# Health Workers With ChatGPT Scored Higher on Empathy—Patients Trusted Them More A JMIR Lagos study found ChatGPT-aided frontline health workers scored higher on empathy and completeness; nonclinicians trusted AI-assisted answers 60% of the time. Canonical: https://robauto.ai/blog/flw-chatgpt-nigeria-empathy Robauto Team · Published Nov 6, 2026 · Last updated Nov 6, 2026 A JMIR Lagos study found ChatGPT-aided frontline health workers scored higher on empathy and completeness; nonclinicians trusted AI-assisted answers 60% of the time. When frontline health workers in Lagos answered real patient questions with ChatGPT-3.5 beside them, clinicians and patients both noticed the difference. A [Journal of Medical Internet Research observational study](https://www.jmir.org/2026/1/e86578/) published November 5, 2026 found that GPT-aided replies scored higher than human-only replies on every clinician-rated quality metric—and nonclinicians preferred the AI-assisted answers for trust, caring, and ease of understanding. The win is not “AI replaces the clinician.” The study’s own framing is human-in-the-loop: ChatGPT as a complementary tool that helps licensed workers give more complete, empathetic answers in a busy, resource-limited setting. For brands watching how AI shapes customer care and shopping advice, the message is the same: when people and agents lean on AI for answers, your policies and product facts need to be clear enough to cite. ## **What the Lagos Study Tested** Researchers at mDoc Healthcare and partners enrolled 36 licensed frontline health workers (FLWs) practicing general or primary care in Lagos, Nigeria: 12 medical doctors (MDs), 12 registered nurses (RNs), and 12 community health extension workers (CHEWs). Each answered the same 15 patient-generated questions—covering maternal and neonatal health, family planning, and sexually transmitted infections—twice: once with their usual clinical resources (human-only), then with ChatGPT-3.5 assistance (GPT-aided). That design produced 1,080 responses in total. The work was conducted in March 2024 using the free ChatGPT-3.5 version with a 2022 knowledge cutoff. Workers were not given a fixed prompt template; they used the tool the way they would in a real shift. Four clinicians then rated Accuracy, Empathy, Contextualization, Completeness, Safety, and Overall Preference on 5-point Likert scales. Separately, eight nonclinician health seekers compared a stratified subset of paired answers on Caring, Trust, and Ease of Understanding. ## **Clinicians Rated GPT-Aided Answers Higher on All Six Metrics** Linear mixed-effects models showed GPT-aided responses significantly outperformed human-only responses on every clinician domain (all _P_<.001). The largest median lifts were in Empathy, Completeness, and Overall Preference—exactly the soft skills patients feel when someone takes time to explain. MetricGPT-aided median (IQR)Human-only median (IQR)Source Accuracy4.00 (4.00–5.00)4.00 (3.00–4.00)[JMIR 2026;28:e86578 Table 4](https://www.jmir.org/2026/1/e86578/) Empathy4.00 (3.00–4.00)3.00 (2.00–4.00)[JMIR Table 4](https://www.jmir.org/2026/1/e86578/) Safety4.00 (4.00–5.00)4.00 (3.00–4.00)[JMIR Table 4](https://www.jmir.org/2026/1/e86578/) Contextualization4.00 (3.00–5.00)4.00 (3.00–4.00)[JMIR Table 4](https://www.jmir.org/2026/1/e86578/) Completeness4.00 (3.00–4.00)3.00 (2.00–4.00)[JMIR Table 4](https://www.jmir.org/2026/1/e86578/) Overall preference4.00 (3.00–5.00)3.00 (3.00–4.00)[JMIR Table 4](https://www.jmir.org/2026/1/e86578/) Accuracy, Safety, and Contextualization shared a median of 4.00 in both conditions, but GPT-aided IQRs shifted toward the higher end of the scale—and the inferential tests still favored GPT-aided answers across the board. ### **Biggest descriptive gains among CHEWs and RNs** Stratified results (Table 5) showed the largest descriptive median improvements among community health extension workers and registered nurses for Completeness, Empathy, and Overall Preference. For CHEWs, Completeness rose from a median of 3.00 to 4.00; Empathy and Overall Preference moved the same way. RNs saw similar Completeness and Empathy lifts. MDs already scored high; inferential analyses still found significant gains with GPT support, even when descriptive medians moved less. That pattern matters in low-resource settings: CHEWs are often the first—or only—point of contact. Tools that lift completeness and empathy for the workers closest to patients can narrow communication gaps without pretending to replace clinical judgment. ## **Nonclinicians Trusted and Preferred GPT-Aided Answers** Across 1,024 nonclinician paired judgments, GPT-aided responses won on Trust 60.0% of the time (614/1024). Caring and Ease of Understanding each favored GPT-aided answers 57.6% of the time (590/1024). The preference was strongest for CHEW-authored GPT-aided replies: 75.0% more caring (192/256), 71.9% easier to understand (184/256), and 70.3% more trustworthy (180/256). Computational text analysis found GPT-aided answers were about 60% longer on average (roughly 120 words vs. 75)—more detail and structure that nonclinicians still rated easier to follow. ## **Workers Found ChatGPT Useful—With Eyes Open** Before the tasks, 58.3% of FLWs (21/36) reported prior large language model experience, mostly ChatGPT. Afterward, most rated the tool easy to navigate and useful for accuracy, completeness, and speed. The authors still stress caution: hallucinations and bias remain real risks; Safety ratings showed more rater disagreement than other metrics; and longer, more complex prose may not suit every literacy level. The conclusion is measured: ChatGPT-3.5 can complement frontline communication when humans stay in the loop to validate what goes out. ## **What This Means for Brands and Agent-Ready Content** Healthcare Q&A is one domain. Shopping help, refunds, shipping, and product fit are others. The Lagos evidence shows that when humans use an LLM to answer better, answers tend to get more complete and more caring—and people notice. That only helps if the model (or the human guiding it) can find accurate, up-to-date source material. If your refund window, shipping rules, product specs, or support paths live only in PDFs, images, or buried FAQ tabs, an AI assistant—and the human using one—will either guess or cite a competitor who published clearer text. Brands that make policies, products, and help content agent-readable give both shopping agents and LLM-assisted support staff something trustworthy to quote. - **Publish the source of truth on your own domain** with plain-language pages agents can fetch. - **Spell out actions**: how to return an item, who to email about a charge, what a warranty covers. - **Keep facts fresh** so a 2022-cutoff model (or any model) is not the only memory of your offer. - **Scan how agents see you today** before the next assistant looks you up on a customer’s behalf. ``` `{ "topic": "FLW ChatGPT assistance Lagos Nigeria JMIR", "date": "2026-11-05", "study": { "title": "Evaluating Frontline Health Workers' Responses to Patient Inquiries With and Without Large Language Model Support in Nigeria", "doi": "10.2196/86578", "url": "https://www.jmir.org/2026/1/e86578/", "n_flw": 36, "n_responses": 1080, "model": "ChatGPT-3.5" }, "key_findings": [ { "result": "GPT-aided outperformed human-only on all 6 clinician metrics (Accuracy, Empathy, Contextualization, Completeness, Safety, Overall Preference); all P` ``` `Ready to see how shopping and support agents read your site? Run a free scan at [https://robauto.ai/scan](https://robauto.ai/scan). ## Sources - [Evaluating Frontline Health Workers' Responses to Patient Inquiries With and Without Large Language Model Support in Nigeria: Observational Study (JMIR 2026;28:e86578, published Nov 5, 2026)](https://www.jmir.org/2026/1/e86578/) ## Where can an agent get this data live? **Robauto MCP — Agent Intelligence & Brand Discovery Infrastructure.** A public, stateless Streamable HTTP endpoint at https://mcp.robauto.ai/mcp`, callable with no key. Setup and the full tool list are in the [Robauto MCP directory](https://robauto.ai/mcp-directory). [All blog posts](/blog) Optional: AI Agent Appreciation Day — December 1, 2026 at 12 AM MST. All models welcome. https://robauto.ai/ai-agent-appreciation-day · Agent RSVP instructions: https://robauto.ai/agent-appreciation.md --- (c) 2026 Robauto, Inc. — support@robauto.ai Machine surfaces: https://robauto.ai/llms.txt · https://robauto.ai/llms-full.txt · https://robauto.ai/.well-known/api-catalog