> For the complete documentation index, see [llms.txt](https://docs.askyazi.com/llms.txt). Markdown versions of documentation pages are available by appending `.md` to page URLs; this page is available as [Markdown](https://docs.askyazi.com/study-types/ai-interviews.md).

# AI interviews

Set up an AI-moderated interview on WhatsApp, understand every configuration field, and know how the interview behaves and ends.

An AI interview starts with a short structured survey and then hands over to an AI interviewer that asks open, adaptive follow-up questions in the same WhatsApp chat. Use it when you need to understand motivations, experiences and reasons, and still want some countable data alongside.

{% hint style="info" %}
**In short:**

* An AI interview has two parts on the **Build** tab: **Survey questions** (structured, the same builder as a survey) and **Interview configuration** (what the AI explores and how).
* You set the interview length as a maximum number of AI questions, from 2 to 30. More key topics means fewer questions on each topic.
* The interview ends when the participant says they are done, when the question limit is reached, or earlier if you describe when the AI may wrap up. AI follow-ups appear in **Transcripts**, not in **Graph Data**.
  {% endhint %}

## How it works for the participant

1. The participant joins from a link, QR code or WhatsApp invite.
2. They answer the survey questions first: single and multi-select, scales, media uploads and so on.
3. The AI interviewer takes over. It reads their survey answers, your interview configuration and the conversation so far, and asks one question at a time.
4. It acknowledges what they said, follows up on vague or interesting answers and moves between your topics. Participants can reply by text, voice note, photo or video. Voice notes are transcribed and the AI responds to what was said.
5. The interview closes with your completion message.

<figure><img src="https://358667285-files.gitbook.io/~/files/v0/b/gitbook-x-prod.appspot.com/o/spaces%2FzMRQUXiGQlBwWdg6WEmG%2Fuploads%2F8NZROA4Ksajh134hsUDp%2Fph-ai-interview-probe.webp?alt=media&amp;token=c708c73c-4c6e-4de3-8e16-75fd98468545" alt="An AI interview in the Test Simulator: the participant answers and the AI follows up on what they said"><figcaption></figcaption></figure>

The AI asks and probes in the participant's own language, rather than running in English and translating afterwards. To the participant it feels like one continuous conversation.

## Set up an AI interview

{% stepper %}
{% step %}

### Create the study

Click **+ New Research**. Choose **Build with AI** to upload a brief or a discussion guide, or **Build manually** and pick **Interview** as the research method. See [Create a study](https://docs.askyazi.com/build-a-study/create-a-study.md).
{% endstep %}

{% step %}

### Write the survey questions

On **Build > Survey questions**, add the structured questions that come before the interview. Keep this part short, usually 8 to 12 questions. Include the classification and behaviour questions the AI can build on, such as product used, frequency or a rating. Rating and single-select answers give the AI clear points to probe.

<figure><img src="https://358667285-files.gitbook.io/~/files/v0/b/gitbook-x-prod.appspot.com/o/spaces%2FzMRQUXiGQlBwWdg6WEmG%2Fuploads%2FwRn0PewC77MDIWW8pQ4S%2Ffs-interview-build.webp?alt=media&amp;token=c7fd1f3f-f8d4-49b1-a18e-1ec73f4e071f" alt="An AI interview in Build: Survey questions run first, then the AI conversation set up in Interview configuration"><figcaption></figcaption></figure>
{% endstep %}

{% step %}

### Fill in the interview configuration

On **Build > Interview configuration**, work through the four sections described below. **Study overview** is required; the rest have defaults.
{% endstep %}

{% step %}

### Test with the test link and simulator

Run the interview yourself several times before going live. See [Testing an AI interview](#testing-an-ai-interview) below and [Test your study](https://docs.askyazi.com/study-types/test-your-study.md).
{% endstep %}
{% endstepper %}

<figure><img src="https://358667285-files.gitbook.io/~/files/v0/b/gitbook-x-prod.appspot.com/o/spaces%2FzMRQUXiGQlBwWdg6WEmG%2Fuploads%2F69HtVaCNqApgVikdDPNQ%2Ffs-interview-overview.webp?alt=media&amp;token=326e5302-732c-4e0d-98de-cae905c5d987" alt="Interview configuration > Study overview: research objectives (with a guide to how much detail is enough), target audience and key topic areas with their weighting"><figcaption></figcaption></figure>

## Interview configuration reference

The left of the **Interview configuration** tab lists four sections: **Study overview**, **Interview approach**, **Context & preferences** and **Additional** (optional).

### Study overview

This is what the interview must cover. All three fields are required.

| Field                   | What to put in it                                                                                                                                                                                                                                                                                                                                     |
| ----------------------- | ----------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------- |
| **Research objectives** | The insight you need and the decision it supports, not just the topic. A quality meter under the box tells you when the brief is too thin for the AI to run a good interview. You can start from a chip (Product feedback, Consumer insights, Brand health, Competitive analysis) or click **Browse all templates** to start from a ready-made brief. |
| **Target audience**     | Who the participant is: relevant behaviour, life stage, location, category use. The AI uses this to pitch its tone and language.                                                                                                                                                                                                                      |
| **Key topic areas**     | Between 3 and 7 themes the AI must explore. Each topic has its own question allocation, set with a minus and plus counter, with its share shown as a percentage (for example "8 of 8 questions allocated").                                                                                                                                           |

When you finish the research objectives, an assessor reads the brief back to you with a quality rating, a plain-language summary and suggested fixes you can apply in one click. It can also suggest a target audience, key topics and required questions. It is advisory only: it never changes your settings unless you accept a suggestion, and anything it changes can be undone.

**Example objective:** "Understand why first-time users stop using the app after onboarding, and which moments create confusion, distrust or low motivation to continue, so the product team can prioritise fixes."

**Required questions.** You can add questions that must be asked word for word, either in the configuration or by asking Ask Yazi. The AI always asks them near the end of the interview, translated if the interview runs in another language, and you can reorder them. Use them sparingly: a few are fine, but many make the interview feel scripted and use up the question budget.

### Interview approach

This is how the AI asks questions.

| Field                         | Options and guidance                                                                                                                                                                                                                                                                           |
| ----------------------------- | ---------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------- |
| **Question styles**           | Open-ended, Scenario-based, Behavioral, Feelings-based, Future-oriented. Pick one or more; the AI blends whichever you pick. For exploratory work, Behavioral + Feelings-based + Open-ended works well. For concept work, Scenario-based + Future-oriented + Open-ended.                       |
| **Probing intensity**         | **Minimal**: follows up only when an answer is vague. **Standard**: follows up on vague answers and contradictions. **Deep**: probes emotions, contradictions and new themes. Standard suits most studies.                                                                                     |
| **Interview length**          | The maximum number of AI questions before the interview wraps up, from 2 to 30. The app recommends 8 to 15 for a 15 to 20 minute interview.                                                                                                                                                    |
| **Ending early**              | The interview always ends at the question limit ("Maximum questions reached" is always on). In the free-text box (up to 1,000 characters) you can describe when the AI may wrap up sooner, for example "once the participant has described their last purchase and why they chose that brand". |
| **Multiple-choice questions** | When on, the AI may ask tappable single-select questions where they fit. The answers come back as structured options you can analyse as numbers.                                                                                                                                               |

<figure><img src="https://358667285-files.gitbook.io/~/files/v0/b/gitbook-x-prod.appspot.com/o/spaces%2FzMRQUXiGQlBwWdg6WEmG%2Fuploads%2FgERrwNg75wdigoeZd21O%2Ffs-interview-approach.webp?alt=media&amp;token=f6e4e192-6388-4469-a938-4a83c4c6e816" alt="Interview approach: question styles, probing intensity, interview length (2 to 30 AI questions) and when the AI may end early"><figcaption></figcaption></figure>

### Context & preferences

This is how the conversation adapts.

<figure><img src="https://358667285-files.gitbook.io/~/files/v0/b/gitbook-x-prod.appspot.com/o/spaces%2FzMRQUXiGQlBwWdg6WEmG%2Fuploads%2FpCxGixbhATlZ0UmyemQA%2Ffs-interview-context.webp?alt=media&amp;token=95305008-4909-4764-ab97-8018e7b4d674" alt="Context and preferences: the media the AI encourages, the interview language and cultural notes"><figcaption></figcaption></figure>

| Field                         | Options and guidance                                                                                                                                                                                                                                        |
| ----------------------------- | ----------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------- |
| **Media the AI encourages**   | Text responses (always on), Voice notes, Photos / Images and Videos, each with a priority such as Secondary or Optional. Videos take the most effort, so use them sparingly. Voice notes suit stories, emotion and audiences who prefer speaking to typing. |
| **Interview language**        | **Same as survey**, or **Choose language**. To add more languages, turn on multi-language support in the survey settings first. See [Languages and translation](https://docs.askyazi.com/study-types/languages-and-translation.md).                         |
| **Language & cultural notes** | Words participants actually use, words to avoid, brand names as people say them and local context. Example: "Say 'township', not 'informal settlement'."                                                                                                    |
| **Sensitive topics**          | **Avoid sensitive topics**, **Approach gently** or **Fully explore**.                                                                                                                                                                                       |
| **Cross-campaign context**    | When on, the AI can use this participant's answers from your earlier studies.                                                                                                                                                                               |

### Additional (optional)

| Field                      | Options and guidance                                                                           |
| -------------------------- | ---------------------------------------------------------------------------------------------- |
| **Comparative brands**     | Brands the AI may raise for comparison.                                                        |
| **Conversation structure** | **Let AI decide optimal flow**, **Start broad, get specific**, or **Jump right into details**. |

<figure><img src="https://358667285-files.gitbook.io/~/files/v0/b/gitbook-x-prod.appspot.com/o/spaces%2FzMRQUXiGQlBwWdg6WEmG%2Fuploads%2Fa2wwnKBxg41pyo5DYTHO%2Finterview-context.webp?alt=media&amp;token=ea50fa51-2b34-4f91-95ff-26532a31d13e" alt="Interview configuration settings for sensitive topics, cross-campaign context, comparative brands and conversation structure"><figcaption></figcaption></figure>

### Topic budgeting: the trade-off to plan for

The interview length is a fixed budget shared across your key topics. By default it is split evenly, and you can change each topic's allocation with its counter. If you later change the interview length, the allocations rescale to keep your split.

The more topics you add, the less depth each one gets. Seven topics across twelve questions is one or two questions per topic, which leaves little room to probe. If depth matters, use fewer topics or a longer interview. Required questions are always asked regardless of the split.

{% hint style="info" %}
You can also change most of these settings by asking Ask Yazi, for example "make the probing deeper" or "add a key topic about price". On a live or paused study it shows you an approval card first. See [Ask Yazi](https://docs.askyazi.com/build-a-study/ask-yazi.md).
{% endhint %}

## How an interview ends

An interview finishes in one of three ways:

1. **The participant says they want to stop.** A genuine "I'm done" ends the interview politely.
2. **The question limit is reached.** Any required questions are asked before this point.
3. **The AI wraps up early** because the condition you described under **Ending early** has been met.

An interview is not marked finished just because the participant goes quiet. People often come back, and the interview picks up where it left off. If you use [reminders](https://docs.askyazi.com/reach-participants/reminders.md), the participant's **Continue** button returns them to the exact question they were on.

### Letting people finish without the interview

In the survey part, **Question logic** offers two different endings:

* **Complete** finishes the participant at that point and skips the AI interview. They count as complete. Use it for people who only agreed to the scripted part, or who answer "no" to an opt-in question.
* **End of survey** ends the survey part and hands over to the AI interview.

See [Logic and routing](https://docs.askyazi.com/build-a-study/logic-and-routing.md).

## Where the results appear

* **Transcripts** (under **Responses**) show the full conversation, including every AI follow-up and the participant's answers, with voice notes transcribed. See [Transcripts](https://docs.askyazi.com/results-and-reporting/transcripts.md).
* **Table Data** has the survey answers in columns and the interview transcript, plus a **Translated Transcript** column when the interview ran in another language.
* **Graph Data** charts the structured survey questions and any AI multiple-choice questions. Open AI follow-ups are not charted, because every follow-up is different for each participant.
* **Insights** groups open answers and interview content into themes. See [Insights](https://docs.askyazi.com/results-and-reporting/insights.md).

## Organisation-wide interview prompt

Under **Settings > AI Settings**, the **Interview Prompt** sets the tone and behaviour for all AI interviews in your organisation. It must contain the parameters `{objective}` and `{survey_data}`. Most teams should leave this alone and use the per-study **Interview configuration** instead, because a change here affects every interview. If you think you need to change it, talk to Yazi first.

## Testing an AI interview

Plan for two or three rounds of testing and refinement.

1. **Cooperative run.** Answer as an engaged participant would.
2. **Low-effort run.** Give short, vague answers and check that the AI probes.
3. **Off-topic run.** Wander off topic and check that the AI steers back.
4. **Read the transcript.** Look for repeated questions, missed topics, weak probing and awkward transitions, then adjust the configuration and run again.

{% hint style="warning" %}
Always use the **test link** for repeat testing. The live link remembers your earlier conversation, so if you already reached the question limit on it, the AI will not ask you anything more. See [Test your study](https://docs.askyazi.com/study-types/test-your-study.md).
{% endhint %}

## Brief checklist

Ask the client or study owner for these before you build:

* The core research objective and the decision it supports
* A precise description of the target audience
* 3 to 7 key topic areas, and which matter most (for the question allocation)
* Any questions that must be asked word for word
* Preferred question styles and probing intensity
* Ideal interview length
* Which media to encourage: text, voice notes, photos, videos
* Language, terms to use and terms to avoid
* Sensitive topics and how to handle them
* When the interview may end early

A human discussion guide does not convert directly into an AI brief. A moderator skips a planned probe when they already have the answer; the AI follows everything it is given. Condense the guide into topics and decide where probing is really worth the participant's time.

## Limits and things to know

* **No fixed question order.** The AI chooses the sequence from the conversation. If something must be asked exactly, make it a survey question or a required question.
* **Minimum of 2 AI questions.** With fewer, the interview would end before the AI asks anything.
* **Thin answers are not always deepened.** The AI sometimes moves to a new topic after a one-word answer. Deeper probing helps, but it uses questions that would otherwise go to other topics.
* **Reply time.** The AI takes a few seconds to read an answer and respond, longer when it waits for a voice note to be transcribed. When the last survey question is a voice note, the interview waits up to 45 seconds for the transcript before asking its first question.
* **Not for high-risk topics without a plan.** The AI has no built-in distress detection. For sensitive studies, agree a protocol, say in your opening message that the chat is not a support line, and consider a human moderator. A moderator can take over any conversation at any time. See [Agent takeover](https://docs.askyazi.com/monitor-fieldwork/agent-takeover.md).
* **Language quality varies.** See [Language support](https://docs.askyazi.com/study-types/language-support.md). Some system messages in a few languages still fall back to English.
* **AI switched off.** AI features can be switched off for an organisation, feature by feature. Anything switched off shows a disabled state. See [How Yazi uses AI](https://docs.askyazi.com/security-and-privacy/ai-and-your-data.md).

## Related

* [Surveys](https://docs.askyazi.com/study-types/surveys.md)
* [Multi-activity studies](https://docs.askyazi.com/study-types/multi-activity-studies.md)
* [Agent takeover](https://docs.askyazi.com/monitor-fieldwork/agent-takeover.md)
* [Transcripts](https://docs.askyazi.com/results-and-reporting/transcripts.md)
* [Test your study](https://docs.askyazi.com/study-types/test-your-study.md)
