How do text response summaries work?

Conjointly uses a popular AI language model to provide summaries of respondents’ answers. This provides key insights from respondents’ feedback in cases such as the ‘Other’ option for multiple choice questions, free-form answers on open-ended feedback questions and even comments for different colour highlights in text highlighter and image heatmap questions.

Before sending the responses to the model the answers are preprocessed, mainly to remove answers that do not carry any interesting information, such as “Nothing”, “N/A” and empty answers.

While models like GPT-3 are a great help in working on natural language tasks (that would otherwise require manual human work), it is not always clear exactly why the model provides a certain output. By design, the exact way of “thinking” of AI is not always clear and is a deep research field in itself.

We encourage treating AI-based summaries as high-level insights that take into consideration all answers that provide valuable information for the text input that they are summarizing.