How much water do you actually need?

In short

Heinz Valtin searched for the origin of “eight glasses a day” and found no scientific studies supporting it — while conceding an absence of evidence is not a proof of absence.
The North American reference figures, 3.7 L/day for men and 2.7 L/day for women, are totals from all beverages and all food, set from median observed intake rather than a requirement.
The 2%-of-body-mass threshold comes from fixed-intensity laboratory protocols. Across 15 pooled studies, dehydration changed self-paced time-trial power by +0.09% (p = 0.9) but cut fixed-rate performance by 1.91% (p < 0.05).
Over-drinking has a clinical endpoint under-drinking does not: exercise-associated hyponatraemia, blood sodium below 135 mmol/L, driven mostly by excess body water.

The rule nobody could source

In 2002 Heinz Valtin, a renal physiologist at Dartmouth, published an invited review in the American Journal of Physiology whose object was not to test a hypothesis but to trace a piece of advice: that every adult should drink at least eight 8-ounce glasses of water a day — “8 × 8” — with caffeinated and alcoholic drinks not counting.¹ He examined the scientific evidence that might support it and reached a blunt conclusion: “No scientific studies were found in support of 8 × 8.”

The likeliest ancestor is a 1945 statement by the Food and Nutrition Board of the National Research Council: “A suitable allowance of water for adults is 2.5 liters daily in most instances. An ordinary standard for diverse persons is 1 milliliter for each calorie of food. Most of this quantity is contained in prepared foods.” Valtin relays the suggestion that the last sentence was not heeded, and the allowance read as a volume to drink.

He fenced the conclusion in carefully: healthy, largely sedentary adults in temperate climates only, with higher intakes called for “under special circumstances, such as vigorous work and exercise, especially in hot climates.” And he conceded the logical problem with his own paper — “it is difficult or impossible to prove a negative”. The finding is not that drinking less is harmless. It is that the number was never evidence.

Table 1. Daily water figures and what each covers. “Total water” counts water in food as well as drinks.
Source	Figure	What it covers
Food and Nutrition Board, 1945 as quoted by Valtin	2.5 L/day	A “suitable allowance”, about 1 mL per dietary calorie. “Most of this quantity is contained in prepared foods.”
“8 × 8” as popularly stated	≈1.9 L/day	Eight 8-oz glasses of water, beyond other drinks; caffeinated and alcoholic ones excluded.
Institute of Medicine AI, 2005 men 19 y and over	3.7 L/day	Total water: 3.0 L beverages, 0.7 L food (≈81% beverages).
Institute of Medicine AI, 2005 women 19 y and over	2.7 L/day	Total water: 2.2 L beverages, 0.5 L food (≈81% beverages).

What the reference intakes are, and are not

Three years later the Institute of Medicine issued Dietary Reference Intakes for water: an Adequate Intake for total water of 3.7 L a day for men and 2.7 L for women aged 19 and over.² These are routinely quoted as requirements. They are not, in two senses. First, they are totals for water from all sources, food included, not volumes of drinking water. Second, an Adequate Intake is not derived from a requirement at all: it “is set based on the median total water intake from U.S. survey data” — a description of what healthy Americans were already consuming.

The report draws the consequence itself — “a wide range of intakes is compatible with normal hydration. In this setting, the AI should not be interpreted as a specific requirement” — while noting that higher intakes are required by people who are physically active or exposed to heat. One detail cuts against the 1945 framing: in that survey data, beverages supplied about 81% of total water and food about 19%. Whatever was true of American plates in 1945, most water now arrives in a glass.

Where the 2% threshold came from

Ask an athlete what counts as dehydration and you get a number: 2% of body mass. Its source is traceable. The 2007 American College of Sports Medicine position stand holds that “the goal of drinking during exercise is to prevent excessive (>2% body weight loss from water deficit) dehydration… to avert compromised performance.”³ The same document notes that sweating rates vary considerably between people, so “customized fluid replacement programs are recommended.”

Éric Goulet's 2013 meta-analysis asked where that threshold had been measured.⁴ Pooling 15 articles, 28 effect estimates and 122 subjects, he sorted the studies by protocol: self-paced time trials, where a rider adjusts effort as a competitor would, against clamped-intensity tests, where the workload is fixed.

Figure 1. One exposure, two laboratory designs, from one pooled dataset; the dashed line marks no effect. Goulet reported a spread around each mean, but the abstract does not say whether those are confidence limits, so only point estimates are plotted.

The split was stark, and only in the fixed-rate studies did losses of 2% or more impair performance (p = 0.03). Goulet concluded that the 2% rule “has been established from findings of studies using NEV exercise protocols and does not apply to out-of-doors exercise conditions”, and that losses up to 4% are very unlikely to impair real-world endurance performance. That is a critique of external validity, not of the measurements: he allows the clamped-intensity results may matter for military and occupational work, where the pace is not chosen.

Blinding the athletes did not settle it

The obvious objection is expectation: riders who know they are dehydrated may simply ride worse. Two trials removed that variable, and they disagree.

Table 2. Controlled evidence on dehydration and drinking strategy. Outcome metrics differ between rows and are not comparable.
Study	Design	Result
Wall et al., 2015 Br J Sports Med	RCT; 10 cyclists blinded by intravenous saline at 0%, −2%, −3% body mass	“Performance, physiological and perceptual variables were not different between trials.” Rectal temperature ran higher beyond 17 km at −3% (38.9 vs 38.6 °C, p < 0.05).
Funnell et al., 2019 J Appl Physiol	RCT; 14 cyclists, ~3% body mass loss, 2 h fixed-rate preload then ~15 min trial at 31 °C; nasogastric blinding	Trials were 11.4% slower blinded, 10.1% slower unblinded (both p ≤ 0.013); the two did not differ (p = 0.710).
Goulet & Hoffman, 2019 Sports Medicine	Meta-analysis; 7 publications, 82 subjects, 1–2 h cycling or running	Ad libitum beat programmed drinking by 0.98% (95% CI 0.11–1.84) on half the fluid (505 vs 1,073 mL/h).

Wall and colleagues dehydrated ten trained cyclists, then returned intravenous saline to place each rider, unknowingly, at 0%, −2% or −3% of body mass.⁵ Nothing differed but core temperature late in the ride. Funnell and colleagues delivered water covertly by nasogastric tube to fourteen cyclists who rode two hours at a fixed workload in 31 °C heat before a short time trial.⁶ At roughly 3% body mass loss they were about a tenth slower, and blinded riders were impaired as much as unblinded ones — which rules out expectation.

The disagreement is unresolved but not mysterious. The trials differ in how fluid was restored (intravenous saline, bypassing gut and thirst, against water into the stomach), in heat load, and in what came first: Funnell's riders had already spent two hours at a fixed workload in heat, precisely the condition Goulet identified as the one where dehydration bites. With ten and fourteen subjects, neither settles it alone.

The literature agrees more readily about how to drink. Goulet and Hoffman pooled seven publications comparing drinking to thirst against a prescribed schedule.⁷ Ad libitum drinkers took half the hourly fluid, finished 2.1% down in body mass rather than 1.0%, and performed marginally better: 0.98% (95% CI 0.11–1.84). The interval excludes zero, but only just, and eighty-two subjects is thin. Thirst does not win; programming twice the fluid simply bought nothing.

The failure mode of drinking too much

Under-drinking, in healthy people, mostly produces thirst. Over-drinking during endurance events produces a specific and occasionally fatal condition. The third international consensus statement on exercise-associated hyponatraemia defines it as blood sodium below the laboratory reference range — “for most laboratories… less than 135 mmol/L” — during or up to 24 hours after activity.⁸ Sweat sodium losses contribute, but the panel names the predominant mechanism as “dilutional hyponatremia caused by sustained overdrinking”. Severe cases cause cerebral oedema, which is life-threatening.

Nor is it rare where exposure is greatest: the highest incidence of asymptomatic hyponatraemia after a race has consistently been in 161 km ultramarathons, between 5% and 51%. The panel's prevention advice matches what the performance literature reaches from the other direction — “drinking to thirst will, in most cases, prevent both dilutional EAH and performance decrements due to excessive dehydration” — with exceptions listed for dry mouth, genetic variation and unusual sodium intakes.

The bottom line

For a healthy adult in an ordinary day, no evidence says a particular volume must be drunk, and the reference intakes usually quoted are averages of observed consumption rather than requirements. Thirst is fast and sensitive, and the panels that worry most about fluid balance recommend following it. The cases needing more are recognisable: long efforts, heat, a high sweat rate. There the useful number is not eight glasses but your own — weigh yourself before and after a hard session in the heat and you have a sweat rate to plan around.

Most people have no idea what they drink in a day. Kettle logs water alongside food macros and training on the iPhone, so the record exists if you want it.