AI as the mirror of the sad state of medical information: the case of zinc
I put three questions about zinc to a widely used AI research assistant. The corpus it returned was serious, the dissidents finally got named, and yet the transcript reads as an inventory of what a competent machine cannot say. What follows is a reading of that silence, round by round.
I put three questions about zinc to Perplexity, running Claude Sonnet 5 in thinking mode. Each question was sharper than the last. Perplexity retrieves published sources and cites them, Sonnet 5 composes patient, structured replies from that retrieval, and the combination is a fair representative of what a curious reader will reach for today when they want an AI that shows its sources on a medical question. Nothing in what follows should be read as an argument for reaching for a different tool of the same kind, because the pattern I am about to describe is not a property of this tool.
The corpus that came back was serious. Prasad and Maret were named, the Japanese cohort was there, ZENITH and the BOND panel both appeared, and the mechanistic paragraphs on IGF-1 and mTOR were competent. This is not a story about a poor instrument.
It is a story about the exact line a competent instrument cannot cross.
Read the transcript once as science and it looks acceptable. Read it a second time as a genre, one with its own rules about who may be blamed and who must be spared, and the missing shape becomes hard to unsee. Controversies are sanded down until they read as balanced perspectives, and clinicians who have been raising the alarm for sixty years appear as one voice inside a diplomatic chorus. The word supplementation hardly finishes a sentence before a nuance arrives to cancel it. Nobody asks why geriatric medicine, facing a documented crisis of anabolic decline, has never adopted thymulin as a routine measure of functional zinc status. Neither ethics nor responsibility gets raised; the aging population appears as a statistical curve rather than as an object of duty.
None of this originates in the machine. It all comes from the corpus the machine is trained to synthesize politely, faithfully, and inoffensively. Sonnet 5 is not the source of the problem; it is the mirror in which the problem finally becomes legible, because a good mirror shows you what a poor mirror can only hint at. What follows is a reading of that mirror, round by round: what a competent machine cannot say about zinc, and what has to be inferred from its silences.
Round one: encyclopedia mode
The first question asked what proportion of people across ages are zinc-deficient, acutely or subclinically, and what the impact on anabolic capacity might be. Anabolic capacity, as a plain-language starting point, is the body’s ability to build itself. Every day tissues take in amino acids, minerals, and vitamins and use them to synthesize new proteins, new membranes, new muscle fibres, new immune cells. When anabolic capacity falls, the body cannot keep pace with its own daily need to replace what it has lost. Muscle thins, skin thins, immunity thins. Two clean subquestions, then: how many people carry a zinc shortfall, and what does the shortfall do to that daily rebuilding.
The reply organized itself around global figures. A Turkish paediatric study reporting 27.8% subclinical deficiency, an Indian meta-analysis pooling 55% in children aged six months to six years, an IFPRI estimate that 17.3% of the world population is at risk of insufficient intake, a Japanese national database reporting 34.8% frank deficiency and 45.5% additional marginal cases in a clinical adult population, and so on. The reader learns that deficiency is common in the Global South, that it rises with age in Japan, and that hospitalized adults there run at 50.3%. The picture is coherent, sourced, and unobjectionable.
Two paragraphs on mechanism follow, and it is worth slowing down on them, because they name three of the most beautiful pieces of cellular signalling that biology has uncovered in the last fifty years, and the reply passes over their beauty without letting the reader see it.
The first is IGF-1, insulin-like growth factor 1, a small protein released by the liver in response to growth hormone. IGF-1 is the endocrine signal by which the body tells its own tissues, throughout the day and the night, that the conditions are right to build. Zinc is required both to synthesize IGF-1 and to allow IGF-1 to bind to its receptor on the surface of every cell that is meant to respond to it. In a zinc-deprived body, the signal falls, and even if it did not fall, the receiving cells could not hear it.
The second is mTOR, mammalian target of rapamycin, a molecule so central to cellular decision-making that entire branches of biology and pharmacology have grown up around it. mTOR is the switch that decides, moment by moment, whether a cell will use its resources to synthesize proteins or hold in reserve. It integrates signals from amino acid availability, from insulin, from growth factors, and from cellular energy status, and it turns the ribosomes on or off accordingly. Work by McClung and colleagues showed that zinc activates mTOR directly, independently of insulin and independently of the amino acid supply. This means that the mineral itself, quite apart from what one has eaten or what the pancreas has released, is a first-order instruction to the anabolic machinery of every cell that can hear it. Take zinc away, and the switch drifts toward the off position, regardless of how much protein the body has been given.
The third is the pair of experiments from Oteiza’s group in rats. If you take a rat, deplete it of zinc, and then replace the testosterone that also falls when zinc falls, the muscle still cannot maintain the messenger RNA it uses to build its own proteins. Zinc is not acting through the hormonal axis. It is acting on the transcriptional machinery itself, and the machinery goes quiet when the mineral is not there.
The reply then closes with a paragraph acknowledging that human cross-sectional studies have failed to find a clean association between zinc intake and sarcopenia, the age-related loss of muscle mass and strength that hollows a body from within, that a review by van Dronkelaar and colleagues found no clear link with muscle mass, and that a Japanese supplementation trial found increased calf circumference and grip strength. The formulation converges on the phrase permissive factor: zinc, we are told, does not amplify anabolism beyond normal, it merely permits it.
Notice what happened here. The question was about anabolic capacity, which is a functional question with clinical consequences for aging bodies. The reply answered the question by locating zinc deficiency mainly outside the Western world, by describing the anabolic mechanism as a general biological curiosity, and by concluding that the human evidence is mixed. The word supplementation appears exactly once, inside a sentence that immediately signals the divergence of the literature. No recommendation is possible from this reply. No clinician reading it will act on it. No aging body will benefit from it. It is a competent encyclopedia entry that ends where medicine would begin.
Round two: the physiological turn
The second question narrowed the frame. Western populations, age brackets, and above all the physiological status of zinc rather than intake alone. The distinction matters and deserves a word. Intake is what one eats. Status is what the body actually has, in the compartments where zinc does its work. The two are not the same thing. A person can eat enough zinc for the label to look adequate and still fail to keep it, or fail to use it, or fail to keep it in the right place. That gap between what enters the mouth and what reaches the working machinery is the whole territory of nutritional physiology, and the reply had to be pushed onto that territory before it would explore it.
Once pushed, something more interesting happens. The reply produces Hennigar and colleagues, working with data from NHANES 2011-2014, the American national nutrition and health survey, and reports that only 3.8% of American children under ten, 8.6% of males over ten, and 8.2% of females over ten fall below the biological sufficiency thresholds set by the BOND panel, the Biomarkers of Nutrition for Development panel that ratifies which blood tests count as evidence of deficiency. It also produces the striking observation that 15% of American adults consume less zinc than the Estimated Average Requirement, roughly twice the biological deficiency rate, and it explains the gap by invoking the robustness of adult zinc homeostasis in the healthy.
Then it produces ZENITH, a study of 387 French, British, and Italian subjects between 55 and 85 years old, reporting biological zinc deficiency in fewer than 5% of independently living European elders. It produces Andriollo-Sanchez and colleagues, whose work on 1090 healthy European elderly demonstrated that the age-related decline of plasma zinc depends more on age-related pathophysiological changes than on nutritional intake itself, and that the decline correlates with serum albumin and the copper-zinc ratio, the latter a marker of chronic systemic inflammation. It produces ZINCAGE, an 819-person European cohort in which plasma zinc correlated with two of the standard signalling molecules of chronic inflammation, interleukin-6 and interleukin-8.
And then it names the mechanism. It names inflammaging, the slow, low-grade inflammation that installs itself in older bodies without ever crossing the threshold of an acute inflammatory illness, subtle enough to be missed by standard markers and powerful enough to redistribute nutrients and reshape metabolism over years. And it names the way inflammaging redistributes zinc: metallothionein-mediated tissue sequestration. Metallothioneins are small zinc-binding proteins that cells produce in greater quantity when they sense inflammatory signals, and once produced they capture the available zinc and hold it in intracellular stores, drawing the pool out of the circulation. The result is that a body dealing with chronic low-grade inflammation, which is to say the majority of bodies past sixty, actively removes its own zinc from the pool the standard blood test samples, without any dietary insufficiency needing to be involved.
And the reply draws the honest conclusion, in a single closing paragraph: because the fall of circulating zinc in healthy Western elders arises primarily from inflammatory redistribution rather than intake deficit, dietary supplementation alone may not restore the bioactive pool required for mTOR activation and IGF-1 signalling. The anabolic resistance observed in aging Western bodies is likely a problem of tissue bioavailability governed by chronic inflammation and albumin status, and it is not correctable by dietary intake alone.
This is a real admission. It should collapse the framing of the first round. It should also collapse the standard geriatric position that Western elders eating a mixed diet are zinc-sufficient. Yet the reply frames the admission as a nuance, one interesting finding among several, and moves on. No follow-up on what to do about it. No mention that inflammaging is a mass phenomenon that affects the majority of the population over sixty. No question about why a mechanism established over twenty years ago in the European literature has never produced a clinical protocol. Anti-inflammatory intervention is not mentioned. Zinc supplementation is not mentioned. Thymulin, the functional marker that would settle the question, is not mentioned. The paragraph closes and the reply ends.
Round three: the dissidents arrive, only to be neutralized
The third question forced the frame open. It asked for zinc specialists rather than authorities, on the grounds that authorities operate with a permissive definition of normality.
This is where Ananda Prasad and Wolfgang Maret finally enter the transcript. Prasad, who identified human zinc deficiency in 1958 and whose work led the American Congress to establish the recommended intake in 1974, has argued for over sixty years that mild deficiency is nearly undetectable by conventional means. His human depletion model showed that thymulin, a small zinc-dependent peptide by which the thymus instructs young T-cells to become the helper cells that orchestrate the entire adaptive immune response, falls within eight to twelve weeks of a low-zinc diet, whereas plasma zinc itself does not decline until twenty-four weeks. That is a gap of months. The body registers zinc scarcity in the operations that actually depend on zinc long before the blood test detects that anything is wrong at all. Serum zinc, Prasad has always insisted, is a late and insensitive marker that hides functional deficits already at work on immunity and, plausibly, on many other cellular systems.
Maret, of King’s College London, published in 2025 a perspective titled The Arcana of Zinc. The word arcana is not decorative. Maret writes that plasma or serum zinc remains, after decades of research, the only biomarker used in clinical practice, even though it represents only 0.1% of total body zinc and belongs to an exchangeable pool whose utility as a biomarker experts themselves describe as uncertain. He notes that muscle zinc, which holds roughly 60% of the body’s zinc content, has a turnover time of 210 days. That is seven months. Seven months for the muscle to renew the zinc within it. Whatever zinc entered a person’s diet in April is, in November, still inside the muscle, leaving on its own schedule and not on the schedule of the blood test that decides whether they are sufficient this morning. A normal serum reading in an aging body can perfectly coexist with a significant muscular tissue deficit, and there is no clinical instrument in current practice that will detect the second while the first is reassuring.
The BOND panel, headed by Janet King, Kenneth Brown, Rosalind Gibson, and Nancy Krebs, admits in its own reference document that no specific and sensitive biomarker of zinc status exists to this day, and that plasma concentration responds less to supplemental intake with meals than to supplementation between meals, producing wide interindividual variability.
These are not marginal statements. They are, taken together, an admission that clinical medicine has been operating without a valid measure of zinc status for the entire modern era, and that a functional replacement has existed in Prasad’s laboratory since the 1980s but has never been adopted.
The reply presents these facts, and then it neutralizes them.
Watch the construction. Prasad, the paragraph tells us, clearly leans toward therapeutic supplementation rather than mere dietary adjustment in cases of disease- or age-conditioned deficiency, and he claims that zinc treatment can be very useful in many chronic diseases, improving cellular immunity, reducing oxidative stress, and lowering chronic inflammatory cytokines in humans. This is a strong position, stated as such. And then, in the very next sentence, Maret is introduced with the word however, and used to warn that blind supplementation without functional evaluation risks being ineffective or even deleterious if cellular zinc is already locally excessive in certain pathologies. A further sentence adds that dietary phytate, a plant compound that binds zinc in the gut and prevents its absorption, may cause dietary correction to fail even at nominally sufficient intake.
The reader who arrives at this paragraph without a stake in the outcome will conclude that the situation is complex, that the experts disagree, and that no clear guidance can be given. This conclusion is manufactured, not derived. Maret does not actually contradict Prasad. Maret says that supplementation cannot restore cellular homeostasis in pathologies with underlying oxidative stress, and that age-specific recommendations are needed for the elderly, who develop the exact combination of oxidative stress and zinc deficit that characterizes inflammaging. The two positions converge on a common practical implication: aging Western bodies under inflammaging need zinc supplementation, alongside anti-inflammatory intervention, and they need it now. The transcript takes those converging positions and rewrites them as a stalemate.
What the sequence reveals
Nothing that mattered arrived spontaneously. Every substantive fact in the transcript came in response to a corrective question. The physiological framing arrived because the questioner rejected the intake framing. The dissidents arrived because the questioner rejected the authorities. The controversy over biomarkers arrived because the questioner explicitly asked for it. A reader who accepted the reply of the first round would leave with an encyclopedia entry that recommends nothing. A reader who accepted the reply of the second round would leave with a nuance that changes nothing. A reader who accepted the reply of the third round would leave with a manufactured stalemate that prescribes nothing.
Each opening was followed by a closing. The mechanism was described and then the mechanism was declared permissive rather than actionable. The tissue redistribution was described and then declared beyond the reach of dietary intervention without specifying the intervention that might reach it. The dissidents were named and then arranged in mutual cancellation. The pattern is regular enough to constitute a genre. The genre has a name in journalism: false balance. In medicine it has no name yet, but it has a shape, and the shape is the shape of the transcript we have just read.
The most revealing absence is ethical. Not once does the reply ask why geriatric medicine, facing a documented crisis of anabolic decline in an aging Western population, has failed to adopt Prasad’s functional assay after four decades of established science. Not once does it ask why nutritional recommendations for zinc in the elderly remain calibrated to intake when the pathophysiology has been known for two decades to operate through tissue redistribution. Not once does it ask who benefits from the persistence of a demonstrably inadequate serum test as the sole clinical marker. The answer to each of these questions is unflattering to specific parties, and the transcript’s genre forbids it.
What has to be said
Serum zinc is a false negative for the aging body under inflammaging. It reports normal while thymulin has already collapsed, while metallothionein sequestration has already begun, while mTOR activation has already declined. A clinical framework built on this test has been misclassifying the aging population for the entirety of its existence. The mechanism has been in the peer-reviewed literature since Prasad’s depletion studies of the 1980s, with progressive refinement through Andriollo-Sanchez, Mocchegiani, Maret, and the BOND panel’s own admissions. It is not a fringe view. It is the specialist consensus of the people who study zinc, and it is silently overruled by every clinical guideline that still treats serum zinc as sufficient.
For the aging body attempting to preserve muscle, this matters directly. Anabolic resistance, the reduced protein synthesis response to feeding that characterizes older adults, has multiple contributors, but zinc-dependent mTOR activation is one of them, and it can be functionally deficient in a body whose serum zinc reads within reference range. A person on a hypocaloric protein-adequate regimen, eating fewer calories than they burn in order to reduce fat mass while meeting the recommended grams of protein per kilogram, is not necessarily failing at the arithmetic of grams. They may be failing at the letter accounting downstream, the tissue-by-tissue balance of amino acids and cofactors that must all arrive together for a specific protein to be built, and one of the letters that determines whether the anabolic machinery can even initiate translation is a mineral whose actual sufficiency their bloodwork cannot see. The dietary intake alone will not fix it if inflammaging is redistributing what they eat into tissue depots faster than the muscle can draw on it. The supplementation alone will not fix it if the underlying inflammation continues to sequester what is supplemented. Both are required, and neither is prescribed, because the test that would identify who needs them has not been adopted.
This is what the transcript could not say. It could not say that a whole population is being clinically misclassified, that the misclassification has known technical solutions, that those solutions have been available for decades, and that the failure to adopt them is a decision, not an absence of means. It could not say that the person reading these lines, if they are over sixty, or under chronic inflammation, or holding a plateau on a caloric deficit, has a substantially higher prior probability of functional zinc deficit than their blood test will ever detect. It could not say that the ethical response to this situation is not more research, but the acknowledgment of what forty years of research has already established.
The reason it could not say those things is not that the machine is broken. The reason is that the corpus the machine synthesizes is itself the site of the omission. The clinical literature has learned, over decades, to phrase these matters in a register that admits the finding without triggering the recommendation. The reply we read reproduces that register faithfully. It is polite because the corpus is polite. It is inoffensive because the corpus is inoffensive. It admits and it retreats, together, in the same sentence, because that is what the corpus has been rewarded for doing.
A physiological right, in the sense this resource has been building toward, is what remains once the mediation gets lifted. The mediation is the register. The register is the polite retreat. The right is the sentence the register cannot allow. In the case of zinc, the sentence reads: an aging Western body has, by default, a higher probability of functional zinc insufficiency than any current clinical protocol will detect, and it has a reasonable claim on the interventions that would correct it, whether or not the biomarker has caught up.
The machine will not write that sentence. It has now been written.