Answer expectant mothers by SMS and escalate danger signs
Jacaranda Health's PROMPTS uses an AI help desk to triage mothers' text messages and refer risks fast.
Why it matters
Mothers often seek care too late. Two-way SMS reaches people on basic phones, and AI triage lets a small nurse team focus on the messages that signal danger.
How it works
Mothers receive SMS messages matched to their stage of pregnancy. An AI-enabled help desk reads, triages and answers their questions and starts a rapid referral chain when it spots a risk (Jacaranda Health). A peer-reviewed cluster randomised trial in 40 Kenyan health facilities with 6,139 women, published in PLOS Medicine in February 2025, describes the help desk sending automated guidance in roughly 86% of cases and escalating the rest to clinical agents who staff it 24/7. The trial found small improvements that reached statistical significance (0.06 to 0.09 standard deviations) in knowledge, birth preparedness, routine care seeking, newborn care and postpartum care content, and no statistically detectable effect on care seeking for danger signs. An AWS case study reports 3.8 million mothers reached, 10,000 to 12,000 questions received a day and that the model, Meta Llama 3, can respond to 70% of the questions PROMPTS receives. NPR reported in September 2026 that PROMPTS is used at public clinics in 24 Kenyan counties, with a team of 18 nurses monitoring and answering questions around the clock.
Who it's for
Ministries of health, public-health NGOs and maternal-care programmes.
The prompt
Act as a clinical-safety designer drafting materials for an SMS help line for pregnant women and new mothers in [country], in [languages]. Everything you produce is a draft: a qualified clinician must approve it, and native speakers must check the translations, before any real use. Using only the official danger-sign guidance I paste below, produce: 1. A triage table with three levels: emergency now, see a clinician within 24 hours, general advice. List the exact symptoms that trigger each level and quote the guidance line behind each one. 2. Reply templates under 300 characters for each level, in plain language, in each of my languages. Every emergency reply must tell the mother to go to the nearest health facility now and give [local emergency number]. 3. Escalation rules: which messages go to a human nurse immediately, and what summary the nurse sees. Never diagnose. When a message is ambiguous, escalate. List every situation where my guidance is silent so a clinician can fill the gap. [Paste national or WHO danger-sign guidance]
Open inClaude (opens in a new tab)ChatGPT (opens in a new tab)Gemini (opens in a new tab)Grok (opens in a new tab)
Sources
NPR reported independently (September 2026). The PLOS Medicine trial is peer-reviewed, with authors from Harvard, the World Bank and Innovations for Poverty Action, but Jacaranda Health staff are co-authors. AWS is a technology partner. Usage figures in all sources come from Jacaranda.
1. Jacaranda Health: Two-way SMS; AI-enabled help desk triages and responds; rapid referral chain. PROMPTS (opens in a new tab)
2. AWS: 3.8M mothers reached; 10,000 to 12,000 questions received a day; model can respond to 70% of questions; Meta Llama 3. Jacaranda Health case study (opens in a new tab)
3. PLOS Medicine (via PubMed Central): Peer-reviewed cluster RCT: 40 facilities, 8 counties, 6,139 women; AI-enabled help desk, about 86% automated guidance, escalation to clinical agents 24/7; modest gains; no statistically detectable effect on danger-sign care seeking; outcomes self-reported. Impact evaluation of a digital health platform empowering Kenyan women across the pregnancy-postpartum care continuum: a cluster randomized controlled trial (opens in a new tab)
4. NPR (All Things Considered): PROMPTS used in public clinics in 24 Kenyan counties; AI triages and flags urgent messages to nurses; 18 nurses monitor around the clock; about 2 US dollars per user; outside expert comment (Johns Hopkins). To prevent deaths in childbirth, Kenyan moms turn to an AI powered chatbot (opens in a new tab)