# Agentic interviews

With agentic interviews, an AI interviewer talks to your customers for you. You set up a
research with the objectives you want to explore and the questions you want answered, then
invite contacts. The AI emails each person, follows up on their answers and can offer a voice
conversation instead. Every finished interview is summarized, and you can summarize the whole
research to see the themes that come up again and again.

<Availability
  permission="Agentic interviews: Read to see research, Create or Update to set it up, and Delete to delete it. Customer meetings & support: Read to see the interviews, and Create to invite people."
  plan="Every plan. Your plan sets how many researches a workspace can have, and how many minutes of voice interviews you can use. See Plans and limits."
/>

## Key concepts

**Research.** A study with a title, an optional description, **objectives** and
**questions**. Every agentic interview belongs to one research, and so does every conversation
from the [feedback widget](/user-guide/tools/feedback-widget), and every AI conversation about a
comment on a [shared live preview](/user-guide/sessions/live-preview#sharing-a-live-preview-publicly).
A new workspace starts with one research already created, called **General**.

**Objectives.** What you want to learn, such as "Understand why users abandon onboarding". The
AI interviewer steers each conversation toward them.

**Questions.** The concrete questions the AI interviewer asks each participant, in order. It then
asks follow-up questions based on the answers.

**Interview.** One conversation with one person. Its type is shown as an icon on its row:

| Type | What it is |
| --- | --- |
| **AI Email** | The AI interviewer emails the participant and follows up by email. The email can also offer a voice conversation. |
| **AI Voice** | A spoken conversation with the AI interviewer, from a link in the participant's browser. An **AI Email** interview becomes **AI Voice** when the participant takes the voice conversation. |
| **Feedback widget** | A conversation someone started in the feedback widget on your website or app, or by commenting on a shared live preview whose AI replies are on. |

**Status.** Where an interview is:

| Status | Means |
| --- | --- |
| **Draft** | Created, and nothing has happened yet. |
| **Invited** | The first email has been sent to the participant. |
| **In Progress** | The participant has replied or the conversation is under way. |
| **Processing** | Productmind is writing the interview's summary. |
| **Completed** | The summary is ready. |
| **Invalid** | The conversation didn't contain a usable interview, so there is nothing to summarize. |
| **Failed** | Writing the summary failed. |
| **Cancelled** | The AI interviewer stopped or never started, for example because the conversation stalled, or because billing refused the interview before it started. The interview's panel then says why under its status. |

**Sentiment.** Each summarized interview is **Positive**, **Neutral** or **Negative**. The
research shows how many interviews have each.

## The Agentic interviews page at a glance

In the sidebar, under **Feedback**, choose **Agentic interviews**.

<Screenshot id="agentic-interviews-list" alt="The Agentic interviews page: a list of researches, each with sentiment counts, a completed-of-total badge and a one-line summary" />

**The toolbar** holds **New research** (the **+** button), the search box and **Sort**,
which orders the list by **Date** or **Title**.

**Each row** shows:
- the research's title;
- how many interviews are negative, neutral and positive;
- how many interviews are completed out of how many there are, such as **39/43**. The badge
  turns green when all of them are completed;
- when the research was last updated;
- the research's summary or, until it has one, its description.

Choose a research to open it beside the list. Choose it again, or **Close** (the X), to close it.
Right-click a research to open it in a new tab or to delete it. To
delete several researches at once, select them with the checkboxes that appear when you point at
a row, then choose **Delete selected**.

### The research page

<Screenshot id="agentic-interviews-research" alt="A research open beside the list on its Overview: the description and summary, then Key themes, Question insights and Linked feature requests" />

**The header** shows how many interviews are completed (**39 of 43 interviews**), the sentiment
counts and the research's title, which you can choose to edit. Once the research has a summary,
**Refresh summary** (the circular arrows) writes a new one. A badge such as **3 new** says how many
interviews were completed since the last summary. **Delete research** (the trash icon) deletes the
research. The icon at the header's left shows or hides the list of researches.

Down the left of the research is a list of its sections. When the research has little room, the
list shows only their icons; point at one to see its name.

| Section | What it holds |
| --- | --- |
| **Overview** | The research's description, which you can choose to edit, and its summary. The research opens here. |
| **Interviews** | Every interview in this research. See [Reading an interview](#reading-an-interview). |
| **Feedback widget** | This research's feedback widget: its welcome message, a playground to try it, the widget key and the embed code. See [Feedback widget](/user-guide/tools/feedback-widget). |
| **Settings** | The research's objectives, questions and the opening message for voice interviews. |

## Setting up a research

### To create a research

1. In the sidebar, choose **Agentic interviews**.
2. Choose **New research** (the **+** button).
3. **Name**: enter a **Research title** and, if you like, a **Description**. Choose **Next**.
4. **Objectives**: enter an objective and press Enter. Add as many as you need, then choose
   **Next**.
5. **Questions**: enter a question and press Enter. Add as many as you need.
6. Choose **Create research**.

The research opens. Each step needs at least one entry before you can go on.

:::tip
Ask open questions ("How do you prioritize requests today?") rather than yes-or-no ones, and one
thing at a time. The AI interviewer asks follow-up questions on its own, so a short list of good
questions goes a long way.
:::

### To change objectives and questions

1. Open the research and choose **Settings**.
2. Choose **Expand** next to **Research objectives** or **Questions to ask participants**.
3. Change the list:
   - choose **Add**, enter the text and press Enter;
   - choose an item to edit it;
   - drag an item by its handle to reorder it. The AI interviewer asks the questions in this
     order;
   - point at an item and choose **Remove item** to delete it.

Each change is saved as you make it.

### To set the opening message for voice interviews

The opening message is the first thing a participant hears when a voice interview starts.

1. Open the research and choose **Settings**.
2. Choose **Expand** next to **Opening message for voice AI agents**.
3. Choose the message and enter your own.

You can personalize it with `{{customer_name}}`, `{{company_name}}` and `{{product_name}}`. Leave
it empty to use the default message.

## Running interviews

### To invite people to an interview

The people you invite must be [contacts](/user-guide/feedback/contacts) with an email address.

1. Open the research and choose **Interviews**.
2. Choose **Add interview** (the **+** button).
3. Select the contacts you want to interview. Search by name, email or account. To add someone
   who isn't a contact yet, choose **Create contact**. The new contact is selected for you.
4. Choose **Next**.
5. Check the list. To let participants talk to the AI interviewer instead of writing, select
   **Enable AI voice interviews**.

   <Screenshot id="agentic-interviews-invite" alt="The Confirm invitations step: two contacts, and the Enable AI voice interviews option" />

6. Choose **Send invitations**.

An interview is created for each contact, and the AI interviewer writes and sends each one an
invitation email. The interview moves to **Invited** once the email is sent.

If billing refuses the invitation, because your organization's Pay-as-you-go balance can't cover
the AI interviewer, pay-as-you-go isn't activated, spending is paused or a usage limit is reached,
the interview isn't created. The dialog stays open with the reason, for example "Your
Pay-as-you-go balance is EUR 0.12, and AI work needs at least EUR 0.78 available. Top up to
continue.", and Productmind shows the action that fixes it, such as **Top up**, or the **Usage
limit reached** dialog. Once it's fixed, choose **Send invitations** again.

:::note
The research needs at least one objective and one question before you can invite anyone. Until it
has both, **Invite to interview** says so and offers **Open Settings**.
:::

### How an email interview runs

1. The AI interviewer emails the participant, based on your objectives and questions.
2. When the participant replies, it reads the answer and sends a follow-up, until it has what it
   needs. The interview is **In Progress**.
3. When the interview is finished, the AI interviewer sends a thank-you email and Productmind
   summarizes the conversation. The interview moves to **Processing**, then **Completed**.

If you selected **Enable AI voice interviews**, the invitation also contains a link to a voice
interview. The participant opens it in their browser, allows the microphone and chooses **Start
Interview**. When the conversation ends, Productmind summarizes it.

## Reading an interview

In the research's **Interviews** section, choose an interview to open it beside the list. While
an interview is open, the list of researches folds away to make room, and comes back when you
close the interview.

<Screenshot id="agentic-interviews-interview" alt="An interview open beside the Interviews list, showing its summary, actionable insights and a notable quote" />

**The list** can be searched by title, summary, contact name, email or account. **Filters**
narrows it by **Status**, **Type** and **Sentiment**, and **Sort** orders it by **Title**,
**Contact name**, **Type**, **Status**, **Sentiment** or **Date**.

**Each row** shows the interview's type, its title and contact, its date, its sentiment and its
status. Until an interview is summarized, its title is the contact's name.

**The panel** shows the contact, the sentiment and the type, then the interview's title, which
you can edit, and its status. Once the interview is summarized, the panel has these tabs:

| Tab | What it shows |
| --- | --- |
| **Summary** | What the participant said, the **Actionable insights** and **Notable quotes**. For an **AI Voice** interview, also the interview link, with **Copy**. |
| **Answers** | Each of the research's questions, with the participant's answer, or "Not addressed in this interview." |
| **Feature requests** | The feature requests created from this interview. Appears when there are some. |
| **Bug reports** | The bug reports created from this interview. Appears when there are some. |
| **Conversation** | Every message in the conversation. |

Before an interview is summarized, the panel shows its raw content instead.

### To summarize an interview again

The button is on an interview that has been summarized, and on one that is still **In Progress**.

1. In the interview's panel, choose **Re-summarize this interview** (the circular arrows). On an
   interview that is still **In Progress**, the button reads **Interview still in progress**.
2. Choose **Re-summarize**.

The new summary replaces the old one, with its insights and feature requests. The work continues
in the background, so you can leave the page.

If the interview is still **In Progress**, the summary uses the conversation so far, and the
interview moves out of **In Progress**.

### To delete interviews

1. Select the interviews with the checkboxes at the start of their rows.
2. Choose **Delete selected**.
3. Type **delete**, and confirm.

To delete one interview, you can also right-click it and choose **Delete**.

:::caution
Deleting an interview can't be undone.
:::

## Summarizing a research

A research's summary brings every completed interview together.

### To summarize a research

1. Open the research. It opens on **Overview**.
2. Choose **Summarize this research** (the circular arrows).
3. Choose **Summarize**.

You need at least one completed interview. You can summarize at any time, even while more
interviews are still in progress. The work continues in the background.

The summary has four parts:
- an overview of what the interviews found;
- **Key themes**: the patterns that recur across interviews, each with how many interviews
  mention it and supporting quotes;
- **Question insights**: the key takeaway for each of your questions;
- **Linked feature requests**: the feature requests that came from this research's interviews,
  with how many times each was mentioned.

Choose **Expand** to open a part.

### To refresh the summary

1. In the research's header, choose **Refresh summary**.
2. Choose **Re-summarize**.

The new summary replaces the old one, with its question insights and linked feature requests.

## Deleting a research

### To delete a research

1. Open the research and choose **Delete research** (the trash icon).
2. Enter **delete**, and choose **Delete**.

The research is deleted and the **Agentic interviews** page opens.

You can also delete a research from the **Agentic interviews** page: right-click it and choose
**Delete**. To delete several at once, select them with the checkboxes that appear when you point
at a row, then choose **Delete selected**. Either way, enter **delete** to confirm.

:::caution
Deleting a research also deletes all of its interviews, and its feedback widget stops working on
every site that embeds it. It can't be undone.
:::

## How agentic interviews behave

**Summaries create feature requests and bug reports.** Summarizing an interview works as for a
customer meeting: what the participant asked for becomes
[feature requests](/user-guide/product/feature-requests) and what they ran into becomes
[bug reports](/user-guide/bug-reports), linked back to the interview, the contact and the
contact's account. See
[How customer meetings behave](/user-guide/feedback/customer-meetings#how-customer-meetings-behave).

**Summaries are written in the background.** A summary can take a few minutes. The page updates
on its own when it's ready.

**A summary that seems stuck can be restarted.** While an interview is **Processing**, you can't
start another summary. Once it has been processing for 45 minutes, the circular arrows can be
chosen again, as **This run looks stuck**: choose them, then **Start new run**. Soon after that,
Productmind also gives up on the run by itself: the interview moves back to **Completed** if it
already had a summary, or to **Failed** if it didn't.

**Billing is checked before the AI interviewer starts.** Once when you send the invitations, and
again when the AI interviewer starts, unless the interview runs on your organization's own model
provider key, which uses no AI credits. If that second check refuses an interview, for example
because the balance ran out in between, the interview moves to **Cancelled** and nothing is spent
on it. Its panel says why under its status, such as **Not enough Pay-as-you-go balance**, with the
action that fixes it: **Top up**, **Activate pay-as-you-go**, **Open Usage** or **View plans**. A
member who can't take the action reads who can, for example "Ask an Owner or Admin to top up the
Pay-as-you-go balance." Your Owners and Admins are also notified, at most once a day. See
[Billing notices](/user-guide/profile-and-notifications#billing-notices).

**Comments on a shared live preview become interviews.** When a session's live preview has a
[public link](/user-guide/sessions/live-preview#sharing-a-live-preview-publicly) with **AI replies
to comments** on, the AI interviewer answers each comment an outside viewer leaves, guided by the
research the link names (**General** unless someone chose another). Each comment's conversation is
saved in that research as an interview of the **Feedback widget** type, under a contact for the
email the viewer gave. It ends when your team resolves the comment, when the link or its AI replies
are turned off, when your workspace turns off public preview links, or after 24 hours without a new
message. Productmind then summarizes it, and turns what the viewer asked for into feature requests
and the problems they met into bug reports, as for any interview. The replies use your
organization's AI credits.

**Notifications.** When a voice interview or a feedback-widget conversation is summarized,
everyone in the workspace with **Customer meetings & support: Read** is notified. See
[Profile and notifications](/user-guide/profile-and-notifications).

## Permissions

| To | You need |
| --- | --- |
| See researches and their summaries | **Agentic interviews: Read** |
| Create a research | **Agentic interviews: Create** |
| Rename a research, edit its objectives and questions, summarize it | **Agentic interviews: Update** |
| Delete a research | **Agentic interviews: Delete** |
| See the **Interviews** section | **Customer meetings & support: Read** |
| Invite people | **Customer meetings & support: Create** |
| Rename or summarize an interview | **Customer meetings & support: Update** |
| Delete interviews | **Customer meetings & support: Delete** |
| See the **Feature requests** tab of an interview | **Feature requests: Read** |

By default, the **Owner**, **Admin**, **Power User** and **Product Manager** roles can do all of
these. **Viewer** and **Customer** can see research and interviews but not change them.
**Developer** can't see **Agentic interviews**. See
[Roles and permissions](/user-guide/reference/roles-and-permissions).

## Limits

- **Researches per workspace.** Your plan can limit how many researches a workspace has. When
  the limit is reached, creating one fails with "This workspace has reached its customer
  research limit".
- **Voice interview minutes.** Voice interviews use minutes from your plan's allowance for the
  billing period, billed by the second. If your plan includes none, inviting with voice interviews fails with "Voice
  interviews are not available on your current plan". If you've used them all and your
  organization doesn't allow overage, it fails with a message that says so.

See [Plans and limits](/user-guide/reference/plans-and-limits) and
[Usage](/user-guide/settings/billing-and-payments/usage).

## Troubleshooting

### Next is unavailable in Invite to interview

A contact you selected has no email address. The dialog names the contact, or says how many
contacts have none when there are several. Add an email address to the contact on the
[Contacts](/user-guide/feedback/contacts) page, or clear it from the selection.

**Next** is also unavailable while the research has no objectives or no questions. The dialog says
so: choose **Open Settings** and add at least one of each.

### "AI interviews require objectives and questions"

The research has no objectives or no questions. Add at least one of each in its **Settings**
section, then invite again.

### An interview stays in Processing

Summaries run in the background and can take a few minutes. Once it has been processing for 45
minutes, choose the circular arrows (**This run looks stuck**), then **Start new run**. If you
leave it, Productmind gives up on the run soon after, and the interview moves to **Completed** or
**Failed**.

### An interview shows Failed

Productmind couldn't write the interview's summary. An interview in **Failed** can't be
summarized again from its panel: the circular arrows aren't shown there. The panel still shows
the conversation's raw content.

### An interview shows Cancelled

The AI interviewer stopped or never started. If billing refused the interview, its panel says why
under the status, with the action that fixes it. Other reasons appear under **Why it was
cancelled**. A cancelled interview doesn't start again by itself: once the cause is fixed, invite
the contact again. See [How agentic interviews behave](#how-agentic-interviews-behave).

### Send invitations is refused with a billing message

Billing refused the invitation, so no interview was created. The dialog shows the reason, and
Productmind offers the fix: **Top up**, **Activate pay-as-you-go**, **Open Usage**, or the **Usage
limit reached** dialog. Only an Owner or Admin can top up, activate pay-as-you-go or turn spending
back on. See [Usage](/user-guide/settings/billing-and-payments/usage).

### The Interviews section is missing

You need **Customer meetings & support: Read** to see interviews. Ask an Owner or Admin of your
organization.

## Related topics

- [Contacts](/user-guide/feedback/contacts)
- [Feedback widget](/user-guide/tools/feedback-widget)
- [Live preview](/user-guide/sessions/live-preview#sharing-a-live-preview-publicly)
- [Customer meetings](/user-guide/feedback/customer-meetings)
- [Feature requests](/user-guide/product/feature-requests)
- [Plans and limits](/user-guide/reference/plans-and-limits)
