AI systems can now read continuous glucose data and actually do something with it, like suggest insulin doses or adjust them automatically through closed-loop systems. That means less guessing, less delay, and care that adapts to what’s happening right now, not what was true six months ago.
This shift matters as it is one of the first times individuals living with diabetes will feel a different kind of support. One that listens better, responds faster, and fits the reality of each person’s body and lifestyle.
Want all the details? Grab your free PDF!
From Static Rules to Adaptive Insulin Decisions
For years, insulin management mostly relied on paper correction scales and the occasional clinic visit. It was a system that didn’t leave much room for real-life changes, like a stressful week, a skipped meal, or getting sick. Even when doses were updated by a doctor, those changes often came too late to match what was really going on.
That’s where AI is offering support. Instead of reacting to one-off glucose readings, it looks at the bigger picture, spotting patterns over time and adjusting insulin recommendations to match what your body actually needs.1,2
Some of this happens through clinical decision support systems (CDSS), which act as a kind of digital assistant for healthcare teams. They pull from your medical records and consider things like your weight, the type of diabetes you have, whether you're in hospital, and how much insulin you've needed in the past. Based on that, the software suggests personalized doses, including for meals, background insulin, and corrections.
Trials show these tools can often help lower average glucose levels and improve overall blood sugar control. That said, results depend on how the systems are used, and not every study shows the same outcome.3,4
Making Glucose Data Work Harder: AI and Continuous Monitoring
As AI tools get better at tailoring insulin doses, they also depend on better data. And that’s where continuous glucose monitoring (CGM) finds its role. Unlike fingersticks, CGMs track glucose levels 24/7, creating a stream of information that shows how your blood sugar responds to real life: meals, movement, stress, sleep. It’s detailed, dynamic and sometimes overwhelming.
Raw CGM data already helps people notice patterns. But AI takes it further. Algorithms can detect trends, spot risks, and give clearer feedback in real time. That means more useful insights about things like glucose variability, time spent in range, and early warning signs of highs or lows.
Right now, AI supports CGM systems in a few key ways:
- Closed-loop insulin control: Syncing with pumps to adjust insulin automatically
- Short-term glucose predictions: Forecasting where levels are heading
- Sensor calibration: Improving accuracy with less manual input
Prediction models are especially useful. They use deep learning to anticipate blood sugar changes up to several hours ahead, giving people time to adjust a dose, eat something, or pause activity before problems hit.5,6
For those in early stages of dysglycemia or prediabetes, AI-enabled CGMs can also offer lifestyle nudges or medication reminders, reaching beyond just those on insulin therapy.5,6
And we’re already seeing this in commercial tools. Roche’s Accu-Chek SmartGuide, now CE-marked in Europe, uses onboard AI to analyze glucose levels every five minutes and flag potential future changes based on detected patterns.7
Smart Support for People Using Injections, Not Pumps
Not everyone with type 1 diabetes uses an insulin pump. Many still rely on multiple daily injections (MDI), which makes it harder to tap into the kind of automation that pump users get. But that’s starting to change. AI-based decision support systems are now making personalized dosing advice available to MDI users too, and all it takes is a phone or a browser.
These tools pull together data from CGMs or fingerstick readings, along with info about meals, insulin sensitivity, and daily routines. This now means that individuals can receive clear recommendations around background insulin, meal-time doses, and correction suggestions. These small adjustments reflect what’s going on in your life, not just what’s written in a prescription plan.
One recent study introduced a system called KNN-DSS, which offers weekly insulin adjustments based on trends from the past week. It uses a kind of digital library of dosing options, combined with virtual patient models and expert-approved safety rules.2 Another trial tested a Bayesian decision support app and found it helped lower HbA1c levels while keeping users happy with their treatment. This is a rare combo in diabetes care.8
But numbers aside, what people liked most was the sense of control. In interviews, MDI users said they appreciated getting timely, frequent dose suggestions, the kind that would never happen with check-ins every six months. They also liked being able to override the system when it didn’t quite fit. These experiences suggest that when AI acts more like a guide than a boss, it can make managing diabetes feel more manageable.?8
Learning on the Job: How Closed-Loop Systems Are Getting Smarter
So far, we’ve looked at how AI helps make insulin decisions, whether that be through clinic-based tools or personal apps. But for people using insulin pumps, things are heading into fully automated territory.
Closed-loop systems connect a CGM to an insulin pump, using algorithms that adjust insulin delivery in real time. Most focus on basal insulin (the background dose), but some also tweak bolus doses (the extra insulin you need for meals or corrections). A few systems still ask users to log their meals, while others are experimenting with full automation.6
The idea is simply to allow for less micromanaging and more living.
AI helps by learning from how your body reacts to insulin, adapting over time so that future dosing decisions are better. Reinforcement learning (RL) plays a part here. It’s a type of AI that works a bit like trial and error: the system tests out actions, learns from the results, and improves as it goes. The “reward” that it is chasing is to keep your glucose in range while avoiding dangerous lows.
Early studies using RL with type 1 diabetes data found that these algorithms often landed on dose schedules similar to what doctors would recommend. Newer versions of RL go one step further, looking at more factors, remembering past patterns, and adjusting more efficiently. All of this is still being tested in virtual environments, but the results are promising.9
These systems now tend to look beyond glucose levels. They consider things like HbA1c, physical activity, BMI, and even alcohol use. This wider picture helps to build a more complete picture of what affects your insulin needs. Instead of relying on you (or your doctor) to juggle all those variables, the system starts learning how they work together.
It’s early days, but the direction is looking promising as closed-loop systems are becoming more personalized and smarter, with every data point they take in.
Does it Work? What the Evidence, and People, Say About AI in Diabetes Care
As these AI tools become more common, the big question is, do they actually help? The short answer is yes - but with some variation.
Many clinical trials have shown that AI-supported insulin dosing can bring average glucose levels down and improve HbA1c. But results aren’t always consistent. Outcomes often depend on how the system is designed, how well it fits into real-world care, whether people actually use it regularly, and how long the tools are studied.4
More advanced systems, particularly those that combine CGM with smart analytics, have shown real promise. They can boost time spent in range and reduce big swings in blood sugar for people with both type 1 and type 2 diabetes. Some reviews even suggest that closed-loop platforms help ease the mental load of diabetes management, cutting down on decision fatigue. For people with prediabetes, AI-backed CGMs might help spot early warning signs and support lifestyle changes, though we don’t have long-term data yet.
But beyond the numbers, user experience really matters.
Studies on decision support apps for people using injections show high satisfaction. Users have really appreciated getting personalized guidance without needing to go out of their way for a clinic visit, and they value being able to adjust or reject the app’s suggestions when needed. That flexibility builds trust, and it is something no algorithm can take for granted.8
And of course, safety, as always, still comes first. Developers have been building in hard limits and rules that run separately from the AI models. The way this works is kind of like having a second set of eyes, helping to prevent risky dosing changes. On top of that, these systems go through strict approval processes, ongoing monitoring, and tons of documentation before they reach anyone’s device.5,6
The bottom line here really is that AI tools for insulin management are showing real benefits, especially when they fit into people’s lives and when safety and user control are built in from the start.2,3,7
What’s Next for AI in Diabetes Care?
The future of AI in diabetes care is less about new gadgets and more about connection.
Instead of isolated tools, we’re moving toward systems that bring everything together: glucose data, insulin history, meals, sleep, exercise, stress. When all of that can be seen in one place, AI has the potential to offer truly personalized coaching and smarter, real-time decisions both for everyday use and for clinic visits.5,6
Researchers are also tackling how to build these systems safely and fairly. One promising approach is called federated learning, where AI models learn from data stored on local devices rather than collecting everything in one central system. It’s a way to protect privacy while still improving the technology. This is especially important for global access, where data laws vary.
But for all this progress to reach people who need it, we also have to think about practical barriers: cost, internet access, and the digital literacy needed to use these tools.
That’s where collaboration becomes essential. Endocrinologists, engineers, data scientists, and patient advocates will need to work together, not just to build better algorithms, but to make sure they’re usable and equitable.5,6,8
The tech is moving fast. The challenge now is making sure it works for everyone.
References and Further Reading
- Tyler, N. S., & Jacobs, P. G. (2020). Artificial Intelligence in Decision Support Systems for Type 1 Diabetes. Sensors, 20(11), 3214. DOI:10.3390/s20113214. https://www.mdpi.com/1424-8220/20/11/3214
- Tyler, N. S. et al. (2020). An artificial intelligence decision support system for the management of type 1 diabetes. Nature Metabolism, 2(7), 612. DOI:10.1038/s42255-020-0212-y. https://www.nature.com/articles/s42255-020-0212-y
- Nimri, R. et al. (2020). Decision Support Systems and Closed Loop. Diabetes Technology & Therapeutics, 22(1). DOI: 10.1089/dia.2020.2504. https://www.liebertpub.com/doi/10.1089/dia.2020.2504
- Jia, P. et al. (2020). The effects of clinical decision support systems on insulin use: A systematic review. Journal of Evaluation in Clinical Practice, 26(4), 1292-1301. DOI:10.1111/jep.13291. https://onlinelibrary.wiley.com/doi/10.1111/jep.13291
- Ji, C. et al. (2025). Continuous glucose monitoring combined with artificial intelligence: Redefining the pathway for prediabetes management. Frontiers in Endocrinology, 16, 1571362. DOI:10.3389/fendo.2025.1571362. https://www.frontiersin.org/journals/endocrinology/articles/10.3389/fendo.2025.1571362/full
- Medanki, S. et al. (2024). Artificial intelligence powered glucose monitoring and controlling system: Pumping module. World Journal of Experimental Medicine, 14(1), 87916. DOI:10.5493/wjem.v14.i1.87916. https://www.wjgnet.com/2220-315X/full/v14/i1/87916.htm
- Roche receives CE Mark for its AI-enabled continuous glucose monitoring solution offering critical predictions to people living with diabetes. (2024). Roche. https://www.roche.com/media/releases/med-cor-2024-07-09
- Kobayati, A. et al. (2025). A Bayesian decision support system for automated insulin doses in adults with type 1 diabetes on multiple daily injections: A randomized controlled trial. Nature Communications, 16(1), 8593. DOI:10.1038/s41467-025-63671-0. https://www.nature.com/articles/s41467-025-63671-0
- Mohammad Javad, M. O. et al. (2019). A Reinforcement Learning–Based Method for Management of Type 1 Diabetes: Exploratory Study. JMIR Diabetes, 4(3), e12905. DOI:10.2196/12905. https://diabetes.jmir.org/2019/3/e12905/
Disclaimer: The views expressed here are those of the author expressed in their private capacity and do not necessarily represent the views of AZoM.com Limited T/A AZoNetwork the owner and operator of this website. This disclaimer forms part of the Terms and conditions of use of this website.