Airgentic Help
Categories & Themes on the Insights screen helps you understand what your customers are asking about — organised into stable themes you can track over time and governed categories that match your organisation's structure.
While the other Insights tabs show individual conversations and point-in-time metrics, Categories & Themes reveals patterns: which topics are growing, which are declining, and what new questions are emerging.
Categories & Themes uses AI to analyse the questions users ask your service. It provides two complementary approaches:
Themes — Discovered automatically by clustering similar questions together based on their meaning. Themes are generated bottom-up from the data itself, so they surface topics you might not have anticipated.
Categories — Defined by you through taxonomies that reflect your organisation's structure. Questions are automatically classified against your categories using semantic matching.
Both approaches are updated nightly as new conversations come in, so emerging topics are surfaced within a day of appearing.
At the lowest level, the system captures the standalone question from each user interaction — a clear, self-contained version of what the user was asking. Questions that are semantically similar (asking about the same thing in different words) are grouped together.
Clusters are groups of very similar questions discovered during a monthly analysis period. The system uses AI to identify questions that share the same intent and groups them together.
For example, these questions would likely end up in the same cluster:
Clusters are refreshed each calendar month. The current month's data is marked as preliminary since new conversations are still being added.
Themes are stable, persistent categories that exist across time. While clusters are recalculated monthly, themes provide continuity for tracking and reporting.
Examples of themes might include:
Each theme has an AI-generated label and description that captures the common intent behind the questions it contains. When new clusters are discovered, the system automatically maps them to existing themes using embedding similarity — or creates new themes if the cluster represents something genuinely new.
Themes are designed to stay consistent so you can track trends month over month.
In addition to automatic theme discovery, you can define your own taxonomy — a structured set of categories that match your organisation's terminology and ownership structure.
For example, a taxonomy for a university might include:
Or a taxonomy for a customer service team might include:
Each node in the taxonomy can have an owner (the team or person responsible for that topic) and a description that helps the AI understand what belongs there.
You can create multiple taxonomies for the same data. For instance, one taxonomy might organise by department while another organises by enquiry type — giving you different views of the same conversations.
Questions are automatically classified against your published taxonomies every night. This gives you a governed, consistent view that aligns with your business structure.
Understanding how clustering relates to date filters is important for interpreting your data correctly.
Topic clusters are generated on a calendar month basis. Each month, the system analyses all questions from that month and groups them into clusters. These clusters are then mapped to themes.
Within each month, clusters are updated nightly as new conversations arrive, but the clustering period always aligns with the calendar month boundary.
When you apply a date range filter (e.g., "15 March to 15 April"), the theme and cluster data you see is based on the clustering runs that overlap your selected dates. This means:
For example, if you filter to "15 March – 15 April":
- You'll see themes that appeared in March's clustering AND/OR April's clustering
- Question counts will be filtered to your exact date range
- But a theme's existence depends on whether it was discovered in either monthly run
This design ensures themes remain stable and comparable across time while still allowing you to filter and slice the data.
The Categories & Themes tab shows your themes and categories in three connected views, accessed via the Categories, Themes, and Overlap buttons:
The Categories view displays your taxonomy categories as horizontal bar charts:
Each category bar shows:
- The category name with indicator badges (⚠ needs attention, 😟 strong negative sentiment)
- Question count and percentage of total
- Trend indicator if historical data is available
Hover over any bar to see detailed metrics including answer rate and sentiment breakdown.
Click any bar to open a drill panel showing the questions assigned to that category. From there, you can click View Conversations to navigate to the Conversations tab filtered to show only conversations in that category.
The summary strip at the top shows:
- Total questions analysed
- Number of categories tracked
- Categories needing attention (low answer rate)
- Categories with strong negative sentiment
Use the filter buttons (Spikes, Drops, Need Attention, Strong Negative) to highlight categories matching specific criteria.
The Themes view displays automatically discovered themes:
Each theme bar shows:
- The theme name with indicator badges
- Unique question count and percentage
- 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
Expand any cluster to see its questions, and click View Conversations to navigate to the Conversations tab filtered to that theme.
The Emerging Demand section highlights new topics that aren't yet mapped to established themes — potential new issues or trends worth investigating.
The Overlap view shows where themes and taxonomy categories intersect. This answers questions like:
The matrix displays categories as rows and themes as columns, with cell values showing the number of questions that belong to both.
Click any cell to see the questions matching that category-theme combination. From there, click View Conversations to navigate to the Conversations tab with both filters applied.
Display options let you:
- Toggle between question counts and percentages
- Sort rows/columns alphabetically or by total
- Hide empty rows or columns
Throughout Categories & Themes, clicking on items lets you drill through to see the underlying conversations:
This lets you seamlessly move from aggregate patterns to individual conversations and back.
Themes move through a lifecycle as your service evolves:
| Status | Meaning |
|---|---|
| Active | This theme has been seen consistently and represents an established topic. |
| Emerging | This theme is new or growing — it has appeared recently or is seeing increased activity. Watch these closely. |
| Declining | This theme has seen reduced activity recently. Users may be asking about this topic less often. |
| Inactive | This theme hasn't been seen in several analysis periods and may no longer be relevant. |
The system automatically updates theme statuses based on usage trends.
New themes can appear in two ways:
Seeding — When you run your first analysis, the system creates initial themes based on the clusters it discovers. These become your baseline.
Emergence — In subsequent runs, if a cluster is discovered that doesn't match any existing theme well, the system creates a new theme for it. These start with "Emerging" status.
Emerging themes are worth watching closely — they can indicate new issues, changing customer needs, or gaps in your content.
Categories & Themes updates automatically every night:
The current month's data is marked as preliminary since new conversations are still being added. Previous months show final, stable results.
You can also trigger a manual analysis from the Topics and Themes admin screen using the +New Analysis button.
All views in Categories & Themes respect the global filters at the top of the Insights screen:
When you apply filters, the theme counts, category counts, and overlap matrix all update to reflect that filtered slice of data.
A loading indicator appears when filters are being applied to help you know when data is being refreshed.
The summary strip shows key metrics at a glance:
Use the filter buttons to focus on specific conditions:
| Button | Shows |
|---|---|
| All | All items (default) |
| Spikes | Items with unusual increases in volume |
| Drops | Items with unusual decreases in volume |
| Need Attention | Items with low answer completeness |
| Strong Negative | Items with high negative sentiment |
Hover over any filter button to see a tooltip explaining the criteria and how many items match.
Hover over any bar to see detailed metrics:
If a theme or category shows high volume but low answer completeness (visible via the ⚠ badge or Need Attention filter), you may need to add or improve content on that topic.
After updating prompts, adding content, or making configuration changes, use the trend indicators and evolution charts to see whether the affected topics show improved outcomes.
Watch for new "Emerging" themes that suddenly appear with high volume — these often indicate a new product issue, a marketing campaign driving questions, or a gap in your knowledge base.
Use taxonomy categories to report on customer demand by department, product line, or any other organisational dimension. The governed structure ensures consistent categorisation over time.
Use the Overlap view to see which discovered themes span multiple business units. This can reveal topics that need cross-functional ownership or indicate where your taxonomy categories may need refinement.
Click the Export button on any chart to download the data as CSV for further analysis or sharing with stakeholders.