AI summary of open-ended text responses


With Conjointly’s AI summary of open-ended text responses functionality, you can automatically obtain high-level summaries and sentiment analyses for all open-ended questions in your reports.

From USD 0.15 per respondent, the function summarises open-ended responses to the following questions, including:

Example of AI summary of open-ended text responses

How to enable AI summary of open-ended text responses?

Here are two intuitive ways to enable AI summary of open-ended text responses in your experiment reports.

Method 1: Via Experiment Settings

Navigate to the Analytics options tabs under Advanced Settings of your experiment, and click on the checkbox next to Summarise text responses .

Locating the account settings to set up AI summaries of open-ended text responses on Conjointly’s platform

Method 2: Via Report Dashboard

When viewing reports from experiments without AI summary enabled, the Enable AI summary of responses button will appear on applicable sections and questions. Click the button to activate the function.

Enabling AI summary of text responses through report

For comments in text highlighter and image heatmap questions, the Enable AI summary of responses button is located on the question tab.

Enabling AI summary of text responses for text highlighter and image heatmap questions

If you are the primary owner of the experiment, you will be prompted to set up auto top-up if you haven’t already, and you must maintain a minimum of USD 100 in prepaid balance to enable AI summaries of open-ended text responses. Once enabled, your prepayment balance will be charged USD 0.15 per respondent included in analysis.

If you are not the primary owner of the experiment, you can still enable the feature, but the charges will be applied to the primary owner’s prepaid balance.

Sentiment analysis

Sentiment analysis helps you analyse respondents’ text responses to determine whether the dominant emotion is positive, negative, mixed, or neutral. The analysis is automatically included when performing an AI summary of text responses, and there are no additional charges for this functionality.

Interpreting sentiment analysis results for individual questions

For every open-ended question, the sentiment analysis outputs show the distribution of sentiments, which includes:

  • The average sentiment score of responses, ranges from the most negative (-1) to the most positive (1).
  • A coloured bar chart showing the percentage of responses by sentiment category, with green for positive sentiment, pink for negative sentiment, blue for neutral sentiment, and violet for the mixed sentiment.
The distribution of sentiments

The following information is also displayed for individual responses to short text, paragraph input, positive/negative open-ended feedback, and the ‘Other’ option for multiple choice and dropdown menu questions:

  • The prevailing sentiment determined by the natural language processing model
  • The sentiment score, ranges from -1 (indicating a 100% likelihood that the response has a negative sentiment) to 1 (indicating a 100% likelihood that the response has positive sentiment).

You can also filter the responses by sentiment using the dropdown menu.

The prevailing sentiment and sentiment score of responses

Interpreting sentiment analysis results for monadic blocks

For monadic blocks, the prevailing sentiment and the sentiment score are displayed for each stimulus in open-ended questions.

The distribution of sentiments

Using sentiment analysis results for segmentation

Follow these steps to create segmentation in your report using sentiment analysis results.

  1. In your report, navigate to Segmentation .
  2. Click on Add segment .
  3. On the pop-up, select Sentiment Analysis .
  4. Specify the question, item (only if applicable), operator, and sentiment from the dropdown menus.
Using sentiment analysis results for segmentation

FAQs

How do AI summaries of open-ended text responses work?

The AI summaries of open-ended text responses are powered by large language models operated by our subprocessors such as OpenAI and Anthropic. All responses are preprocessed to remove answers that do not contain valuable information, such as “Nothing”, “N/A”, and empty answers. Then, the AI-based model processes and summarises the content.

While large language models 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, AI’s exact way of “thinking” is not always clear and is a deep research field.

Conjointly encourages treating AI summaries as high-level summaries that take into consideration all text inputs with valuable information.

How are AI summaries of responses priced?

The price is USD 0.15 per respondent. The cost is irrespective of the number of questions. Attempts to make excessive use of the feature are subject to the acceptable use policy.

Why do I need to set up auto top-up to enable AI summaries of open-ended text responses?

The AI summaries are charged to your store credit (prepayment balance) on every report calculation. Setting up the auto top-up would ensure your store credit is always sufficient. You will also receive a 5% bonus on top of your subsequent prepayments.

How can I disable the AI summaries of open-ended text responses?

To disable summarisation of your existing reports, you can navigate to the Analytics options tabs under Advanced Settings of your experiment, and uncheck the box next to Summarise text responses. You will still retain access to the summaries of text responses generated for your existing reports. You are also able to re-enable the function at any time.

What languages are supported by the AI summaries of open-ended text responses functionality?

Currently, the functionality generates summaries in English only. It works well for all English and most non-English responses.

Why are the summaries different for the same responses?

A language model may provide different outputs when run multiple times on the same input. Thus, the summaries for the same responses may have different wordings but the same meaning.