UserVoice AI Agent

UserVoice is developing an AI Agent that lets you ask natural-language questions about your data — ideas, accounts, users, and themes — and get answers grounded in your actual UserVoice data.

Instead of manually filtering idea lists or exporting to a spreadsheet, you can ask:

• "What are our top 10 requests from enterprise accounts by revenue?"
• "What is Acme Corp asking for that we haven't responded to?"
• "Which themes have the most supporter activity this quarter?"
• "What has Sarah at Acme been asking for?"

The AI Agent reads your data in real time, reasons over it, and responds with prose summaries and clickable idea tables — all in a persistent, multi-turn chat interface.

What you can access

What it can do — and the limits

You don't need to know how the Agent works under the hood — just ask in plain language and it figures out which data to pull. This section describes what's possible and the limits that apply, so you know what to expect.

Searching your feedback (ideas)

Ask for lists of ideas filtered by status, category, forum, label, account, keyword, or date range, sorted by votes, revenue, recency, activity, comment volume, or what's trending.

• "Top 20 open ideas by votes in the Integrations category."
• "Unanswered ideas from enterprise accounts, sorted by MRR."
• "Ideas marked Planned in the last 30 days."

Limits: Returns up to 10 ideas by default; ask for more, up to 100 in one answer. Searches open ideas by default (published, approved, and under-review) — ask explicitly to include closed or merged ideas.

Reading an idea in depth

Ask a "why" question and the Agent reads an idea's full description, customer comments, and internal admin notes to surface the real painpoints — not just the title.

• "What are customers saying about the top API idea?"
• "Why do customers want this feature?"

Limits: Reads the 10 most recent comments and 5 most recent internal notes; the description is shown up to roughly 1,500 characters.

Finding accounts

Filter your customer accounts by revenue, lifecycle status, engagement, or CRM data, and sort by MRR, LTV, recency, last-active date, or supported-idea count.

• "Top 10 active accounts by MRR."
• "Show me churned accounts that wanted Salesforce integration."

Limits: Returns up to 50 accounts; defaults to active accounts (ask for churned, prospect, or lost). Revenue fields (MRR/LTV) require the revenue permission; health signals require a Gainsight integration.

Finding people (users)

Look up a person by name, email, linked account, or the ideas they've interacted with — useful for resolving "who is Sarah?" before drilling in.

• "Who is John Smith?"
• "Which users at Acme have voted the most?"

Limits: Returns up to 50 users. By default it includes anyone who engaged — voted or commented; ask for contributors to include idea creators who haven't voted or commented. A deep-dive on one person shows up to 10 ideas they created and 10 they support (open ideas).

Working with labels

Reference a label by name in your question and the Agent resolves it automatically — matching is case-insensitive against the label name.

Detecting themes

Ask an analytical question and the Agent clusters the relevant ideas into pain-based themes with impact and a representative quote. See How the Agent detects themes below.

Idea tables

When your question calls for a list of ideas, the Agent renders an interactive, clickable table inline. See Understanding what you see below.

How the Agent detects themes

When you ask an analytical question, the AI Agent reads the actual ideas in scope — their titles, bodies, and customer comments — and analyzes them on the spot to find recurring painpoints. It's not pulling from a pre-built report; the analysis is generated fresh each time you ask, usually in about 5–10 seconds.

The work is split in two:

• The Agent clusters and names. It groups related ideas into themes, writes a specific pain statement for each, judges its intensity, and pulls a real, verbatim quote from a customer comment as evidence.
• The Agent does the math. Every number you see — supporters, accounts, revenue/MRR, dates, lifecycle — is calculated directly from your database, not estimated by the AI.

Two things worth knowing: a theme only forms when at least two genuinely related ideas support it, and ideas that don't fit a clear pattern are intentionally left uncategorized rather than forced into a theme. Because the analysis is generated live, results can vary slightly between runs.

How these differ from your Themes & Trends report

The AI Agent's themes are generated live, scoped to the specific question you just asked, and shaped to drive a decision — so instead of a category label you get a pointed pain statement, enriched with impact (supporters, accounts, revenue, lifecycle) and a representative quote.

Differences are expected: the Agent analyzes the slice of ideas relevant to your question at the moment you ask, using its own method and a more specific naming style, rather than reading back your saved report. The Agent's themes are also ephemeral — they aren't saved and won't appear in Themes & Trends, and your T&T reports won't change because of them. Your Themes & Trends data stays exactly as it is.

Understanding what you see

Streaming responses

The AI Agent shows its reasoning in real time. Short narration appears between steps — e.g., "Let me check the Onboarding category..." — followed by the answer. This lets you catch cases where the Agent interpreted your question differently than you intended.

Tool call details

A collapsible block shows what the Agent looked up and what filters it applied. Expand it to verify exactly what data was used. This is intentionally visible and helpful for checking the Agent's work.

Inline idea tables

When your question calls for a list of ideas, the AI Agent renders a clickable table inline. Columns are chosen to match your question — a revenue query shows MRR; a staleness question shows last activity date. Each row links directly to the idea in UserVoice.

Tables always show live data. The Agent saves which ideas were in the table and re-pulls their current details (votes, status, owner, etc.) each time you reopen a conversation. The written analysis, by contrast, is saved exactly as the Agent wrote it. So the table reflects your data as it is now, while the commentary reflects it as it was then — treat the table as the source of truth for current numbers.

Save as List: The table includes a Save as List button (labeled Save N as List, where N is the number of selected rows). Each row has a checkbox and all rows are selected by default — so you can save the whole table in one click, or uncheck rows to save just a subset. Clicking the button opens a small popover that prompts you to name the list; on save it creates a new list of the selected ideas and shows a "Saved as '…'" confirmation inline — it does not navigate you away. The button is hidden entirely for read-only admins (it requires feedback write access). The list is saved to Idea Lists.

What the AI Agent cannot do

The AI Agent currently can't perform write actions through conversational requests — but many actions are available as buttons on idea tables and within the idea sheet. If you ask the Agent to do something (change a status, send a message), it tells you directly and stops — it won't invent a workaround or tell you how to do it in the UI (it doesn't know your specific configuration).

The Agent only answers questions about your UserVoice data. It won't function as a general-purpose AI assistant.

Tips for better results

Be specific. "Show me top ideas by MRR" returns better results than "what should I work on?" Give the Agent a direction to search in.

Use your own terminology. The Agent knows your category names, forum names, label names, and the contents of your user feedback. Say "ideas in Onboarding" or "the Mobile theme" — it resolves them automatically.

Ask follow-ups. You don't need to restate context. After asking about Acme Corp, you can follow up with "what about ideas with no response?" and the Agent stays on topic.

Respond if the Agent asks a clarifying question. This happens when a term matches multiple things (e.g., a category and a label both named "Integration"). One short answer and it proceeds.

For "why" questions, start with a search. Try "Show me the top 5 API ideas" first, then "What are customers saying in the comments on the top one?" The Agent reads comments to surface the actual pain points, not just the title.

Revenue data requires permission. If MRR/ARR figures are missing, your admin account may not have the revenue permission enabled.

Conversations and history

Conversations are saved automatically. From the left sidebar you can resume any past thread, rename conversations, or delete them — and when you resume a conversation, the Agent retains the context from that thread.

Conversations are private to you; there is no conversation sharing between admins. There is no conversation limit during the early-access period, though we may bulk-remove conversations at certain points as we make larger changes.

Access, data & privacy

The AI Agent is part of the UserVoice AI package and is in early access. For questions about availability and invites, which AI model powers the Agent, data handling, and tenant isolation, see the UserVoice AI Agent FAQ.

For common questions — access, model, privacy, and quick "can it do X?" answers — see the UserVoice AI Agent FAQ.