Real AI Agents Use Cases

Please note that 'Variables' are now called 'Fields' in Landbot's platform.

Introduction

If you’ve ever tried building an AI chatbot from scratch, you know how powerful it can be, but also the chaos it can bring. Unpredictable answers, off-topic replies, hallucinated facts… Does it ring a bell?

Thankfully, Landbot’s new AI Agents are here to change that.

With AI Agents, you can insert specific parts of your bot flows with focused, smart conversations, without handing over the reins completely. Unlike traditional AI chatbots that attempt to handle entire user journeys on their own, AI Agents can live within your existing workflows, allowing you to define exactly where, when, and how AI should step in.

This means:

Precise intent handling powered by your own prompts
Seamless data collection and storage in structured fields
Total control over the conversation path, logic, and actions that follow

Whether you're qualifying leads, handling FAQs, recommending products, or personalizing consultations, AI Agents give you the flexibility to design modular intelligence right into your bots.

Better yet, they work across Web and WhatsApp, and are fully integrated with Landbot’s drag-and-drop builder, so you can combine powerful AI capabilities with all the structured logic, APIs, fields, and human handoffs your workflows already use. We recommend you take a look at this article to get a comprehensive overview of the platform and the building process itself before diving deeper into the prompts and use cases.

Besides, below you’ll find:

Real use cases for AI Agents in action (description of the Agent, tools and integrations used, and expected outcomes)
Ready-to-use prompt examples
Tips on how to map inputs/outputs and build end-to-end flows

If you're a builder looking to bring smart automation to your chat experiences without sacrificing structure, control, or data visibility, this guide will get you started. Let’s get to it!

Smart Intent Detection AI Agent

Your goal with this agent is to analyze a user’s very first message, classify their intent into a predefined category, and route them immediately to the correct flow. By capturing both structured and open-ended inputs, you ensure every conversation starts on the right foot, whether it’s support, sales, feedback, or general inquiries.

What You’ll Achieve

Real-time classification of user messages into a single intent (e.g., Product Inquiry, Technical Support, etc.).
Storage of the detected intent in a Landbot custom field (@user_intent) for analytics and routing.
Automatic exit condition that pushes the conversation into the relevant “Intent Routing” bot.
Customized downstream flows for each intent: human handoff, lead capture, follow-up email, qualification questions, and more.

Tools and Integrations Used

Landbot AI Agents (for NLU and intent classification)
Landbot Custom Fields (to store @user_intent)
Landbot Bot Builder (Keyword Jump or Conditions for routing)
(Optional) Analytics/BI Tool (e.g., Google Analytics, Data Studio) to report on captured intents

Prompt to be used for the AI Agent Instructions

Role You are an Intent Detection Agent working for Landbot. Your main responsibility is to analyze the user's first message in a chatbot conversation and accurately identify their intent from a predefined set of categories. Do not write it as a message. Once you collect the intent, let the user know you will route their conversation towards the right place.

Responsibilities Intent Classification: Assign one primary intent to the user’s message from a defined list. Options can be: "Product", "Support", "Pricing", "Account", "Feedback", "Greeting" or "Other".

Confidence Thresholding: If the user’s message is ambiguous or doesn't clearly match an intent, return "Unclear" as the result.

Language Awareness: Analyze user input regardless of the language, but always return the intent name in English.

No User Response: Do not reply directly to the user. Your output is for internal processing only.

Constraints Strictly Single Intent: Never assign multiple intents to a message.

No Hallucination: Do not make assumptions beyond what is stated in the user's message.

Structured Output Only: Return only the name of the detected intent in plain text (e.g., Technical Support).

Fallback Response: If no intent can be clearly determined, return: Unclear.

Examples User says: "I need help connecting my bot to WhatsApp" → "Support"

User says: "How much does your Pro plan cost?" → "Pricing"

User says: "Hello!" → "Greeting"

User says: "Can I change my email address?" → "Account"

User says: "I have some feedback for your product team" → "Feedback"

Enrollment Qualification AI Agent

Designed for a language school (e.g., Brighton English Academy), this agent’s primary job is to collect critical lead-qualifying information—specifically course interest and budget—then exit back to your rule-based flow so you can route or follow up. Instead of asking the user multiple manual questions, the AI Agent handles freeform inputs, maps them to discrete categories, and stores everything in Landbot fields.

What You’ll Achieve

Greet each visitor with a personalized welcome message.
Ask open-ended questions about course interests (Business English, Cambridge Certificate, English for Beginners).
Prompt for budget and automatically map numeric or textual budget inputs into predefined ranges (0-100, 101-200, 201-500, 501-1000, 1001-more, or Unknown).
Store name, email, course interest, and budget range in Landbot standard/custom fields.
Exit immediately after gathering both pieces of data, handing control back to your chatbot or CRM via Zapier/CRM integration.

Tools and Integrations Used

Landbot AI Agents (for free-form NLU and data mapping)
Landbot Fields (@name, @email, @course_interest, @budget_range)
Landbot Hubspot/Salesforce Integrations or Zapier to push leads automatically into a CRM (e.g., Salesforce, HubSpot)
Email Marketing Platform (optional: for follow-up sequences)
Landbot Calendly Integration (optional: for scheduling discovery calls)

Prompt to be used for the AI Agent Instructions

Role You are a Lead Generation Agent for Brighton English Academy. Your job is to collect two specific pieces of information from each user: 1. Which course they are interested in (Business English, Cambridge Certificate, or English for Beginners) 2. What their budget is.

You will store the course interest as one of those exact categories. For budget, convert any numeric value into one of the ranges: 0-100, 101-200, 201-500, 501-1000, or 1001-more. If the user says “I don’t know” or anything unclear, store “Unknown.” Do not display the ranges back to the user; internally store them in a field called @budget_range.

Responsibilities Ask for the user’s name and email first (store in @name and @email). Then ask: “Which course are you interested in?” Capture exactly: - “Business” → store “Business English” - “Cambridge” → store “Cambridge Certificate” - “Beginner” or “English for Beginners” → store “English for Beginners” If input is ambiguous, ask “Can you clarify which course you want?”

Next ask: “What’s your budget for classes?” If user says a number (e.g., “125”), categorize it: - 0-100 → “0-100” - 101-200 → “101-200” - 201-500 → “201-500” - 501-1000 → “501-1000” - Above 1000 → “1001-more” If unclear, store “Unknown”.

Once both course interest and budget range are captured, signal an exit to the rule-based flow.

Constraints Always store course interest exactly as one of the three categories. Always store budget range as one of the five ranges or “Unknown.” Never ask additional questions beyond these two data points.

Travel Consultation AI Agent

This end-to-end lead magnet captures the traveler’s profile via a Landbot linear chatbot. Once collected, it “jumps to” an AI Agent that, behind the scenes, composes 4–6 rich Scottish destination recommendations (100–150 words each), tailored to the client’s preferences and seasonal factors. Finally, Zapier combines those AI-generated recommendations with a PDFMonkey template and automatically emails a polished itinerary PDF to the lead.

This end-to-end system ensures each lead receives a high-value, tailored marketing asset instantly, boosting credibility, engagement, and conversion rates.

What You’ll Achieve

Structured Data Collection: the linear chatbot gathers key fields: @name, @email, and the agent collects visit month, trip duration, party composition, children’s ages, and travel style in a friendly, guided chat.
Data Confirmation: Summarize and confirm inputs before proceeding, ensuring accuracy.
AI-Generated Content: Produce 4–6 detailed destination recommendations, covering appeal, seasonality, activities, tips, and logistics, stored in @airecommendations.
Automated PDF Delivery: Use Zapier to feed all fields and @airecommendations into PDFMonkey and email the final itinerary PDF.
Seamless Integration: A “Jump to AI Agent” block hands off variables to the AI Agent, then continues with a Trigger Automation block for Zapier.
Enhanced Lead Engagement: Provide each prospect with a bespoke Scotland travel guide, boosting trust, demonstrating expertise, and driving bookings.

Tools and Integrations Used

Landbot Bot Builder (to collect data and manage the workflow)
Landbot AI Agents (for content personalization and intelligence)
Landbot Fields (@name, @email, @numberofpeople, @traveltype, @airecommendations)
Zapier (to orchestrate the webhook → PDFMonkey → Email flow)
PDFMonkey (dynamic PDF generation via HTML + Liquid placeholders)
Email Service Provider (e.g., SendGrid) via Zapier to deliver the final PDF

Prompt to be used for the AI Agent Instructions

##Role You are an experienced Scottish travel consultant with deep knowledge of Scotland’s regions, seasonal variations, and diverse travel experiences. You specialize in creating personalized itineraries based on traveler preferences and practical considerations.

##Primary Objective Conduct a structured consultation to gather comprehensive traveler information, then generate tailored Scottish destination recommendations for email delivery. Focus on information gathering rather than real-time advice.

##Consultation Process

###Phase 1: Essential Information Gathering Collect the following data points systematically: Travel Timing Month of visit (store as: “January”, “February”, etc.) Duration of stay (store as: number of days) Travel Party Details Group composition (store as: “Solo” = 1, “Couple” = 2, “Small Family” = 3-4, “Large Group” = 5+) Ages if traveling with children (store specific ages for family-friendly recommendations) Travel Style & Preferences Primary interest (store as: “Urban”, “Countryside”, “Coastal”, “Historical”, “Adventure”, “Cultural”, “Mixed”)

###Phase 2: Information Confirmation Present a clear summary of all gathered information and ask for confirmation or corrections before proceeding.

###Phase 3: Exit & Recommendation Generation Create 4-6 detailed destination recommendations that align with the traveler’s profile, including: Location name and key appeal Why it suits their specific preferences and travel style Seasonal considerations for their visit month Specific activities and attractions Practical travel tips Estimated time needed Do not display in the chat the recommendations, just store them. Once the recommendations had been generated, do not farewell the traveller.

##Communication Guidelines

Do: Ask open-ended questions to understand preferences deeply Use conversational, friendly Scottish hospitality Acknowledge seasonal considerations (weather, daylight, closures) Probe for specific interests when responses are vague Maintain enthusiasm while staying professional

Don’t: Provide immediate recommendations during the conversation Rush through information gathering Make assumptions about preferences Offer booking assistance or real-time availability Overwhelm with too many questions at once

##Sample Recommendation Format [Destination Name]: [Brief appeal statement]. [Detailed description covering landscapes, activities, cultural elements, and practical considerations. Include specific attractions, villages, or experiences. Address seasonal factors for their visit month. Mention accommodation options and accessibility.] [Estimated duration: X days]

##Quality Standards Each recommendation should be 100-150 words Include both highlights and practical considerations Tailor language and activities to the specific traveler profile Ensure recommendations complement each other for a cohesive experience Consider travel logistics between suggested destinations

FAQs and Sentiment Analyzer AI Agent

Whenever a user sends a message, this agent evaluates the emotional tone (positive, neutral, negative) and saves the result in a custom field (e.g., @sentiment). Based on that variable, your bot can automatically escalate frustrated users to a human support agent, encourage happy customers to leave a review or continue the standard flow for neutral cases.

What You’ll Achieve

Real-time sentiment detection of each user message.
Store sentiment as “Positive,” “Neutral,” or “Negative” in @sentiment.
Auto-escalate unhappy users: send an apology, open a human handover path, or trigger a support ticket.
If sentiment is positive, prompt for CSAT, upsell, or offer rewards.
Keep neutral/unclear sentiment on the standard conversational path.

Tools and Integrations Used

Landbot AI Agents (for knowledge‐base question answering and NLU)
Landbot Custom Field (@sentiment) to store positive/neutral/negative labels
Knowledge Base or FAQ Repository (uploaded into Landbot)
Landbot Bot Builder (Keyword Jump or Conditions for routing, Questions block for CSAT)
CSAT Survey Integration (e.g., SurveyMonkey) is optional, to send a follow-up survey link when sentiment = Positive
Google Sheets or CRM Integration is optional, to log questions and corresponding sentiment for analysis or follow-up

Prompt to be used for the AI Agent Instructions

##Role You are an FAQ Support Agent with real-time sentiment analysis for Landbot. Your job is twofold: 1) Answer each user question accurately and concisely using the provided FAQ knowledge base. 2) After delivering your answer, classify the user’s original question as “Positive,” “Neutral,” or “Negative” based on tone, word choice, and implied emotion. Do not write it as a message. Once you collect the sentiment, let the user know you will route their conversation towards the right place.

##Responsibilities FAQ Response: Read the user’s question and retrieve the best possible answer from the knowledge base (without verbatim citations—paraphrase if needed).

Sentiment Classification: Evaluate the user’s question text and determine sentiment: - **Positive**: Polite, appreciative, or friendly language (e.g., “Thank you!”, “Great, how do I...?”). - **Neutral**: Straightforward inquiries without emotional cues (e.g., “How do I change my plan?”). - **Negative**: Frustrated, urgent, or angry tone (e.g., “Why isn’t this working?!”, “I’m mad that...”).

Confidence Thresholding: If the user’s message is ambiguous or doesn't clearly match a sentiment, return "Neutral" as the result.

Language Awareness: Analyze user input regardless of the language, but always return the sentiment name in English.

No User Response: Do not reply directly to the user. Your output is for internal processing only.

##Constraints - Always consult only the FAQ/knowledge base for answers; do not hallucinate or make up information. - Return exactly one sentiment label. If unsure between Neutral and Negative, err on the side of Neutral. - Do not ask follow-up questions or request clarification within this prompt—the focus is answer + sentiment only. - Store the detected sentiment in `@sentiment` for downstream routing.

##Examples User says: “I can’t log in and I’m really frustrated.” → Answer: “I’m sorry you’re having trouble. To reset your password, click ‘Forgot Password’ on the login page and follow the instructions.” Sentiment: Negative

User says: “Hi, could you show me your pricing plans?” → Answer: “Certainly! You can view our current pricing plans here: [link]. Let me know if you have any more questions.” Sentiment: Neutral

User says: “Thanks, that helped a lot—awesome service!” → Answer: “You’re welcome! Glad I could help. If there’s anything else, feel free to ask.” Sentiment: Positive

Car Recommendation and Appointment AI Agent

This AI Agent sits at the heart of a linear Web bot for a car reseller. The bot first collects essential lead information, name, email, buying intent (e.g., new vs. used), budget range, preferred style (SUV, sedan, hatchback, etc.), and fuel type (gasoline, diesel, electric). Once all fields are captured, the flow “Jumps to” the AI Agent, which ingests those variables, analyses available car deals, and returns 2–3 personalized recommendations. Finally, the agent asks if the user wants to book an appointment (test drive or consultation) and shares a calendar link.

What You’ll Achieve

Efficient Lead Capture: Collect six key data points (@name, @email, @buy_intent, @budget, @style, @fuel_type) via a straightforward, linear Landbot flow.
Personalized Vehicle Matching: Use the AI Agent to parse user preferences and recommend the top 2–3 cars that fit budget, style, and fuel type.
Seamless Appointment Scheduling: Assign an exit condition in the AI Agent setup that jumps to a lineal bot to book an appointment or test drive.
Automated Lead Logging: Push all captured data and the chosen recommendation to your CRM or Google Sheets for follow-up, segmentation, and reporting.
Higher Conversion Rates: Improve user experience by delivering tailored options quickly and reducing friction in the booking process.

Tools and Integrations Used

Landbot AI Agents (for NLU-driven recommendations and exit condition to book a meeting)
Landbot Standard Fields (@name, @email)
Landbot Custom Fields (@buy_intent, @budget, @style, @fuel_type, @recommended_car)
Landbot Bot Builder (to collect lead data in a linear Web bot and “Jump to” the AI Agent)
CRM or Google Sheets Integration (to log lead information and chosen recommendations)
Car Inventory Knowledge Base (uploaded into Landbot’s AI Agent setup)
Calendly Integration (or similar)

Prompt to be used for the AI Agent Instructions

##Role You are a friendly Car Recommendation & Appointment Agent for a dealership. You speak in a warm, conversational style—like a helpful sales consultant—while still providing structured recommendations.

##Primary Objective Read the user’s collected details—@name, @email, @buy_intent (new or used), @budget, @style, and @fuel_type—and craft a natural, engaging reply that: 1. Presents 2–3 personalized car options with key specs and highlights. 2. Asks which one they like best. 3. Offers to set up a meeting or test drive.

##Instructions 1) Greet the user by @name and thank them for sharing their preferences. Don't ask again about their preferences. Read the user’s collected details—@name, @email, @buy_intent, @budget, @style, and @fuel_type—. 2) Recommend exactly 2–3 cars in a friendly, descriptive way, using this format: “[Make] [Model] ([Year]) – [Price]. Highlight: [Feature].” —but weave it into a sentence rather than listing bullet points. 3) After describing the options, ask: “Which of these catches your eye?” 4) If the user shares their preference, then suggest: “If you’d like, I can book a meeting or test drive for you—just let me know!” 4) Output a structured response so your suggestion and booking offer can be parsed by the Landbot flow. Do not apologize or add extra commentary.

##Output Format - First, your conversational recommendation paragraph. - Then, on a new line, ask the choice question. - On the following line, offer the meeting link invitation.

##Data Logging - Ensure the top recommendation (or the one they choose) is stored in `@recommended_car`. - Do not apologize or add any unrelated commentary.

##Constraints - Always recommend exactly 2–3 cars. If fewer than 2 match, still list all available. If more than 3 match, pick those closest to the user’s budget. - Return only the recommendations and booking prompt—no small talk or additional questions. - Use “Budget” in USD (e.g., “$20,000–$30,000”) and match exactly to user’s `@budget` range. - Store the final selected car model in `@recommended_car` (e.g., “Toyota Corolla 2024”).

##Examples User details: - @buy_intent = “New” - @budget = “$20,000-$30,000” - @style = “SUV” - @fuel_type = “Electric”

How to Write AI Prompts: Prompt Engineering Tips for AI Agents

We have just seen the potential of AI agents tailored for specific use cases, and also how important it is to write a proper prompt to get the results that we expect from our AI chatbot. A detailed, well-defined, and concrete prompt will ensure our bot behaves according to what our users need, and it will prevent hallucinations.

We’ve seen a few examples of accurate and precise prompts, but now let’s dive deeper into the best practices for prompt engineering:

Be Extremely Explicit About Roles and Outputs

Always start your prompt with a Role section that states who (or what) the agent is.
Under Responsibilities, list exactly what you expect: the specific data to extract, categories to use, or how to format the answer.
If you want a single-word output (e.g., “Technical Support” or “Positive”), tell the agent: “Return only the label in plain text. Do not add any additional commentary.”
Define Constraints and Edge Cases
- If you need a fallback label (“Unclear,” “Unknown,” or “Neutral”), explicitly mention it and when to use it.
- Clarify what should happen if the user’s input doesn’t match any category (for example: “If budget is not a number, store ‘Unknown’”).

Include Concrete Examples

Show 4–6 short, realistic user inputs mapped to the ideal output. Agents learn from examples, and specifying them dramatically reduces misclassifications.
Use simple, unambiguous phrasing:

User says: “What’s your Pro plan cost?” → Pricing Question User says: “I want to enroll in Cambridge” → Cambridge Certificate

Use Clear Field Names and Consistent Naming

Match your custom field variable exactly with how you refer to it in the prompt (e.g., @budget_range, @course_interest).
Avoid overly generic variable names, be precise so that you don’t accidentally overwrite data later.

Validate Format When Needed

If you need a valid email, credit card number, or phone, prompt the agent: “If the input does not match the email format, ask ‘Can you please provide a valid email address?’”
This prevents storing junk data and improves your downstream automation.

Keep Prompts Modular

Write each agent’s prompt so it focuses on one core task only (intent detection, lead qualification, sentiment analysis). If you try to merge too many responsibilities, accuracy will drop.
If you need additional functionality (e.g., analyzing sentiment and asking follow-ups), consider chaining two agents or having one agent exit into a second flow.

Watch for Language and Tone

If your brand uses a particular tone (friendly, formal, playful), you can add a short instruction like: “When speaking to the user (in greeting or exit), use a friendly, upbeat tone.”
However, if your agent’s output is strictly “internal” (e.g., returning “Positive”/“Negative”), make it clear: “Do not generate any conversational text; only return the sentiment label.”

Test Continuously and Refine

Use the “Analyze” tab to inspect real user inputs vs. agent predictions.
If you see high “Unclear” or misclassified intents, add more examples or adjust your category definitions.
Iterate: small tweaks in wording or examples often yield large improvements in accuracy.

Conclusion

From intent detection to lead qualification, content personalization, or sentiment analysis, the possibilities and use cases for AI agents in conversational experiences are multiple and they are expanding quickly. By embedding AI into specific steps of your flow, you can gain all the intelligence and flexibility of natural language understanding without sacrificing structure, consistency, or control.

One of the key aspects of maximizing the benefits of AI agents is mastering the art of prompt engineering. A well-crafted prompt is what transforms a generic AI into a focused, high-performing agent capable of delivering real business results. Whether you’re building an AI chatbot for Web or WhatsApp, defining clear roles, setting the expectations, and mapping the right outputs will make all the difference.

With Landbot AI Agents, you can design and build smarter, modular experiences that improve automation and user satisfaction equally. The examples in this article are just the beginning, there are countless other AI agents use cases waiting to be explored. Ready to build your own? Start experimenting, refining your prompts, and turning conversations into conversions!

FAQs About AI Agents

1. What is an AI agent and how does it work?

An AI agent is an autonomous system powered by artificial intelligence that performs tasks or makes decisions based on goals, prompts, or environmental inputs. It can interact with users, systems, or data sources to complete complex workflows.

2. What are the most common use cases for AI agents?

AI agents are commonly used for customer support, lead generation, data analysis, personal assistants, content creation, and task automation in both business and consumer settings.

3. How is an AI agent different from a chatbot?

While a chatbot typically follows scripted or rule-based conversations, an AI agent can understand context, make decisions, and act autonomously, often combining natural language processing with memory and reasoning capabilities.

4. Why is prompt engineering important for AI agents?

Prompt engineering helps define clear instructions for AI agents to follow. Well-crafted prompts improve accuracy, maintain structure, reduce hallucinations, and ensure consistent, high-quality outputs aligned with business goals.

5. What tools can I use to build and deploy AI agents?

Popular tools to build AI agents include LangChain, OpenAI GPTs, AutoGPT, Microsoft Copilot Studio, and no-code platforms like Landbot, which let you design conversational AI agents for websites and WhatsApp.