/mo, that's ~$130/hour effective rate.

The Update Triggers To Watch For

Beyond the monthly cadence, certain triggers require immediate updates:

• New service launched
• Service discontinued
• Pricing change
• Hours change (especially holiday closures)
• New insurance plan accepted (or dropped)
• New staff with public-facing role
• Address change or expansion to second location
• Brand color change (affects widget styling but not knowledge base directly)

Most local businesses email or text these updates to you in real-time once they're trained to. Set the expectation at install: "Text me when anything changes — I'll update within 24 hours."

The Quality Test

Before going live with a new knowledge base, run this 10-question test. The chatbot should answer all 10 confidently:

1. What are your hours today?
2. Do you take [common insurance name in their area]?
3. Where are you located and is there parking?
4. How much is [their most common service]?
5. Can I make an appointment?
6. What payment methods do you accept?
7. Do you take new patients?
8. What's the difference between [Service A] and [Service B]?
9. Are you open on [upcoming holiday]?
10. How do I reach you outside business hours?

If the chatbot fails 2+ questions, the knowledge base needs work before going live.

The Cross-Sell

The AI Chatbot for Local Business in 60 Minutes guide ($6.99) includes the knowledge base template as a fillable markdown document, the Claude extraction prompt, the monthly maintenance checklist, the owner training script, the 10-question quality test, and the $99/mo install-and-handoff offer template.

$6.99 once. Most operators recoup the cost on the first install where the templates save 2+ hours of manual knowledge base writing.

The actionable next step: if you've already installed any chatbots, audit them against the 6-section structure this week. Most installs will have weak Common Questions sections (the most common gap). Strengthen those first — it's the single highest-leverage knowledge base improvement.