Your first topic

A topic is a clustered set of questions your category gets asked. Mention rate, share of voice, sentiment, and citations all roll up at the topic level — so the topic you pick determines what you can answer.

What a good topic looks like

A good topic is narrow enough that the prompts cohere but broad enough to matter commercially. Examples:

Rule of thumb: if a real prospect would Google your topic before they Google your product, it's the right scope.

The topic → keyword → query hierarchy

Workspace (your company)
  └── Topic (category — e.g. "Voice AI for healthcare")
        └── Keyword (sub-theme — e.g. "HIPAA compliance")
              └── Query (actual prompt — e.g. "How do I make voice AI HIPAA compliant?")

You won't always use all three layers. Most workspaces start with one topic and 10–20 queries directly under it. Add keywords later when you want to slice the data ("how do we do on HIPAA-specific questions vs general healthcare questions").

Adding a topic

In your workspace sidebar, go to Topics → New topic.

1. Name — the category as you'd describe it to a new employee
2. Description (optional) — one sentence on what this topic represents
3. AI-suggested queries — click the "Suggest queries" button. We'll generate 10–15 starter prompts based on the topic name and your brand context

The AI suggester pulls from common search patterns for your category. It's a starting point — edit aggressively. A starter prompt like "What's the best voice AI platform?" should usually become "What's the best voice AI platform for patient intake?" — narrower, more like what real prospects ask.

Writing prompts that produce signal

The job of a prompt is to resemble a real customer question. Three rules:

1. Use real intent verbs. Not "tell me about voice AI" — "how do I choose a voice AI vendor for my hospital?" Customers ask "how," "what's the best," "compare X to Y," "is X right for my use case." Generic descriptions don't get the same kind of answer.

2. Include the relevant context. "Voice AI" alone gets you generic answers. "Voice AI for outbound sales in financial services" gets you a list with vendor-specific recommendations.

3. Keep a mix of buyer-stage prompts. Awareness ("what is voice AI"), consideration ("voice AI vs IVR"), decision ("Get Vocal vs Bland vs Vapi"). Different stages surface your brand in different ways.

How many prompts to start with

10 is the sweet spot. Enough to see meaningful mention-rate signal; few enough that the free tier or a $5 OpenRouter top-up covers a month of scans.

Once you start running scans:

• Add prompts that surface gaps you didn't expect
• Remove prompts that produce noise (e.g., LLM keeps changing answer because the question is too vague)
• Mark high-value prompts as "core" so the dashboards highlight them separately

After your first topic

• Run your first scan — single-query scan to see what the data looks like
• Bulk scans — run every query in a topic in one click
• Scheduled scans — daily / weekly / monthly automation
• How insights work — what the AI narrative is actually doing