Semantic Themes on the Insights Demand tab shows what people asked, organised into topics the system discovers by grouping similar questions.

Themes are generated from the conversations themselves, so they surface subjects you might not have put in a taxonomy. For reporting against your organisation’s own structure, use Taxonomy Categories instead.

Opening the view

1. Open Insights.
2. Open the Demand tab.
3. Under View by, click Semantic Themes.

The date range and other filters at the top of Insights still apply.

You need a completed theme analysis for the period you are looking at. If clustering has not been run yet, run it from Clustering & Taxonomies.

The same tab also has Taxonomy Categories and, when visitor profiles are enabled, Voice of Customer.

How themes work

Themes are built in three layers, from individual questions up to stable topics you can track over time.

Questions

The system captures a standalone question from each user interaction — a clear, self-contained version of what they were asking. Questions that mean the same thing in different words are grouped together.

Semantic clusters

Clusters are groups of very similar questions discovered during a calendar month. The system identifies questions that share the same intent.

For example, these would likely land in the same cluster:

• "How do I reset my password?"
• "I forgot my password, how do I get a new one?"
• "Password reset not working"

Clusters are refreshed each calendar month. The current month is marked preliminary because new conversations are still arriving.

Semantic themes

Themes are stable topics that persist across months. Clusters are recalculated monthly; themes give you continuity for tracking and reporting.

Examples:

• Connectivity Issues
• Password Reset
• Returns & Refunds
• Product Setup

Each theme has an AI-generated label and description. When a new cluster appears, it is mapped to an existing theme by similarity — or a new theme is created if the cluster is genuinely new.

Themes are meant to stay consistent so you can compare month over month.

How clustering works with date ranges

Topic clusters are generated on a calendar month basis. Each month, the system analyses that month’s questions, groups them into clusters, and maps those clusters to themes.

• January’s clusters contain questions asked in January
• February’s clusters contain questions asked in February

Within a month, clusters update nightly, but the period always aligns with the calendar month.

When you apply a date range (for example 15 March to 15 April), the themes you see come from the monthly clustering runs that overlap those dates:

• If the range spans two months, you see themes from both months’ runs
• Question counts are filtered to your exact dates
• Whether a theme exists in the view depends on whether it was discovered in either monthly run

Current month is preliminary — clusters are still being refined.

Taxonomy Categories are different: those assignments are per-question, so category filters follow your date range exactly. Theme membership is decided monthly.

The Semantic Themes view

The view shows automatically discovered themes:

• Theme Demand — A horizontal bar chart of question volume per theme
• Theme Evolution — A bubble chart of how theme volumes change across monthly periods

Each theme bar shows:

• The theme name with indicator badges
• Unique question count and percentage
• A trend indicator

Click any bar or bubble to open a drill modal showing:

• The clusters that make up this theme
• The questions within each cluster
• Confidence scores for cluster-to-theme mapping

From the modal you can:

• Click View Categories to switch to Taxonomy Categories, filtered to categories that have questions in this theme. The Theme dropdown is set automatically. If you have several published taxonomies, you will see overlapping categories across each of them.
• Expand a cluster, then click View Conversations to open the Conversations tab filtered to that theme.

The Emerging Demand section highlights new topics that are not yet mapped to established themes — possible new issues or trends.

The summary strip at the top shows questions analysed, themes identified, items needing attention, and items with strong negative sentiment. Use Spikes, Drops, Need Attention, and Strong Negative to focus the list.

Understanding theme status

Themes move through a lifecycle:

Status updates automatically from usage trends.

How new themes appear

1. Seeding — The first analysis creates initial themes from the clusters it finds. That is your baseline.
2. Emergence — Later runs create a new theme when a cluster does not match any existing theme well. Those start as Emerging.

Emerging themes often mean a new issue, a campaign driving questions, or a gap in your knowledge base.

Drilling through

Clicking a bar or bubble opens the detail modal. From there:

• View Conversations — Jump to the Conversations tab with the theme filter pre-applied. A banner shows the active filter and a Back to Demand button.
• View Categories — Switch to Taxonomy Categories filtered to categories that overlap this theme. A banner on the categories view shows the active theme filter and a Clear button.

The Theme dropdown at the top of Insights is set so you can always see what you are looking at. The filter clears when you switch views again, so you do not get stuck.

Use View Categories when you want to ask "which governed categories does this theme touch?" without leaving the page.

Nightly updates

Clustering runs automatically every night at 03:30 UTC — grouping the current month’s questions into clusters and mapping them to themes.

The current month is preliminary. Previous months are final.

You can also trigger a manual analysis from the Topics and Themes admin screen with + New Analysis.

Monthly insights digest (email)

On Conversational Intelligence, the Insights Subscriptions tab can email a monthly digest of the strongest signals — themes with large answer gaps, unusual volume spikes or drops, new or lost themes, and emerging clusters.

• The digest covers the previous calendar month (stable data, not the preliminary current month).
• Delivery defaults to the 1st of the month at 09:05 Australia/Sydney (you can change day and hour in the same tab).
• Preview shows an HTML sample without sending; Send test email sends a one-off copy to one address.

The email links back into Insights with the same date slice. The dashboard remains the place to explore full detail.

Using filters

This view respects the global filters at the top of Insights:

• Date range — Focus on a period (see how this interacts with monthly clustering above)
• Taxonomy / Category / Theme — Restrict to a theme or to overlapping categories
• Sentiment — Positive, negative, or neutral conversations
• Geography — Country, region, or city
• Channel — Website or entry point
• Agent — Responding agent

A loading indicator appears while the charts refresh.

Metrics and indicators

Quick stats

• Questions analysed — Total questions in the current view
• Themes identified — Number of active themes
• Need attention — Items with answer rate below 85%
• Strong negative sentiment — Items with negative sentiment above 5%

Filter buttons

Hover a filter button for the criteria and how many items match.

Hover tooltips

Hover a bar for unique question count and percentage, occurrence count (if different), answer rate, negative sentiment, and any attention or sentiment warnings.

Practical applications

Identify content gaps

A theme with high volume and low answer completeness (⚠ badge or Need Attention) often means missing content, a curated answer, or a prompt that does not cover the topic.

Spot emerging issues

Watch Emerging themes that appear with high volume — a new product issue, a marketing campaign, or a knowledge-base gap.

Track the impact of changes

After you update prompts or content, use trend indicators and the evolution chart to see whether that theme’s outcomes improved.

Export data

Click Export on a chart to download CSV for stakeholders who do not have admin access.

Tips

• Check preliminary data — The current month will still move as more conversations come in.
• Understand monthly boundaries — Date range filters for themes reflect which monthly clustering runs overlap your dates. Matching the picker to a full calendar month makes comparison easier.
• Start with high-volume themes to see what is driving the most questions.
• Review emerging themes — They may be genuine new topics worth investigating.
• Drill through to conversations before you brief a content owner.
• Pivot to categories with View Categories when you need to see which teams own the overlapping topics.
• Cross-reference sentiment with the Strong Negative filter.

Related documentation

• Taxonomy Categories — Governed categories for the same conversations
• Voice of Customer — How to interpret the audience report on the same tab
• Clustering & Taxonomies — How to run clustering and maintain taxonomies
• Visitor Profiles — How profile fields are collected in chat

← Back to Insights overview