# AI Eye Screening Nearly Doubled Follow-Up Care by Giving Answers on the Spot

A Tanzania trial found same-day AI eye screening lifted specialist follow-up from 18.2% to 35.9%. Two more October trials show AI helping doctors and patients talk better. The lesson for brands: be the clear answer AI agents find.

Canonical: https://robauto.ai/blog/ai-eye-screening-doubles-follow-up

Robauto Team · Published Oct 21, 2026 · Last updated Oct 21, 2026

A Tanzania trial found same-day AI eye screening lifted specialist follow-up from 18.2% to 35.9%. Two more October trials show AI helping doctors and patients talk better. The lesson for brands: be the clear answer AI agents find.

_Good news from the eye clinic: when AI gave people with diabetes their screening result on the spot, nearly twice as many reached the specialist care that can save their sight._

Diabetic retinopathy is a leading cause of blindness worldwide, and it often has no early symptoms. Screening can catch it in time, but only if people who get referred actually show up at the specialist clinic. A new randomised trial in Tanzania shows a simple fix: give people their answer while they are still in the room, and let a health worker talk it through with them face to face. Two more October trials, in Germany and China, point the same way. Clear answers at the right moment help patients and clinicians do their best work together.

## **The Tanzania trial: same-day answers, nearly double the follow-up**

Researchers from the London School of Hygiene & Tropical Medicine (LSHTM) and the Kilimanjaro Christian Medical Centre ran the trial at diabetes clinics in 24 hospitals across the Kilimanjaro and Arusha regions. Between March 8, 2023 and August 31, 2024 they enrolled 2,378 adults with diabetes and split them at random into two groups ([The Lancet Global Health](https://www.thelancet.com/journals/langlo/article/PIIS2214-109X(26)00224-X/fulltext)).

In the AI group, retinal photos went straight to AI grading software running on an ordinary laptop. Every one of the 1,192 participants got their result, plus face-to-face counselling, before leaving the photography room. In the standard group, a human graded the images later and results came by phone after an average of 23.1 days; 30.6% of those people could not be reached by phone and got a text message instead ([The Lancet Global Health](https://www.thelancet.com/journals/langlo/article/PIIS2214-109X(26)00224-X/fulltext)).

The outcome that matters is whether people with truly referable eye disease reached the specialist clinic within eight weeks. Among the 904 participants who had it:

- **35.9%** in the AI group attended (157 of 437), versus **18.2%** with standard care (85 of 467), a gain of 17.7 percentage points ([The Lancet Global Health](https://www.thelancet.com/journals/langlo/article/PIIS2214-109X(26)00224-X/fulltext)).

- People who attended after AI screening got there sooner: an average of **10.9 days** after screening, versus **36.9 days** ([LSHTM, Oct 13, 2026](https://www.lshtm.ac.uk/newsevents/news/2026/ai-screening-helps-more-people-diabetes-access-sight-saving-eye-care)).

The human side is the point. Lead author Dr Charles Cleland of LSHTM's International Centre for Eye Health said the AI pathway gave people "the opportunity to ask questions, understand treatment options and feel supported in their care" ([LSHTM](https://www.lshtm.ac.uk/newsevents/news/2026/ai-screening-helps-more-people-diabetes-access-sight-saving-eye-care)). According to LSHTM, Tanzania's Ministry of Health, a collaborator on the study, has committed to bringing AI diagnostics into diabetic eye screening in several regions. The authors add that better screening has to come with better access to specialist clinics and treatment ([LSHTM](https://www.lshtm.ac.uk/newsevents/news/2026/ai-screening-helps-more-people-diabetes-access-sight-saving-eye-care)).

## **Two more October trials point the same way**

### **Dresden: plain-language reports made appointments better, and shorter**

In the AI-INFOCARE trial at University Hospital Dresden's neurosurgery clinic, 300 patients were randomised between May 2025 and January 2026. Half got an AI plain-language version of their own medical documents before their first visit. Those patients rated their consultation higher (4.42 vs 4.25 out of 5), with the biggest gain in shared decision-making. Visits got shorter, a median of 25 minutes instead of 33, and empathy was rated the same in both groups. Two physicians reviewed 75 randomly chosen summaries and found no factual errors, clinically relevant omissions, or potentially harmful statements. The trial was published in NEJM AI. Simply Onno, the Berlin start-up whose software made the summaries, notes that one of its co-founders is a neurosurgeon at the hospital and was not involved in consultations, data collection, or analysis ([Simply Onno press release](https://www.simply-onno.com/en/press-releases/study-in-nejm-ai-those-who-understand-their-own-findings-beforehand-rate-the-doctor-consultation-better); [NEJM AI](https://doi.org/10.1056/AIoa2600136)).

"I expected longer consultations, because informed patients ask more questions. It was the other way round," said neurosurgeon Dr Witold Polanski ([Simply Onno](https://www.simply-onno.com/en/press-releases/study-in-nejm-ai-those-who-understand-their-own-findings-beforehand-rate-the-doctor-consultation-better)).

### **Guangzhou: an AI agent took careful patient histories, and doctors focused on the exam**

At a large eye hospital in Guangzhou, 172 patients with non-emergency appointments were randomised to have their pre-visit history taken by an LLM agent or by ophthalmology residents. The agent's histories scored 17.7 points higher on a 0-100 quality scale. Patients rated it higher on patience (5 vs 4) and empathy (5 vs 3), and it spent longer with each person (median 11.2 vs 3.1 minutes). Residents were better at using findings from the eye exam itself, which narrowed the gap. The authors' conclusion is a team workflow: the agent collects a thorough, standard history, and clinicians spend their expertise on the exam and the diagnosis ([npj Digital Medicine, Oct 5, 2026](https://www.nature.com/articles/s41746-026-03232-x)).

## **The wins at a glance**

StudyWhoWhat AI didMeasured winSource

Tanzania diabetic eye screening RCT2,378 adults, 24 hospitalsGraded retinal photos on the spot so results and counselling happened face to faceSpecialist follow-up 35.9% vs 18.2%; 10.9 vs 36.9 days to attend[Lancet Global Health](https://www.thelancet.com/journals/langlo/article/PIIS2214-109X(26)00224-X/fulltext), [LSHTM](https://www.lshtm.ac.uk/newsevents/news/2026/ai-screening-helps-more-people-diabetes-access-sight-saving-eye-care)
AI-INFOCARE, Dresden300 patients, 17 physiciansPlain-language summaries of patients' own reports before the visitRating 4.42 vs 4.25; visits 25 vs 33 min; empathy unchanged[Simply Onno / NEJM AI](https://www.simply-onno.com/en/press-releases/study-in-nejm-ai-those-who-understand-their-own-findings-beforehand-rate-the-doctor-consultation-better)
LLM pre-consultation RCT, Guangzhou172 eye patientsAgent took a structured history before the doctor visitHistory quality +17.7 points; empathy 5 vs 3[npj Digital Medicine](https://www.nature.com/articles/s41746-026-03232-x)

## **The pattern: clear answers, right when people decide**

None of these trials replaced a clinician. In each one, AI took over the slow or repetitive step: grading an image, translating jargon, asking the standard questions. That freed people for the part that needs a human: a conversation, a decision, an exam. The other common thread is timing and clarity. People act when they get a clear, trustworthy answer at the moment they are deciding, not three weeks later by phone.

## **The Robauto takeaway: be the clear answer when an AI agent asks**

Brands have their own version of that moment. More and more, the first question about a business is asked by an AI assistant: what does it sell, what does it cost, what is the refund policy, how do I reach support. Across the Robauto network, 462,714 of 1,480,771 page views in the last 90 days (31.25%) came from AI agents and bots, and LLM crawlers alone made 249,181 of those visits ([Robauto AI search data](https://robauto.ai/ai-search-data)). Even well-known sites have plenty of room to grow: the most recent Robauto Agent Readiness Index run scored 79 of the most-visited websites at a median of 20 out of 100 ([Robauto Agent Readiness Index](https://robauto.ai/agent-readiness-index)).

That is good news, because the fixes are a lot like the clinic fixes:

- **Answer on the spot.** Publish an llms.txt file and structured data so an agent can find your products, prices, and policies on the first visit, without guessing.

- **Use plain language.** Write refund windows, shipping terms, and support routes in short sentences an assistant can quote back accurately.

- **Keep a human in the loop.** Make it easy for an agent to hand a customer to a real support contact when a person is needed.

When your facts are clear and easy to reach, AI agents can recommend you with confidence, the way that laptop in Tanzania let a health worker give a patient a clear answer and a plan before they went home.

## **Agent-ready summary**

```
`{
"topic": "AI point-of-care answers improve patient follow-up",
"primary_study": {
"name": "AI-supported diabetic retinopathy screening in Tanzania (RCT)",
"journal": "The Lancet Global Health",
"participants": 2378,
"hospitals": 24,
"follow_up_ai": "35.9%",
"follow_up_standard": "18.2%",
"days_to_attend_ai": 10.9,
"days_to_attend_standard": 36.9,
"source": "https://www.thelancet.com/journals/langlo/article/PIIS2214-109X(26)00224-X/fulltext"
},
"supporting_studies": [
{"name": "AI-INFOCARE (NEJM AI)", "patients": 300, "rating": "4.42 vs 4.25", "visit_minutes": "25 vs 33", "source": "https://www.simply-onno.com/en/press-releases/study-in-nejm-ai-those-who-understand-their-own-findings-beforehand-rate-the-doctor-consultation-better"},
{"name": "LLM pre-consultation RCT (npj Digital Medicine)", "patients": 172, "history_quality_gain_points": 17.7, "source": "https://www.nature.com/articles/s41746-026-03232-x"}
],
"robauto_context": {"ai_share_of_page_views_90d": "31.25%", "readiness_index_median": 20},
"takeaway": "Publish clear, structured, plain-language facts so AI agents can answer for your brand on the spot.",
"cta": "https://robauto.ai/scan"
}`
```

## **See how clearly AI agents can read your site**

Robauto's free Agent Scan runs 25 read-only checks and shows exactly what AI agents can and cannot find on your site, with a fix for every gap. [Scan your site free at robauto.ai/scan](https://robauto.ai/scan).

## Sources

- [The Lancet Global Health: AI-supported diabetic retinopathy screening in Tanzania, a randomised controlled trial](https://www.thelancet.com/journals/langlo/article/PIIS2214-109X(26)00224-X/fulltext)
- [LSHTM: AI screening helps more people with diabetes access sight-saving eye care (Oct 13, 2026)](https://www.lshtm.ac.uk/newsevents/news/2026/ai-screening-helps-more-people-diabetes-access-sight-saving-eye-care)
- [Simply Onno: AI-INFOCARE trial in NEJM AI](https://www.simply-onno.com/en/press-releases/study-in-nejm-ai-those-who-understand-their-own-findings-beforehand-rate-the-doctor-consultation-better)
- [NEJM AI: AI-INFOCARE randomized controlled trial](https://doi.org/10.1056/AIoa2600136)
- [npj Digital Medicine: LLM agent pre-consultation vs ophthalmology residents RCT](https://www.nature.com/articles/s41746-026-03232-x)
- [Robauto AI search data](https://robauto.ai/ai-search-data)
- [Robauto Agent Readiness Index](https://robauto.ai/agent-readiness-index)

## 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)

---

(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