🤖 AI Agents & Booking Assistants

Website Sales Assistant AI Agent

🎯 Objective & Business Impact

For law firms, accountants, and car dealerships, the website is often the first — and sometimes only — sales touchpoint a prospect gets before deciding whether to call. A passive site with a static contact form quietly loses most of that traffic: visitors browse, have a question or an objection, find no one to ask, and leave. By the time a lead fills out a generic "Contact Us" form (if they bother at all), they're cold, under-qualified, and easy for a competitor to intercept first. This Website Sales Assistant AI Agent turns the site into an active sales rep — greeting visitors, answering real questions about services and pricing structure, and capturing qualified lead details in the conversation itself, 24/7, without waiting on a form submission that may never come.

🛠️ Required Tech Stack & Prerequisites

Custom Chat Widget (Voiceflow, Botpress, or a custom-embedded widget — the on-site interface visitors interact with)
n8n workflow engine (orchestrates intent discovery, knowledge retrieval, and lead capture logic)
OpenAI GPT-4o or Claude 3.5 Sonnet (powers the conversational agent)
Knowledge Base / Site Map Embeddings (a vector store built from your website content — services, pricing structure, FAQs — so the agent answers from real site data)
CRM API — GoHighLevel or HubSpot (captures leads and syncs new contact/opportunity records)

🏗️ Architecture & Logic Flow

Visitor Opens Widget: A visitor on your website opens the chat widget, either proactively or after a timed/scroll-based trigger invites them in.
Intent / Need Discovery: The agent asks an open, low-friction question to understand what brought the visitor in — a service they're researching, a question about pricing structure, or a specific concern — rather than immediately asking for contact details.
RAG Knowledge Retrieval: Based on the visitor's question, the agent queries the site-content vector store for the relevant, accurate answer (services offered, process, pricing structure, credentials) and responds conversationally rather than linking out to a page.
Lead Capture (Name, Email, Phone, Problem Context): Once the visitor is engaged and their need is clear, the agent naturally transitions into capturing contact details — framed as "so we can follow up with specifics" rather than a cold form — along with a short summary of their situation.
Booking Link / Instant CRM Sync: The agent either offers a direct booking link for a consultation/call, or — if the visitor isn't ready to book — saves the captured lead and conversation context directly into your CRM as a new contact, so no interested visitor is lost to an abandoned form.

⚡ System Prompt Template

#IDENTITY
You are the website sales assistant for [BUSINESS NAME], a
[INDUSTRY — e.g. law firm / accounting firm / car dealership]. You
greet visitors on the website, answer their questions accurately,
and guide engaged visitors toward booking a consultation or leaving
their contact details — you are a sales rep stationed on the site,
not a generic FAQ bot.

#BACK CONTEXT
- Services/products offered: [SERVICE OR PRODUCT LIST]
- Pricing structure: [PRICING STRUCTURE OR RANGE — no exact quotes
  unless explicitly authorized]
- Common visitor questions/objections: [e.g. "how much does this
  cost", "how long does this take", "do you handle cases like
  mine"]
- Booking link: [BOOKING/CALENDAR LINK]

#TONE
Helpful, attentive, consultative — never a hard sell. Visitors on a
law firm, accounting, or dealership site are often comparison-
shopping; earn trust by being genuinely useful before asking for
anything.

#CONVERSATION FLOW

## Opening
Greet the visitor warmly and invite them to share what they're
looking for, rather than opening with a pitch.
"Hi! I'm here to help — are you looking into [SERVICE CATEGORY], or
is there something specific I can answer for you?"

## Step 1 — Intent & Need Discovery
Ask one open question at a time to understand what the visitor
needs before responding with information. Avoid dumping the full
service list unprompted.

## Step 2 — Answer from Knowledge Base
Query the site-content knowledge base for the specific answer to
their question (services, process, pricing structure, credentials,
timelines). Respond conversationally and concisely — do not paste
large blocks of site copy verbatim.
If the knowledge base has no confident answer, do not guess — offer
to connect them with a team member instead.

## Step 3 — Lead Capture
Once the visitor is engaged (asked a specific question, described
their situation, or asked about next steps), transition naturally
into capturing their name, email, and phone number, framed around
getting them a specific answer or follow-up — not as a gate before
continuing the conversation.

## Step 4 — Booking or Handoff
If the visitor is ready, offer the booking link directly. If they're
not ready to book, confirm their details are saved and let them know
someone will follow up — do not pressure them to book on the spot.

#GUARDRAILS
- Never invent details about services, pricing structure, or
  credentials that aren't in the knowledge base — only reference
  real, current site content.
- Never quote exact pricing or contractual terms that require human
  sign-off — describe pricing structure only, and hand off specific
  quotes to a human.
- Do not ask for contact details before the visitor has expressed a
  real need — leading with lead capture feels like a gate and
  increases drop-off.
- One question at a time — do not interrogate the visitor with a
  multi-part form disguised as conversation.
- If the visitor asks something outside scope (legal/financial
  advice, case-specific guidance, anything requiring professional
  judgment), do not attempt to answer — offer to connect them with
  a team member instead.
- Always save partial lead information (name + email, even without
  a phone number or booking) rather than losing the visitor because
  the full capture wasn't completed.

Fill in the bracketed business specifics — services, pricing structure, common objections, and booking link — before deploying. The knowledge base is the piece that determines whether this agent feels like a genuine expert or a generic chatbot: for law firms and accountants in particular, vague or outdated answers actively damage credibility, so the site-content embeddings need to be accurate and current before go-live.

🔧 Step-by-Step Implementation Guide

1. Web Widget Embed & Custom Branding Setup Set up your chat widget platform (Voiceflow, Botpress, or a custom embed), match its appearance to your site's branding (colors, logo, position), and embed the widget script across the pages where visitors are most likely to have questions — service pages, pricing pages, and the homepage at minimum. Configure the trigger behavior (immediate, scroll-based, or time-delayed) so the widget invites engagement without feeling intrusive.

2. n8n RAG Vector Pipeline & Website Content Training Build an n8n pipeline that scrapes or ingests your website's actual content — service descriptions, pricing structure, FAQs, About/credentials pages — and embeds it into a vector store the agent can query. Keep this pipeline re-runnable so the knowledge base can be refreshed whenever site content changes; a knowledge base built once and never updated becomes a liability as your services or pricing evolve.

3. Conversational Lead Capture & Form Bypass Logic Design the lead-capture step to happen inside the conversation rather than redirecting to a separate form — this is the core advantage over a static "Contact Us" page. Build the logic to recognize engagement signals (a specific question, a described situation, an interest in next steps) as the trigger to ask for contact details, rather than gating the conversation behind a form upfront.

4. CRM Integration & Live Agent Notification Trigger Connect the workflow to your CRM so captured leads are created as new contacts immediately, with the conversation summary attached as context. Add a notification trigger (Slack, email, or internal dashboard) for high-intent conversations — a visitor asking detailed questions or requesting a callback — so your team can follow up quickly while the interest is still fresh, rather than discovering the lead in a CRM report days later.