Are AI nutrition plans safe? The danger of LLM coaching
Why KEXBI uses fixed rules for safety-critical numbers like energy availability and hydration limits, and keeps language models to writing only.
Why KEXBI uses fixed rules for safety-critical numbers like energy availability and hydration limits, and keeps language models to writing only.
The new wave of chatbot-first AI nutrition tools is handing safety-critical numbers to a language model. Calorie floors. Hydration ceilings. Sodium targets. KEXBI's architecture refuses that choice. Safety-critical calculations are fully deterministic, built on peer-reviewed research. AI is used for creative synthesis only. Here is exactly where the line is drawn, and why it matters.
The AI-nutrition market is not one category. It is three.
There are the established tracking apps (MyFitnessPal, Cronometer, Lifesum) whose core logic is deterministic and which have begun layering AI features on top. There are data-first coaches (Zoe, Levels, MacroFactor) built on specifically trained models over microbiome data, continuous glucose, or adaptive expenditure. And then there is a third category, the chatbot-first cohort, where the nutrition plan itself is generated by a frontier language model on demand, reasoned-to rather than computed.
It is this third category we are arguing against.
The specific problem inside that cohort is not what most people worry about. It is not that the advice is bland. It is not that the meal plans are generic. It is not that the onboarding is slow.
The problem is that the numbers themselves are being generated by a language model. The calorie floor. The sodium target. The hydration ceiling. The protein dose.
The model has no concept of what happens to someone training for endurance whose energy availability drops too low. It has no model of gastric emptying. It has no fixed rule for how last night's sleep should change today's training. It produces plausible-sounding text that is statistically close to the training corpus, and then it stops.
That is fine when the answer is “a recipe for roast aubergine.” It is not fine when the answer is “how many calories you can cut this week before your cycle stops arriving,” or “how much sodium to consume across a 3-hour ride in 28°C heat.” These questions carry real physiological risk at the extremes. Get them badly wrong, and people get hurt.
KEXBI was built on one engineering conviction, stated plainly: the numbers are never hallucinated. Only the language is modelled. Here is exactly how that split is implemented, and why every piece of peer-reviewed research it rests on is one we can name.
Inside the KEXBI pipeline, every calculation that can harm someone if it is wrong runs through a deterministic specialist. Every calculation that can be creative without consequence runs through a language model. The two layers are not mixed. The outputs of the synthesis layer are bounded at runtime by the outputs of the deterministic layer.
If the language model tries to plate a 420 kcal breakfast on a day where the deterministic engine has calculated a 620 kcal target, a final deterministic check rejects the plate. The meal gets rebuilt until the constraint is satisfied. You never see the rejected version.
The four calculations below are representative, not exhaustive. Each one is built on peer-reviewed research. Each one would be unsafe to delegate to a language model. Each one is, in KEXBI, a closed-form equation or rule that runs the same way every time and can be audited line-by-line.
Energy availability (EA) is one of the most important numbers in endurance nutrition. It is also the one most likely to be wrong in a chatbot-generated plan.
EA is defined as dietary energy intake minus the energy cost of exercise, divided by fat-free mass. Research, including the 2023 IOC consensus, treats low energy availability as a continuum rather than one universal cut-off. The lower and longer it runs, the higher the risk of the cluster of symptoms now formalised as Relative Energy Deficiency in Sport, or RED-S.
The possible consequences are not subtle: suppressed reproductive function, thyroid downregulation, bone-mineral-density loss, impaired immunity, and elevated cortisol [1][2][3].
In controlled feeding studies in women, reproductive and thyroid hormones and bone turnover markers shifted below about 30 kcal/kg FFM. KEXBI therefore uses a planning guardrail: it works to keep intake above about 30 kcal per kg of lean mass for women and 25 for men, for people training at a moderate level or above. A language model, by contrast, has no such guardrail. It will cheerfully output a 1,600 kcal target for a 65 kg runner burning 900 kcal a day, because nothing in its weights objects.
Cites: Loucks & Thuma 2003 [1]; Mountjoy et al. 2014/2018/2023 IOC consensus [2]; De Souza et al. 2014 [3].
Hydration is the other place a hallucinated number becomes a medical problem.
Under-drink in the heat and you cook. Over-drink and you dilute plasma sodium into exercise-associated hyponatraemia, which is a hospitalisable event. The corridor is not a matter of taste. It is bounded by three numbers with decades of peer-reviewed characterisation: gastric emptying, intestinal absorption, and sweat sodium.
Gastric emptying is rate-limited. Research suggests the stomach empties fluid during exercise at about 1 L/hr, a little more at best, with individual variation, and more slowly once you are dehydrated [4][5]. KEXBI therefore caps planned drinking during a workout at 0.8 to 1.2 L/hr, scaled by body mass. The 150% post-exercise replacement figure is the Shirreffs & Maughan finding that has stood for two decades: drink only what you lost and you are still in negative water balance two hours later [6]. The carbohydrate defaults follow Jeukendrup's multi-transportable carb work [7]: about 90 g an hour is KEXBI's default, the usual guidance for mixed carbohydrates; some trained people absorb more. An LLM asked to “plan hydration for a 3-hour ride” has no guarantee it will respect any of these. KEXBI's hydration plan is designed to, because the limits are fixed in code, not prompted.
Cites: Rehrer et al. 1990 [4]; Jeukendrup & Moseley 2010 [5]; Shirreffs & Maughan 1998 [6]; Jeukendrup 2014 [7].
Sleep is not a wellness variable. It is a recovery variable, and KEXBI reads it every morning.
Research suggests a short night changes how the body handles food. The Spiegel and Donga studies reported 25 to 30% drops in peripheral insulin sensitivity after restricted sleep [8][9]. In one study, resting metabolic rate fell by roughly 3% and free T3 by about 5% [10]. Cortisol can rise, and hunger-regulating hormones (leptin, ghrelin) can shift against you. KEXBI does not turn those findings into a carb cut or a macro multiplier. By the time last night's sleep is known, the day's food is already placed around your training, and moving it would break limits that rest on firmer evidence: an easy stomach before a workout, the fat floor, protein spread across meals. So sleep acts on the training suggestion instead.
Handing that decision to a language model and asking it to “modify today's plan for a bad night's sleep” is not engineering. It is vibes.
Each line is a fixed rule, not a prompt, and each is offered, not forced. If you take the easier day, the lighter training load flows back through the day's blueprint and the macros are worked out again by the same deterministic pipeline. A user can audit the adjustment. A regulator can audit the adjustment. An LLM-only competitor cannot, because no such fixed chain exists in their system.
Cites: Spiegel et al. 1999 [8]; Donga et al. 2010 [9]; Nedeltcheva et al. 2010 [10].
The least glamorous deterministic layer is also the one most often botched by LLM nutrition tools: the basic arithmetic of calories and macros.
KEXBI uses Mifflin-St Jeor for resting metabolic rate. It derives a physical activity factor (PAL) from your training blocks. It sets protein per kilogram of lean mass, by goal: 1.9 g at maintenance, 2.2 g for recomposition or lean gain, 2.4 g during fat loss and 2.6 g for performance, with a ceiling of 3.1 g per kilogram of lean mass drawn from Helms and colleagues [11][12][13][14].
None of this is novel. That is the point. These are the same equations a registered sports dietitian has used for decades, implemented exactly, every time, with no drift. An LLM will approximate them. It will occasionally forget to subtract protein calories before computing carbohydrate. It will round in ways that move protein under the floor. KEXBI does not approximate. It calculates once, and hands the fixed numbers to the synthesis layer to build meals within.
Cites: Mifflin et al. 1990 [11]; FAO/WHO/UNU PAL tables [12]; Helms et al. 2014 [13]; Phillips & Van Loon 2011 [14].
None of the above means KEXBI is anti-AI. The opposite is true. AI is indispensable exactly where deterministic logic fails: in the creative, culturally-rich, palate-sensitive act of building a meal you will actually eat.
Once the deterministic engine has produced a constraint set (“this meal must be 620 kcal, 45 g protein, 80 g carbohydrate, ≤18 g fat, draws only from your pantry, follows your taste profile, lands between 13:30 and 14:15”), a language model does exactly what it is good at. It assembles a plausible, appealing, cohesive dish. It names it. It writes the three-sentence plan note. It explains why this plate fits the biological window you are in.
Then the output passes through a final deterministic check, which re-verifies macro totals, pantry availability, and safety-check flags. Anything that drifted during the synthesis step is rebuilt. You only see plates that survive both layers.
In practice, the synthesis layer is where personality, cuisine, and variety live. The deterministic layer is designed to stop that personality putting anyone at risk.
In the next eighteen months, the number of nutrition apps claiming “AI coaching” will continue to multiply. A significant share of them will be a single API call to a frontier model, a system prompt, and a subscription page.
The margin of error on those products is exactly the margin of error on the underlying language model. For the calculations that matter, that margin is not bounded by anything physiological. One hallucinated sodium target in a heat-wave training block is one product-ending event.
KEXBI's position is not that AI is dangerous. It is that AI is incorrectly scoped in the chatbot-first segment of the market.
Safety-critical calculations belong in deterministic specialists with rules drawn from peer-reviewed research that can be named, cited, and audited. Creative synthesis belongs in a language model with its outputs gated. The moat is not a secret algorithm. It is the discipline of having drawn that line in the first place, and of having built the twenty-six-specialist pipeline that enforces it.
The numbers are never hallucinated. Only the language is modelled. That is the entire argument, and the one we expect to survive regulatory contact.
The research figures in this article are drawn from peer-reviewed primary literature and major consensus statements. KEXBI's own defaults, such as its energy-availability guardrail, are set against these sources and subject to ongoing refinement as new data arrives.