Themes Editor - AI Suggestions

Thematic's AI Suggestions feature uses AI to analyze your theme structure and your untagged feedback, then surfaces actionable improvements directly inside the Themes Editor. Instead of manually scanning hundreds of themes looking for problems, you get a prioritized list of suggestions you can apply or dismiss with one click.

How to access AI Suggestions

1. Open the Themes Editor from your dataset's Manage Themes area.
2. The editor now uses a split layout: your theme list on the left, and the AI Suggestions panel on the right.
3. Suggestions stream in as they're generated -- they may not all appear at once.

If you've clicked into a specific theme (which switches the right panel to theme details), you can click the button to return to the suggestions panel.

What the AI looks for

Suggestions fall into two tabs: Discover (finding new themes and phrases) and Improve (fixing structural issues). Discover is the default tab.

Discover suggestions

These help you increase your tagging coverage by finding patterns in comments that aren't currently tagged.

• New Phrases -- The AI finds untagged comments that match an existing sub-theme but aren't being captured. For example, 50 comments mentioning "bill pay" that aren't tagged under your "Bill Payment" theme. Accepting adds the new phrases so those comments get tagged.
• New Themes -- The AI identifies clusters of untagged comments that represent a topic you don't have a theme for yet. For example, 30 comments about "two-factor authentication" with no matching theme. Accepting creates a new sub-theme with the discovered phrases.

Improve suggestions

These help you fix structural quality issues in your existing theme hierarchy.

• Organization (Misplaced themes) -- A sub-theme sitting under the wrong base theme. For example, "Mobile App Crashes" filed under "Billing" instead of "App Experience". You can move it to the right base, create a new base, or delete it.
• Merges (Overlapping themes) -- Two sub-themes that are semantically similar and should be combined. For example, "Bill Payments" and "Payment of Bills". Accepting merges them into one.
• Mappings (Mismerged themes) -- A merged theme assigned to the wrong sub-theme. For example, the theme "App crash" merged into "App Features" instead of "App Bugs". You can move, delete, or split it into a new sub-theme.
• Generic themes -- Themes too vague to be useful in analysis, such as "General Issues" or "Other Feedback". Accepting removes them.

How to act on suggestions

Each suggestion card shows its category, a short description of the recommended action, and two buttons:

• Apply -- Adds the change to your current themes draft. This does not immediately retag your data. You still need to use the standard Apply flow to commit your changes.
• Dismiss -- Removes the suggestion from the list. The AI remembers your dismissals and avoids re-suggesting the same thing.

Getting more suggestions

When you've worked through all visible suggestions, click Show More Suggestions to load a new batch. If no more suggestions are available, you'll see a message confirming that.

You can also click the Regenerate button in the panel header to clear your dismissed suggestions and get a completely fresh set.

Guided suggestions

Use the freeform input at the top of the panel to focus the AI on a specific area. For example, typing "focus on delivery issues" will regenerate suggestions scoped to that topic.

How it works behind the scenes

The AI runs 18 evaluation checks across your theme hierarchy and your untagged comments. It analyzes themes for overlaps, misplacements, and quality issues at multiple levels of your hierarchy (across the full tree, within a base theme, and within a sub-theme).

For discovering new themes and phrases, the system samples from your untagged comments using relevance scoring to surface the most pertinent feedback. It also estimates the percentage of comments that would be affected by each suggestion, so you can prioritize high-impact changes.

The system keeps a memory of suggestions you've dismissed, so repeated evaluations won't resurface the same recommendations.

Tips for getting the most from AI Suggestions

• Start with Discover -- New phrases and new themes directly increase your tagging coverage, which improves every downstream analysis.
• Sort by impact -- Suggestions that affect a higher percentage of comments will improve your data quality the most.
• Don't over-accept -- Review each suggestion before applying. The AI doesn't have your business context, so a suggestion that looks reasonable may not fit your analytical goals.
• Use guided mode -- If you know a specific area needs work (e.g., you've noticed gaps in "delivery" feedback), type it into the prompt to get targeted suggestions.
• Re-run after major edits -- After applying a batch of changes and retagging, open the suggestions panel again. The AI will evaluate your updated hierarchy and may find new opportunities.

FAQs

Do I need special permissions? You need theme editing permissions for your dataset. If you don't see the Manage Themes option, check with your organization's administrator.

Will applying a suggestion immediately retag my data? No. Applying a suggestion adds the change to your draft. You still need to click Apply (the standard themes apply flow) to commit changes and trigger retagging.

Can I undo a suggestion I applied? Yes, as long as you haven't committed the draft yet. You can discard the draft to revert. If you've already applied, Thematic keeps version history so you can roll back.

Why do I see different suggestions each time? The AI samples from your feedback data each time it runs. Because sampling is randomized for diversity, different runs may surface different patterns. This is by design -- it ensures you discover a broad range of issues over multiple sessions.

Is AI Suggestions available for all datasets? AI Suggestions is available for datasets with a theme structure. If your dataset hasn't had initial themes generated yet, set those up first.