Ask AI About Your Survey Results

Stop scrolling through charts looking for the answer. Open the Ask AI panel from any survey's analytics dashboard and ask questions in plain English, and get back grounded answers with clickable citations that jump you straight to the cited question or theme.

Important: AI Can Make Mistakes

Ask AI uses a large language model to generate its answers. We work hard to keep responses accurate (every claim is grounded in your actual survey data, every cited question or theme is verified before it appears, and numbers in the response are checked against the underlying stats), but AI-generated answers can still contain mistakes. Always verify important numbers and conclusions against the raw data on your analytics dashboard before sharing them with stakeholders or making decisions based on them. Use Ask AI as a starting point and a productivity boost, not as a substitute for reviewing the data yourself.

Where to Find It

When you open a survey's analytics dashboard, look for the floating "Ask AI" button in the bottom-right corner. Click it to slide open the right-rail chat panel. The panel stays open as you switch between Overview, Questions, Responses, and Quality tabs, so keep asking while you explore.

Plans That Include Ask AI

• FREE: Not included — upgrade to Starter to unlock Ask AI.
• STARTER, PRO, BUSINESS: Included. Each question draws on your organization's token balance; there is no separate daily question cap.
• ENTERPRISE / White-label: Partners can enable Ask AI per provisioned organization from their partner dashboard.

How It Works

Every question you ask is grounded in two places: pre-computed statistics from your survey responses (counts, percentages, NPS, completion rates) and, when available, the themes, segments, and executive summary from your full AI analysis. The model is instructed to cite every quantitative claim using inline pills you can click to jump to the source:

• `[Q:question_id]` renders as a citation pill that takes you to the matching question card in the Questions tab.
• `[Theme:theme_id]` takes you to the matching theme in the Overview tab.

If the model invents a citation that doesn't exist in your data, we drop it silently before you see it. If it invents a *number* that doesn't appear in the grounding, we flag it as a warning.

Multiple Conversations

Each conversation is one thread. To start a fresh line of inquiry, click New in the panel header. Browse, switch between, or delete past conversations from the Conversations dropdown next to the title.

Suggested Starter Questions

When you open a new conversation, you'll see suggested prompts to try:

• *"What are the top themes in negative responses?"*
• *"Which segment is most satisfied?"*
• *"What's driving NPS down?"*
• *"Summarize the key takeaways in three bullet points."*

Click one to send it immediately, or type your own.

What Ask AI Sees in Your Survey

Every question you send is answered using a fresh, structured snapshot of your survey built on-the-fly:

• Per-question distributions: option percentages on multiple-choice and checkbox questions, average/median/distribution on rating questions, NPS scores with promoter/passive/detractor splits, word frequencies on text questions, and the equivalent for matrix, ranking, number, slider, and date questions.
• Top-line metrics: total responses, completion rates, and (if your responses have sentiment scores attached) an aggregated sentiment signal.
• Recent text samples: up to 25 of the most recent text responses, always PII-redacted (emails, phones, credit-card numbers, US SSNs).
• Cross-question themes, segments, and an executive summary: *only after you've run a full AI analysis*.

Even without a full AI analysis, Ask AI can answer the vast majority of questions about your survey accurately, because distributions and counts come straight from your real responses.

When to Run a Full AI Analysis

You'll see a banner in the Ask panel when no full AI analysis exists for your survey. Running one adds three things to the grounding:

1. Cross-question themes: clusters of feedback that span multiple questions.
2. Respondent segments: auto-discovered groups (e.g. "vocal detractors", "satisfied power users").
3. Executive summary: a high-level narrative.

To run one:

1. Open your survey's analytics dashboard
2. Click Run AI Analysis in the Insights area
3. Wait for the analysis job to complete (you'll see progress in real time)
4. Future Ask AI answers will automatically include the deeper grounding

Privacy and PII

Raw response text is always redacted for emails, phone numbers, credit card numbers, and US social-security numbers before it reaches the model. By default, raw response text is excluded from the grounding entirely: the model only sees pre-computed aggregates and (if a full AI analysis exists) the theme-level summaries.

If your organization wants the model to see (already-redacted) response samples, an organization admin can opt in via API.

Token Cost

Questions start at 3 tokens. That covers 4,000 tokens of grounding context; beyond that, each additional 1,000 tokens of context adds 1 token. Small and mid-size surveys usually ground under the baseline and cost exactly 3, while a large survey with a full AI analysis attached can reach 5-7. There is a hard ceiling on the amount of context a single question may use, so no question can ever cost more than 11 tokens no matter how large the survey. Before your first question the composer shows "From 3 tokens per question"; after that it quotes your last answer's actual cost, which is the best predictor of the next one. Every answer also shows exactly what it cost, and hovering that figure breaks it into base plus context.

If you run out of tokens mid-conversation, you'll see a friendly top-up prompt; past questions and answers in your conversation history are not lost.

Rate Limits

There is no daily cap on questions — your token balance is the real limit, so you can spend it as fast as you like.

• Per user: up to 30 questions per minute. This is a burst guard against runaway scripts, not a usage limit; normal analysis never approaches it.

If you hit it, you'll see exactly when you can ask again.

Tips for Better Answers

• Be specific: "What did detractors say about pricing?" beats "What's wrong?"
• Reference what you see: "Why is Q3's score lower than Q4's?" The model can reach into per-question stats.
• Ask for citations: "And which questions back that up?" prompts the model to cite specific questions you can click.
• Iterate: Each turn keeps context. Follow up with "drill into that" or "give me an example quote" to go deeper.

What Ask AI Does Not Do (Today)

• It does not modify your survey or responses.
• It does not compare across multiple surveys in v1 (single-survey scope only).
• It does not stream answers token-by-token; you'll see the full reply once it's ready (usually under 10 seconds).
• It does not currently surface as a tool inside surveys filled out by respondents; it's for survey owners and analysts only.

Troubleshooting

"Analytics Ask is not enabled for this partner."

If your organization was provisioned by a partner (white-label customer), the partner must explicitly enable Ask AI for your account. Contact your account manager.

"Out of tokens".

Top up your tokens, or upgrade for a larger monthly allowance.

"Insufficient tokens."

Buy a token pack from Settings → Billing, or wait for your monthly allocation to refresh.

"This feature requires a paid plan."

Ask AI is available from the Starter plan up. Upgrade from Settings → Billing.

"Rate limit exceeded."

Wait until the displayed reset time. This is the per-user burst guard (30 questions per minute); it resets within a minute.