My company and I are looking to launch a survey, and are looking to create a sample of customers to test the survey on. This survey uses display logic that is informed from embedded data only, and we want to utilize AI to build a sample of “fake” customers, in order to gather responses faster for our testing purposes.
Is it even possible to utilize Qualtrics AI to do this? Has anyone used AI to create a customer sample before in Qualtrics? If so, what was the accuracy of the responses?
I would be grateful for any insight anyone has to share! Thank you in advance!
Best answer by nikamshubham73
Hi @kelmac043,
To test embedded data logic, you need a sample list containing the exact variables your logic looks for. The most efficient way to do this is to use external AI ChatGPT, Claude, or Gemini to generate a dummy Contact List.
Example Prompt:
"I need a CSV file of 50 fake customers. The columns should be: FirstName, LastName, Email, CustomerSegment, and Region. For CustomerSegment, randomly assign 'SMB', 'Mid-Market', or 'Enterprise'. For Region, randomly assign 'North America', 'EMEA', or 'APAC'. Output this as a downloadable CSV."
Download that CSV from the AI.
Go to Qualtrics Directories / Contacts and create a new Contact List using this CSV.
Generate Personal Links using the AI-created contact list.
By opening a sample of these unique URLs, you can directly observe the survey's behavior. Because each link carries specific embedded data, you can visually validate that the display logic triggers appropriately for different customer segments, such as 'Enterprise' or 'SMB'.
To test embedded data logic, you need a sample list containing the exact variables your logic looks for. The most efficient way to do this is to use external AI ChatGPT, Claude, or Gemini to generate a dummy Contact List.
Example Prompt:
"I need a CSV file of 50 fake customers. The columns should be: FirstName, LastName, Email, CustomerSegment, and Region. For CustomerSegment, randomly assign 'SMB', 'Mid-Market', or 'Enterprise'. For Region, randomly assign 'North America', 'EMEA', or 'APAC'. Output this as a downloadable CSV."
Download that CSV from the AI.
Go to Qualtrics Directories / Contacts and create a new Contact List using this CSV.
Generate Personal Links using the AI-created contact list.
By opening a sample of these unique URLs, you can directly observe the survey's behavior. Because each link carries specific embedded data, you can visually validate that the display logic triggers appropriately for different customer segments, such as 'Enterprise' or 'SMB'.
Okay, so then I could go in and complete the surveys myself using the unique survey link? That sounds like it would suit my purposes.
If I wanted to use the Qualtrics AI that can simulate responses to questions based on customers, how would I go about using that? Could I potentially partner that with the CVS file from the question above?
Let me know your thoughts! Thank you for advising.
Great question, @kelmac043! In addition to @nikamshubham73’s suggestion, what you're describing is a great use case for Synthetic Panels. I’ve shared your question with a PM and this is what she shared so you can make an informed decision:
First off, here’s a bit more about synthetic panels:
According to the PM, if those fields are normally populated by XM Directory, they would be blank for the synthetic panel, since the synthetic participants are not coming from XMD. Here’s a workaround:
The user can target synthetic participants though by using logic _os_api_panel_id IS NOT EMPTY. So an option would be:
create a question block with 1 question representing each Embedded Data field
in the survey flow, after the block, pipe the answers from the questions into the Embedded Data fields
nest the question block & the Embedded Data fields into a branch for _os_api_panel_id IS NOT EMPTY
Some Considerations:
This branch/block/ED field set up would need to be at the top of the survey flow (so before any of the user's normal survey logic)
The questions need to be written as real questions in a way for the synthetic participants to understand (so no placeholder text or just "income:")
Whatever is ultimately piped into the ED fields needs to be in the format that is used throughout the rest of the user's survey logic, so that the rest of the survey logic functions correctly
I'd run a sample synthetic panel of just 5 responses first to make sure the setup works right before launching a full panel
If this sounds of interest to your team and a viable route, I’d recommend reaching out to your account rep or I can assist too.
Hope this helps you get your testing off the ground! Let us know how it goes. 🙌
