> ## Documentation Index
> Fetch the complete documentation index at: https://docs.buildbetter.ai/llms.txt
> Use this file to discover all available pages before exploring further.

# AI Surveys Overview

> Send targeted surveys, collect conversational responses with AI follow-ups, and turn them into signals alongside the rest of your feedback

AI Surveys (`/surveys`) is BuildBetter's native survey product. You build a survey, target an audience using the same signal filters you use everywhere else, deliver it by email or inside your own app, and let an AI ask clarifying follow-up questions when an answer is thin. Completed responses flow into the same signal tables as calls, documents, and imported feedback.

<Info>
  AI Surveys is gated by the organization-level `surveys` feature flag. When it is off, authenticated survey mutations are rejected, the survey worker holds pending invitations instead of sending them, and public respondent links return **Survey not found**.
</Info>

## Three Ways Responses Arrive

BuildBetter treats all three as the same kind of data once collected:

| Path                   | How it works                                                                                                                 | Best for                          |
| ---------------------- | ---------------------------------------------------------------------------------------------------------------------------- | --------------------------------- |
| **Imported**           | Responses synced from a provider like [Typeform](../Integrations/other/typeform) or [Pendo](../Integrations/analytics/pendo) | Surveys you already run elsewhere |
| **Native email / URL** | Built and sent from BuildBetter, answered on a hosted respondent page                                                        | Targeted outreach to known people |
| **Native in-app**      | Delivered through the installed BuildBetter feedback widget                                                                  | Catching users in product context |

Imported surveys stay readable in BuildBetter but are not editable here — edit those in the source tool.

## How the Pieces Fit

Every collection path converges on the same records:

```
survey -> survey_version -> survey_question -> survey_response -> survey_response_answer
```

The important consequence: **questions are versioned and immutable once sent.** When you add a recipient, that recipient is pinned to the `survey_version` that existed at the time. Editing questions later creates a new version and does not retroactively change what an in-flight respondent sees, or how their answers are interpreted.

Completed non-preview responses are queued for signal extraction, so they show up in Signals, extraction tables, and datasets exactly like any other source.

## The Lifecycle

<Steps>
  <Step title="Create">
    Start from a template or build your own. Pick question types, write descriptions, and decide which questions allow AI follow-ups. See [Creating Surveys](./creating-surveys).
  </Step>

  <Step title="Target">
    Build the audience with the Signals v2 filter builder, review the matched people with their matching evidence, and approve who gets added. See [Audience & Delivery](./audience-and-delivery).
  </Step>

  <Step title="Brand and preview">
    Set theme, primary color, logo, intro copy, and thank-you copy. Preview renders the real respondent page. See [Branding & Respondent Experience](./branding-and-respondent-experience).
  </Step>

  <Step title="Activate">
    Activation queues invitations for pending recipients and surfaces the estimated credit exposure before you commit.
  </Step>

  <Step title="Read the results">
    Watch delivery and respondent progress, open individual responses with their full AI follow-up history, and connect the survey to a View so its signals join that analysis. See [Responses & Signals](./responses-and-signals).
  </Step>
</Steps>

## What Makes It Different

**Targeting is signal-led, not list-led.** The primary way to build an audience is to describe who you want in signal terms — the same fields, nested rules, custom person and company properties, and hierarchy filters available on the Signals page. BuildBetter returns matched people *with the signals that matched them*, so you can audit why someone is in the audience before you send. Manual email entry exists for one-offs, but it is not the main model.

**Follow-ups are model-decided, not scripted.** When a question allows follow-ups and an answer is low-detail, a model decides whether a clarifying question is worth asking. It may ask zero. There is no canned fallback prompt — if the model declines or fails, the answer simply saves and the respondent moves on.

**Responses and signals are counted separately.** "Responses received" and "signals generated" are different outcomes and the UI keeps them distinct, so a survey that collected 40 responses but produced few signals is visible as exactly that.

## Where to Go Next

<CardGroup cols={2}>
  <Card title="Creating Surveys" icon="pen-to-square" href="./creating-surveys">
    Templates, question types and variants, and AI follow-up configuration.
  </Card>

  <Card title="Audience & Delivery" icon="paper-plane" href="./audience-and-delivery">
    Signal-led targeting, email and in-app channels, throttles, and suppression.
  </Card>

  <Card title="Responses & Signals" icon="chart-line" href="./responses-and-signals">
    Reading responses, follow-up history, and routing signals into Views.
  </Card>

  <Card title="Branding & Respondent Experience" icon="palette" href="./branding-and-respondent-experience">
    Theme, color, logo, intro and completion copy, and what respondents see.
  </Card>

  <Card title="Settings & Billing" icon="gear" href="./settings-and-billing">
    Sender identity, organization settings, credit rates, and limits.
  </Card>
</CardGroup>

<Note>
  Surveys are also fully available through [BuildBetter MCP](../MCP/tool-reference), including creating a complete survey in one operation, sending a preview, activating, and retrieving responses with their follow-up answer pairs.
</Note>