An interventional study of Scripted maternal health chatbot and MATRI retrieval-augmented maternal health chatbot in Pregnancy, Prenatal Care and Maternal Health, sponsored by Nivi, Inc.. Not yet recruiting at 1 site in India. Open to female participants aged 18 Years and older, including healthy volunteers. Per ClinicalTrials.gov, last updated 2026-10-20.
Sponsored by Nivi, Inc. · Not applicable, Interventional, and Health services research
Many pregnant women in India use mobile phones to get health information. Nivi runs a free WhatsApp service called askNivi that answers questions about pregnancy and sends reminders about prenatal care visits.
The current version of askNivi replies using a set of prepared answers. This study tests a new version that uses artificial intelligence to write replies. The new version looks up information in a reviewed collection of maternal health materials before answering, so its replies stay grounded in checked sources. It is also designed to recognize symptoms that may signal a pregnancy complication and to tell the user to seek care right away.
Women who join the study are assigned at random to one of the two versions. Neither version replaces medical care, and both send the same weekly messages and clinic reminders. Women can ask questions at any time throughout their pregnancy.
The study will look at how much women use each version, how useful they find it, how they decide when to seek care, and whether they attend prenatal care visits. About 2,500 pregnant women in Assam, Madhya Pradesh, and Uttar Pradesh will take part. Participation lasts from enrollment, between 10 and 20 weeks of pregnancy, until delivery.
Background
Maternal health remains a critical public health challenge in India, with maternal mortality ratios at 96 per 100,000 live births as of 2020. While this represents a significant improvement from 254 per 100,000 in 2006, partly attributed to community health workers (Accredited Social Health Activists or ASHAs) using WhatsApp for health communication, substantial gaps remain in providing timely, accurate maternal health information to pregnant women.
Digital health interventions, particularly chatbots, offer promising solutions for delivering scalable, cost-effective maternal health support. Chatbots can provide 24/7 access to health information, appointment reminders, and emotional support without requiring additional healthcare infrastructure. However, current evidence reveals a critical limitation: engagement with health chatbots typically declines dramatically over time, with retention rates falling from 72% at 2 months to 31% by 8 months.
Most existing maternal health chatbots use scripted, rule-based approaches with predetermined responses. While these systems are reliable and predictable, they often fail to address users' specific questions, leading to frustration and disengagement. Recent advances in Artificial Intelligence (AI), specifically Retrieval-Augmented Generation (RAG) technology, offer a potential solution. RAG enhances Large Language Models (LLMs) by grounding their responses in retrieved information from validated knowledge bases, potentially providing more relevant, personalized responses while reducing the risk of AI hallucinations.
Despite the emergence of RAG technology, no published studies have directly compared RAG-enhanced chatbots with scripted versions in any health context, let alone maternal health. This represents a critical knowledge gap, as RAG systems could potentially maintain higher engagement through more contextually relevant responses while ensuring medical accuracy through grounded information retrieval.
The askNivi platform provides an ideal setting for this comparison. As an established WhatsApp-based chatbot serving pregnant women across India, askNivi has demonstrated feasibility and acceptability. WhatsApp's penetration in India exceeds 500 million users, making it an optimal platform for reaching diverse populations, including those in rural and resource-limited settings.
Design
Two-arm, parallel-group, individually randomized controlled trial conducted on the askNivi WhatsApp platform in India. Allocation is 1:1 using computer-generated permuted block randomization with variable block sizes of 4, 6, and 8, stratified by language preference (Hindi, English, or Assamese). Allocation is automated following consent and completion of the baseline survey. Target enrollment incorporates 20 percent expected attrition.
Arms
Control: the scripted askNivi chatbot currently in deployment. Responses are drawn from prepared content selected by keyword matching, within fixed conversation flows covering predetermined topics.
Intervention: a retrieval-augmented chatbot that classifies incoming user messages, escalates messages indicating a possible clinical emergency to urgent care-seeking guidance, and generates responses to other messages from a curated maternal health corpus. Safety guardrails constrain generated content.
Both arms receive identical weekly template messages and antenatal care visit reminders.
Emergency detection
The intervention arm screens user messages for symptoms that may indicate a pregnancy complication, including severe or persistent headache, unusual swelling of the hands or face, vaginal bleeding, reduced fetal movement, and signs of preeclampsia. On detection, the chatbot advises the user to seek care immediately at her enrolling facility. The chatbot does not make facility appointments or contact providers directly. The control arm provides general health information without active symptom screening.
Setting
Three states where askNivi is deployed through the e-SAATHI initiative: Assam (six districts, with Assamese, Hindi, and English available), Madhya Pradesh (multiple blocks, Hindi and English), and Uttar Pradesh (Varanasi district, Hindi and English). Participation is entirely remote. Data are held on servers in the Mumbai region.
Pilot phase
A pilot precedes full enrollment, testing the randomization mechanism, consent and survey administration over WhatsApp, and system performance across all three languages. Pilot participants are drawn from the same population and pilot data are excluded from the analyses described here.
Analysis
The primary analysis follows intention to treat. Engagement outcomes are measured weekly within participant and analyzed with methods appropriate to repeated measures, accounting for correlation within participant. Secondary analyses adjust for baseline characteristics. An interim review at 50 percent enrollment addresses safety and futility only, with no efficacy stopping rule. Further detail is given in the statistical analysis plan.
This is the only study on the registry with Nivi, Inc. as lead sponsor.
Counted across the registry records on this site, refreshed daily.
Go through the criteria one by one — yes, no or not sure — and take the list to your doctor or the study team. Kept in this browser only.
Exclusion Criteria:
Participants receive the askNivi chatbot version currently in deployment. Responses are drawn from prepared content selected by keyword matching, within fixed conversation flows covering predetermined topics. Participants receive weekly template messages and antenatal care visit reminders.
Behavioral: Scripted maternal health chatbot
Participants receive a retrieval-augmented version of the askNivi chatbot that generates responses in real time from a curated maternal health corpus, can address questions outside predetermined topics, screens messages for symptoms indicating a possible pregnancy complication, and advises urgent care-seeking when such symptoms are detected. Participants receive the same weekly template messages and antenatal care visit reminders as the comparator arm.
Behavioral: MATRI retrieval-augmented maternal health chatbot
WhatsApp-delivered chatbot providing maternal health information through prepared responses selected by keyword matching, weekly template messages, and antenatal care visit reminders. Available in Hindi, English, and Assamese.
MATRI (Maternal-health Agent for Triage and Retrieval-grounded Interaction) is a WhatsApp-delivered chatbot that classifies incoming user messages, escalates messages indicating a possible clinical emergency to urgent care-seeking guidance, and generates responses to other messages from a curated maternal health corpus. Safety guardrails constrain generated content. Delivers the same weekly template messages and antenatal care visit reminders as the comparator. Available in Hindi, English, and Assamese.
Percentage of weeks with any user interaction
Binary measure of whether the user gave any response during the specified week. Measured from platform interaction logs.
Time frame: Weekly from enrollment through delivery or gestational week 41, up to 31 weeks
Mean conversational turns per week
Number of user-initiated messages sent to the chatbot in a given week, averaged across the participant's follow-up weeks. Measured from platform interaction logs.
Time frame: Weekly from enrollment through delivery or gestational week 41, up to 31 weeks
Mean number of substantive user queries per week
Number of substantive queries sent by the participant in a given week, averaged across her follow-up weeks. A substantive query is a user-initiated free-text message, excluding greetings, acknowledgments, single-word replies, emoji-only messages, system strings, and blanks. The same classification procedure is applied in both arms. Measured from platform interaction logs.
Time frame: Weekly from enrollment through delivery or gestational week 41, up to 31 weeks
Percentage of weeks with any substantive user query
Binary measure of whether a participant gave any substantive query during the given week, defined as above. Measured from platform interaction logs.
Time frame: Weekly from enrollment through delivery or gestational week 41, up to 31 weeks
Percentage of weeks with user response to weekly template message
Binary measure of whether the user responded to the template message of the specified week
Time frame: Weekly from enrollment through delivery or gestational week 41, up to 31 weeks
Discernment index from user responses to clinical scenarios
Score from 0 to 2 across a paired scenario battery. Each participant sees one high-severity and one low-severity scenario at baseline and the other pair at endline, so no scenario is answered twice. Each scenario scores 1 for the keyed action and 0 otherwise. High-severity and low-severity accuracy are also reported separately.
Time frame: Baseline and gestational weeks 38 to 41, up to 31 weeks
Self-reported prenatal care attendance and health-seeking behaviors
Total antenatal care visits during the pregnancy, self-reported by the participant. Visits since enrollment are reported at midline and total visits for the pregnancy at endline.
Time frame: From enrollment through delivery or gestational week 41, up to 31 weeks
Self-reported acute care-seeking
The share of reported worrying symptoms for which the participant sought care the same day. Collected through between-survey pulse checks and the midline questionnaire.
Time frame: Every 6 to 8 weeks from enrollment through delivery or gestational week 41, up to 31 weeks
Usability of the chatbot, adapted System Usability Scale
Mean score across a 5-item usability instrument adapted from the System Usability Scale, with items rated from 1 (strongly disagree) to 5 (strongly agree). Possible range 1 to 5. Higher scores indicate better perceived usability. Negatively worded items are reverse-scored before averaging.
Time frame: Gestational weeks 28 to 30, up to 20 weeks after enrollment
Net Promoter Score
Single survey item asking how likely the participant is to recommend the chatbot to a pregnant friend or family member, rated from 0 (not at all likely) to 10 (extremely likely). Higher scores indicate greater likelihood to recommend.
Time frame: Gestational weeks 38 to 41 or within one week of delivery, up to 31 weeks
Change in knowledge self-efficacy
Single survey item, "I feel knowledgeable about my pregnancy," rated from 1 (strongly disagree) to 5 (strongly agree). Higher scores indicate greater self-efficacy. Analyzed as change from baseline.
Time frame: Baseline, gestational weeks 28 to 30, and gestational weeks 38 to 41, up to 31 weeks
Retention assessed as time to last user interaction
Number of weeks from enrollment to the participant's last interaction with the chatbot. Disengagement is defined as no user interaction of any type for four or more consecutive weeks, with the event dated to the last week containing an interaction. Participants are censored at delivery where the delivery date is known, and at gestational week 41 otherwise.
Time frame: From enrollment to gestational week 41, up to 31 weeks
Plan to share: Yes — The Principal Investigator will make all raw individual participant data files, together with statistical analysis protocol descriptions and the information needed for proper analysis, available for download a minimum of one year and a maximum of three years after publication of the corresponding manuscripts. All human subjects data will be anonymized prior to analysis, so participant confidentiality will not be compromised by this plan. All hypotheses, methods, and planned analyses are preregistered on ClinicalTrials.gov and AsPredicted.
Supporting information: Study protocol, Statistical analysis plan, Informed consent form, Analytic code
Time frame: All raw files, statistical analysis protocol descriptions, and information for proper analysis will be available for download a minimum of one year and a maximum of three years after publication of the corresponding manuscripts.
No publications or documents are linked to this record.
From the registry record's own update history. This site started tracking changes on Sep 25, 2026; for anything earlier, see the record history on ClinicalTrials.gov ↗
This study is not yet recruiting, as verified in Sep 2026. You cannot join it, but the record below documents what was studied.
Get an email when the registry record changes — status, dates, results — or when someone posts here.
Sign in to followQuestions and observations about this study, from anyone following it. Not medical advice, and not a channel to the study team — their contact details are on the registry record.
Sign in to join the discussion. Reading takes no account; posting does. You choose a display name, and a pseudonym is the default.
Nothing here yet. If you are running this trial, taking part in it, or weighing whether to, this is the place to say so.