The Problem with Generic AI Fitness Advice for Diabetics

Why 'Walk 30 Minutes Daily' Is Not a Diabetes Fitness Protocol

The American Diabetes Association (ADA) 2024 Standards of Medical Care in Diabetes recommends at least 150 minutes per week of moderate-to-vigorous aerobic activity, plus 2–3 sessions of resistance training for most adults with diabetes. This baseline matches general exercise guidelines. The complexity begins with everything that surrounds that baseline.

For insulin-dependent users (Type 1, LADA, and some Type 2), exercise isn't just health-promoting. Without proper management, it's a direct metabolic event that requires specific pre-session screening, real-time monitoring, and post-session protocols. A generic AI that prescribes exercise without collecting this information is, in a clinical sense, operating outside the standard of care.

The Pre-Exercise Blood Glucose Protocol

Before any exercise session, a diabetic user on insulin or insulin-stimulating medications must know their current blood glucose and trend. The ACSM/ADA Joint Position Statement (Colberg et al., 2016) provides the clinical framework. Here's what a diabetes-aware AI agent must apply:

Blood Glucose Level	Protocol	Status
<70 mg/dL (<3.9 mmol/L)	Do not exercise. Consume 15–30g fast-acting carbohydrates. Recheck in 15 minutes. Wait for BG to reach >100 mg/dL before starting.	Stop
70–90 mg/dL (3.9–5.0 mmol/L)	Consume 10–15g carbohydrates before beginning. Short sessions (<30 min) may proceed with close monitoring.	Caution
90–250 mg/dL (5.0–13.9 mmol/L)	Safe to exercise. Standard session. Monitor CGM trend if available (falling arrow = different protocol than stable).	Safe Window
>250 mg/dL (>13.9 mmol/L)	Check for ketones. If ketones present: do not exercise. If no ketones: light-to-moderate activity may be acceptable with close monitoring.	Evaluate

A generic AI fitness app has none of this decision logic. It has no mechanism to ask for blood glucose before a session, no knowledge of the user's CGM data, and no protocol for elevated or low readings. This isn't a gap in sophistication — it's a safety gap.

Aerobic vs. Resistance: Opposite Effects on Blood Glucose

The most counterintuitive finding in diabetes exercise science — and the one most AI agents miss — is that aerobic and resistance exercise have opposite acute effects on blood glucose for insulin-dependent users.

Aerobic exercise increases glucose uptake via non-insulin-dependent GLUT-4 translocation in working muscle — it actively drives blood glucose down during and immediately after the session. For users starting in the lower safe range (90–130 mg/dL), prolonged aerobic exercise carries significant hypoglycemia risk mid-session.

Resistance exercise causes a transient rise in blood glucose during the session, driven by catecholamine release (epinephrine stimulates hepatic glucose output). Blood glucose can rise 30–50 mg/dL during heavy strength training. This makes resistance exercise a useful tool for T1D users who want to manage the glucose drop from subsequent aerobic work — but it requires specific sequencing knowledge.

The Delayed Hypoglycemia Problem

The most dangerous aspect of exercise for insulin-dependent diabetics is not what happens during the session. It's what happens overnight.

Exercise-induced muscle glycogen depletion continues to drive glucose uptake from the bloodstream for 6–12 hours post-session, as the body replenishes depleted stores. For T1D users with active basal insulin, this creates a significant window of delayed hypoglycemia — including nocturnal hypoglycemia if an evening session depletes glycogen without corresponding carbohydrate or insulin adjustment.

MacDonald (1987, Diabetes Care) first documented late-onset post-exercise hypoglycemia in insulin-dependent patients, observing episodes occurring 6–15 hours post-exercise. More recent CGM-based studies confirm this window remains clinically significant. The ADA recommends a bedtime snack and possible basal insulin reduction after vigorous evening exercise for T1D patients.

Not All Diabetes Is the Same

Generic AI often treats 'diabetes' as a single category. The clinical reality is 4+ distinct types with fundamentally different exercise management requirements.

The Medication Layer Generic AI Cannot Navigate

Diabetes medications have direct interactions with exercise physiology. A diabetes-aware AI agent must screen for the full medication list before prescribing exercise intensity or timing:

• Metformin: Minimal hypoglycemia risk during exercise. No pre-session adjustment needed in most cases.
• Sulfonylureas (glibenclamide, gliclazide): High hypoglycemia risk during and for hours after exercise. May require pre-session carbohydrate intake even with normal BG.
• Rapid-acting insulin: Active insulin on board from a recent bolus dramatically increases hypoglycemia risk during exercise. Timing of last injection matters enormously.
• SGLT-2 inhibitors (dapagliflozin, empagliflozin): Increased DKA risk with prolonged high-intensity exercise (even with 'normal' BG). ADA recommends caution with vigorous exercise.

What a Diabetes-Aware AI Agent Actually Looks Like

Before any exercise prescription, a diabetes-aware AI agent collects: diabetes type, current medications, typical pre-exercise BG range, CGM usage, exercise timing preferences, and history of hypoglycemic episodes. It uses this to build a personalised pre/during/post exercise protocol — not a generic program with a diabetes disclaimer appended.

It knows that a T1D user doing a 60-minute run at 7pm needs a different set of instructions than a T2D user on metformin doing a 30-minute walk at noon. These are not the same clinical situation. Generic AI treats them identically.