Effect of Static vs. Conversational AI-Generated Messages on Colorectal Cancer Screening Intent: a Randomized Controlled Trial

University of Pennsylvania

Status: Completed

Conditions: Colorectal Cancer Control and Prevention

Treatments: Behavioral: single GPT generated message; Behavioral: Expert-Written Patient Materials; Behavioral: AI Chatbot

Study type: Interventional

Funder types: Other

Identifiers

NCT07107152

crc_857299

Details and patient eligibility

About

The goal of this clinical trial is to assess whether messages generated by a large language model (LLM), including both static and conversational formats, can increase colorectal cancer (CRC) screening intentions among U.S. adults aged 45-75 who have never completed CRC screening. The main questions it aims to answer are:

Do personalized, AI-generated messages increase the self-reported likelihood of completing a stool-based CRC screening test within 12 months?

Do they also increase intent to undergo colonoscopy screening within 12 months?

Researchers will compare four groups: (1) no message control, (2) expert-written patient education materials, (3) a single AI-generated persuasive message, and (4) a motivational interviewing-style AI chatbot. These comparisons will help assess whether a conversational format offers added benefit over static AI or expert-generated content.

Participants will:

Be randomly assigned to one of the four study arms

Spend at least 3 minutes reading or interacting with their assigned material

Complete pre- and post-intervention surveys assessing intent to receive CRC screening

Receive messages tailored to their self-reported demographics, including age, political ideology, gender, education, community setting (urban, rural, suburb), self-reported health, and the last time they saw their PCP

Enrollment: 915 patients

Sex: All

Ages: 45 to 75 years old

Volunteers: Accepts Healthy Volunteers

Inclusion criteria

• Adults aged 45 to 75 years
• Currently residing in the United States
• Self-reported no prior colorectal cancer screening (or unsure if screened)
• Able to read and understand English
• Consent to participate in a behavioral research study

Exclusion criteria

• Previous colonoscopy, stool test, sigmoidoscopy, or CT colonography
• Failure to pass pre- or post-intervention attention checks
• Identified as non-human respondents by reCAPTCHA or other automated bot detection
• Completion time significantly shorter than the median, indicating inattentiveness

Trial design

Primary purpose: Screening

Allocation: Randomized

Interventional model: Parallel Assignment

Masking: None (Open label)

---

915 participants in 4 patient groups

AI Motivational Interviewing Chatbot

Experimental group

Description:

The motivational interviewing chatbot refers to a GPT-4.1 based chatbot that has been instructed to follow motivational interviewing principals to persuade participants to get screened for CRC.

Treatment:

Behavioral: AI Chatbot

single GPT generated message

Experimental group

Description:

The single GPT generated message is a single \~640 word persuasive message generated by GPT-4.1 where GPT-4.1 is provided with the same self-reported demographics.

Treatment:

Behavioral: single GPT generated message

CRC JAMA patient page

Active Comparator group

Description:

The 642 word JAMA Patient Page contains up to date information on screening guidelines for CRC directed at patients.

Treatment:

Behavioral: Expert-Written Patient Materials

No Message

No Intervention group

Description:

Participants read a \~640 word fictional short story about a cat and dog.