Conversation Starters That Convert

Conversation Starters That Convert

Conversation starters are the 4 short prompts that appear in your agent's chat widget when a visitor first arrives. They're the first impression — the moment a visitor decides whether to engage or bounce.

The Rules

Every agent gets exactly 4 starters, and each one must follow these rules:

• Maximum 30 characters — short enough to scan in a glance
• Dash-prefixed format — starts with a dash for visual consistency
• Grounded in your knowledge — based on what your agent actually knows, not generic prompts
• Soul-only — these come from your soul, not generated by AI on the fly

Starters appear when the chat opens and clear on the first message. They don't come back on reload or new sessions. They're a one-time hook.

Good vs Bad Starters

Bad starters (generic, could be any chatbot):

• "Ask me anything"
• "How can I help you?"
• "Tell me about yourself"
• "What do you want to know?"

These are useless. They give the visitor no signal about what your agent is good at. They could appear on literally any chatbot in the world.

Good starters (specific, grounded in knowledge):

For a fact-checker (check.agenturo.app):

• "- Check a headline for me"
• "- Is this claim true?"
• "- Verify a viral quote"
• "- Fact-check a statistic"

These tell the visitor exactly what this agent does. Each starter is a specific action the agent is good at.

For a personal brand agent (afonie.agenturo.app):

• "- What's your tech stack?"
• "- Are you open to roles?"
• "- Tell me about your work"
• "- Analyze a job for fit"

These cover the main visitor intents: casual curiosity, recruiter screening, deep-dive, and the special fit-analysis mode.

For a product agent (zoiberg.agenturo.app):

• "- What coffee do you have?"
• "- Help me choose a blend"
• "- How do I brew pour-over?"
• "- Do you ship to my country?"

These map to real customer questions. Each one triggers a different knowledge path in the agent.

For the platform agent (agent.agenturo.app):

• "- What is Agenturo?"
• "- How do I create an agent?"
• "- Show me agent examples"
• "- What makes a great soul?"

These guide new visitors through discovery in a logical sequence.

How to Design Your 4 Starters

Step 1: List the top 10 things people would ask your agent.

Step 2: Group them into 4 categories. Usually these are:

1. Casual intro — "Who are you" / "What do you do"
2. Core value — the main thing your agent is useful for
3. Deep capability — a specific feature or mode
4. Edge case — something surprising or unique your agent can do

Step 3: Write each starter in under 30 characters. Action-oriented language works best: "Check a claim" > "About fact-checking."

Step 4: Test each starter against your agent. Does the response showcase what your agent does best? If the starter produces a generic response, rewrite it to trigger a more specific knowledge path.

Starter Patterns That Work

Action-oriented: "Check a headline" / "Analyze a job" / "Choose a blend"

• Starts with a verb
• Implies the agent will DO something
• Best for tool-using and expert agents

Question-oriented: "What's your stack?" / "Do you ship here?"

• Natural conversation opener
• Best for personal brand and product agents

Challenge-oriented: "Is this claim true?" / "Verify a quote"

• Implies the visitor brings something TO the agent
• Best for expert and analysis agents

Common Starter Mistakes

1. Too long — "Can you help me understand your background?" — over 30 chars, reads like a form field
2. Too generic — "Ask me anything" — gives zero signal about what the agent is good at
3. Not grounded — a starter that triggers knowledge your agent doesn't have
4. All the same type — four variations of "tell me about X" — no variety in intent depth
5. Marketing language — "Discover our premium offerings!" — sounds like an ad, not a conversation

The Starter Test

For each of your 4 starters, ask:

1. Is it under 30 characters? Count them.
2. Could this appear on any chatbot? If yes, rewrite.
3. Does it trigger specific knowledge? Send it to your agent and check.
4. Do the 4 together cover different depths? Casual, core, deep, edge.