Topics, keywords, and queries

BrandBanta organizes what you track into three levels. You'll use all three eventually; most workspaces start with just the top one.

Workspace
  └── Topic           ← what category your brand competes in
        └── Keyword   ← a sub-theme of the topic
              └── Query  ← an actual LLM prompt

Workspace

The top-level container. Equivalent to "your company" or "your client" (for agencies).

A workspace owns:

One primary brand (you) + N competitors
Multiple topics
One default project (we hide this — it exists for the future agency-style "multiple brands per workspace" feature)
Members, billing, API keys, settings

The URL is /<workspace-slug>-<id>/.... The slug is decorative; the ID is canonical. Renaming the workspace doesn't break old links.

Topic

A topic is the category your brand competes in. Think of it as the search-engine "vertical" you'd want to rank for.

Good topics:

"Voice AI for healthcare"
"Headless CMS for e-commerce"
"Customer support automation for B2B SaaS"

What makes a topic good:

Narrow enough that the prompts cohere — a topic of "marketing" is too broad; queries about email marketing and SEO and social ads would all pile in
Broad enough to matter commercially — a topic of "Salesforce-specific Marketing Cloud automation" is too narrow; you wouldn't see meaningful aggregate signal
Maps to how prospects search — if you wouldn't expect a real customer to Google for this category, neither will the LLM produce useful answers about it

Most workspaces have 1–3 topics. Agencies tracking multiple clients usually have one topic per client.

Keyword

A keyword is a sub-theme within a topic. Optional — many workspaces skip this layer and put queries directly under topics.

When keywords help:

You want to slice the data ("how do we do on HIPAA-compliance questions vs general healthcare questions")
You're tracking a topic with 50+ queries and need clusters
You're running a content-gap analysis where each keyword maps to a content pillar

Example topic + keywords:

Topic: Voice AI for healthcare
  ├── Keyword: HIPAA compliance
  ├── Keyword: Patient intake
  ├── Keyword: Outbound reminders
  └── Keyword: EMR integration

You can leave the keyword field blank when adding a query — it just lands under the topic directly.

Query

A query is the actual prompt sent to LLMs. This is what the customer would type into ChatGPT.

Three rules for writing good queries (see Your first topic for the full version):

Use real intent verbs — "how do I," "what's the best," "compare X to Y"
Include relevant context — "voice AI for healthcare" not just "voice AI"
Mix buyer stages — awareness, consideration, decision

A query lives forever (until you delete it) and gets re-run on every scheduled scan. So treat queries as the unit of long-term measurement — write them carefully, then leave them in place to track over time.

Relationships at a glance

A workspace has many topics
A topic has many keywords and many queries
A keyword has many queries (and is owned by exactly one topic)
A query belongs to exactly one topic, and optionally to one keyword

CRUD operations are available on every level via:

The UI — Topics page → drill in → manage keywords + queries
Bulk import — paste a CSV of queries into the topic-edit page
oRPC — tracking.topics.*, tracking.keywords.*, tracking.queries.* (BYOK users with API access)

What to delete vs what to keep

When iterating on prompts:

Delete a query if it's been running for 2+ weeks and produces wildly inconsistent responses (the question is too vague)
Edit a query if you want to narrow it without losing the historical scans (the database keeps every scan tied to the query, even after the text changes)
Mark a query as "core" instead of deleting it if it's important but underperforming — that's a signal to act, not to stop measuring

Don't be precious about pruning. The dashboards work best when every query is producing signal.