AI Summary 
AI Adaptive Follow-Up is a new question type that enables conversational probing powered by artificial intelligence. Starting from an initial question written by the researcher, the AI asks respondents a series of targeted follow-up questions, adapting them in real time based on the answers received, the topics to be explored, and those to be avoided.
Unlike a simple open-ended question, this question type is designed to elicit detailed and comprehensive responses, guiding respondents toward deeper reflection without requiring a human interviewer. The resulting experience is similar to a qualitative interview, but scalable to large samples and fully integrated into IdSurvey’s quantitative survey workflow.
How it works #
At the end of the conversation, IdSurvey automatically generates a faithful summary that combines all exchanges into a single coherent text, written in the first person and incorporating the meaning of the follow-up questions. This summary can be exported and analyzed in the Analysis & Reports section just like any other open-ended question, without requiring any special tools or procedures.
For users who require in-depth qualitative analysis, IdSurvey also provides the option to export the raw conversation data turn by turn, allowing it to be processed with third-party qualitative analysis tools.
In the Contacts Management and Quality Control section, by opening the preview of a completed interview, it is possible to view both the generated summary and the full original conversation, turn by turn.
Add an AI Adaptive Follow-Up question #
- Access the survey’s Question Management section.
- Click + Add Question.
- Select the AI Adaptive Follow-Up question type.
- Click the Configure Adaptive Follow-Up icon to open the configuration panel.
How to set an Adaptive Follow-Up question #
Initial question #
This is the question that starts the conversation with the respondent. It can be created and edited from this window, from the standard question management interface, or through IdCode.
Example: How was your experience with our support team?
Levels #
It defines the total number of questions in the conversation, including the initial question written by the researcher. A maximum of 10 levels can be configured.
- The first level always corresponds to the researcher’s initial question.
- Subsequent levels are generated by the AI based on the respondent’s answers.
For most use cases, a setting between 3 and 4 levels provides a good balance between response depth and interview duration.
Display mode #
It controls how questions are presented to the respondent during the conversation. Two display modes are available:
- Show all
Follow-up questions appear one below the other, keeping the entire conversation history visible. Respondents can review previous questions and their own answers at any time. - One at a time
Only the current question is displayed, while previous questions and answers are hidden. This reduces distractions and helps respondents focus on the current question.
- Show all
Consistency check #
Before accepting each response and proceeding to the next follow-up question, the AI can evaluate its relevance and completeness. If a response is considered inconsistent, off-topic, or too superficial, a message is displayed prompting the respondent to revise or expand on what they have written.
Four validation levels are available:
| Level | Behavior |
|---|---|
| 0 – Disabled | No validation. Any response is accepted without evaluation, and the system always proceeds to the next follow-up question. |
| 1 – Permissive | Accepts very brief responses (e.g., “yes”, “I don’t know”). Completely off-topic responses are not accepted. |
| 2 – Balanced | Accepts only relevant and coherent responses. If the response is too brief or vague, the AI prompts the respondent to provide more details. |
| 3 – Strict | Accepts only well-argued and detailed responses, even when the question is closed-ended. Any superficial response is rejected. |
If the respondent is not cooperative, a too strict consistency control level may hinder or prevent the continuation of the interview.
Target topics (optional) #
Topics that the AI will actively explore in follow-up questions. Enter the topics by typing them into the field and pressing Enter to add each tag.
These are not necessarily single keywords: you can also enter short directive phrases that more precisely guide the AI’s behavior.
Examples of tags:
- Response times
- Ask whether the issue has been resolved
- Agent tone and courtesy
Depending on the number of levels, the AI will attempt to cover all target topics during the conversation, adapting to the respondent’s answers.
Excluded topics (optional) #
Topics that the AI must not address in follow-up questions, regardless of what the respondent answers. This works as an exclusion list.
As with other fields, you can enter both single keywords and more specific directive phrases.
Examples of tags:
- Costs
- Do not ask about the phone channel
Context (highly suggested) #
A free-text field used to describe the survey context, research objectives, target audience, and any other information useful for guiding the AI’s behavior.
This context is not visible to the respondent: it is used exclusively by the AI to calibrate follow-up questions. A well-defined context significantly improves the quality and relevance of the generated questions.
Example:
We are collecting post-support feedback from customers who contacted customer service within the last 7 days. The goal is to understand perceived service quality and identify potential areas for improvement.
Test the configuration #
Before saving, it is possible to test the AI’s behavior by clicking the “Test configuration” button.
This opens a full flow simulation: the initial question, AI-generated follow-up questions, and response fields. You can enter test answers to verify that the AI respects the configured settings, including explored topics, exclusions, and the required coherence level.
To return to the configuration panel without saving, click “Back to configuration”.
Compatibility #
The new Adaptive Follow-Up Questions feature is available for both CAWI and CATI data collection modes.
In CAPI, since it is designed for interviews conducted without an internet connection where AI interaction is not possible, the question is displayed as a simple open-ended text field. The same fallback behavior is automatically applied when AI credits are exhausted or in the event of a temporary service issue: in these cases, IdSurvey shows a standard open-ended question, ensuring that the interview is not interrupted and that no responses are lost.
AI Access License #
Using this type of question requires AI credits. IdSurvey includes a free annual pass and offers additional top-ups to extend AI credit usage.
AI Free Pass — the free allowance included in the plan, which is automatically renewed every year and covers standard use of IdSurvey’s AI features.
Extra Top-ups — allow capacity to be extended at any time. They are ideal for projects with high volumes or surveys that include large-scale follow-up questions.
AI credit consumption is dynamic and varies based on the complexity and configuration of the tool, the amount of data processed, and the length of the conversations.
As an indication, a single extra top-up allows you to manage:
approximately 12,000–18,000 follow-up questions with 3 levels
approximately 8,000–12,000 follow-up questions with 3 levels with coherence control enabled
