In July 2026, the headlines came back around: the sweeteners in your diet soda might be aging your brain. The starting point is a large Brazilian study that tracked nearly 13,000 adults for eight years. It reported that the people who consumed the most low- and no-calorie sweeteners showed faster declines in memory and thinking. The paper was published in the journal Neurology in September 2025, and it drew a fresh wave of coverage in July 2026 [source: Neurology, 2025].
The figure that traveled furthest is this one. The group that consumed the most sweeteners showed a 62% faster decline in overall cognition than the group that consumed the least — equivalent to about 1.6 extra years of brain aging over the eight-year period [source: Neurology, 2025]. And the amounts were not extreme. Even the highest-consuming group took in, in aspartame terms, roughly the equivalent of one can of diet soda a day [source: American Academy of Neurology, 2025].
Before you pour out the diet soda, though, one thing has to be said clearly. This is an observational study. That is, it found an association between sweetener intake and cognitive decline — it did not prove that the sweeteners caused that decline. This article lays out carefully what the study showed, what it did not show, and what remains unknown — keeping "correlation" and "causation" apart from beginning to end.
One note on how this article handles evidence. Figures that came out of the cohort itself — the sample size, the follow-up, the percentage differences — are treated as measured values. Figures describing what people ate are treated as estimates, because they came from what participants reported about themselves. Subgroup results are weaker still, since each rests on a smaller slice of the same data. Where a number could not be confirmed in the primary sources, it is left out rather than rounded into the text.
Table of Contents
- What was observed: what the study actually saw
- "Associated" is not "caused"
- Reading the numbers carefully: sample, confounding, and measurement
- What regulators say
- So what should you do?
What was observed: what the study actually saw
The cohort behind the headline
The paper is titled "Association Between Consumption of Low- and No-Calorie Artificial Sweeteners and Cognitive Decline," and it carries the identifier DOI 10.1212/WNL.0000000000214023. Its lead author is Claudia Kimie Suemoto, MD, PhD, of the University of São Paulo, and the work was funded by Brazil's Ministry of Health, its Ministry of Science, Technology and Innovation, and CNPq, the national council for scientific and technological development [source: Neurology, 2025].
The study's formal name is ELSA-Brasil (the Brazilian Longitudinal Study of Adult Health). It is a multicenter cohort run across six Brazilian cities, and it followed 12,772 adults (average age 52, 55% women) who were free of dementia at baseline. Participants took cognitive tests three times between 2008 and 2019, over a median follow-up of about eight years [source: Neurology, 2025].
Three waves of testing over eight years
The three rounds of cognitive testing were spread across the study's life: 2008–2010, 2012–2014, and 2017–2019 [source: Neurology, 2025]. That spacing is what makes the design longitudinal — each participant is compared against their own earlier scores, not only against other people. And the eight-year figure is a median, meaning half the cohort was followed longer and half for less.
How intake was estimated
Sweetener intake was estimated from a single baseline food frequency questionnaire (FFQ). The study covered seven sweeteners — aspartame, saccharin, acesulfame-K, erythritol, xylitol, sorbitol, and tagatose. Of these, erythritol, xylitol, and sorbitol are strictly classified as "sugar alcohols," and the study analyzed them together within the broader category of low- and no-calorie sweeteners (LNCS). The lowest-intake group averaged about 20 mg a day, and the highest about 191 mg a day [source: American Academy of Neurology, 2025].
A food frequency questionnaire asks how often someone ate particular foods over a set period — here, the previous year — and turns those answers into estimated daily intakes. Within the seven, sorbitol was the most consumed on average across the cohort, at roughly 64 mg a day [source: American Academy of Neurology, 2025]. It is worth holding on to what this instrument is: a year of eating, reconstructed from memory in a single sitting.
The headline result, stated precisely
The core result is this. The group that consumed the most sweeteners declined 62% faster in overall cognition than the group that consumed the least, equivalent to about 1.6 extra years of brain aging over eight years. The middle-intake group declined 35% faster than the lowest, corresponding to about 1.3 years [source: Neurology, 2025]. The decline was most pronounced in memory and verbal fluency.
Two things are being compared here, and naming them precisely matters. The comparison runs between intake groups inside the same cohort — highest against lowest, middle against lowest — not between sweetener users and a separately recruited control group. And "62% faster" and "about 1.6 extra years" are one finding on two scales, not two findings stacked together. For memory and verbal fluency the study reported the direction of the effect; this article attaches no percentages to those domains, because the domain-level figures could not be confirmed in the primary sources.
Six of the seven, one at a time
When the sweeteners were analyzed one by one, six of the seven were individually associated with cognitive decline. The lone exception was tagatose [source: American Academy of Neurology, 2025]. That is the summary of "what was measured." How to read that result now matters far more.
Analyzing the sweeteners individually asks a narrower question than the pooled analysis: does each substance, on its own, track with faster decline? For six of them the answer was yes — again, in the sense of association. Tagatose was the exception, and the exception deserves care. One cohort reporting no association for one substance is not the same as that substance being shown to be safe; it is a single result in a single population, carrying all the limits that apply to the other six.
Where the result split
One more point: the result split by subgroup. The association was clearest in adults under 60, and there was no statistically significant association in those 60 and older. It was also stronger among participants with diabetes [source: American Academy of Neurology, 2025]. Both of these branches lead straight into the interpretive traps discussed below.
Subgroup results sit lower on the evidence ladder than the headline comparison, because each is computed on a smaller slice of the same participants, and smaller slices give noisier estimates. They are best read as directions for the next study rather than as conclusions that stand on their own.
"Associated" is not "caused"
What an observational cohort can and cannot do
The nature of this study can be summed up in one sentence. An observational cohort can measure an association; it cannot prove causation. The researchers themselves noted that the results do not mean the sweeteners caused the decline, and they cautioned against firm conclusions about cause and effect [source: Neurology, 2025].
A cohort study watches what people do and records what happens to them. Nobody was assigned to consume sweeteners or to avoid them, so the groups being compared formed themselves — and people's choices come bundled with everything else about them. Breaking that bundle apart is exactly what a randomized trial does. It is also why a cohort can establish that two things moved together while staying silent on whether one moved the other.
Reverse causation and hidden common causes
The first alternative explanations to reach for are reverse causation and a hidden common cause. People with metabolic risk — diabetes or obesity, say — tend to switch from sugar to low- and no-calorie sweeteners. If so, the real driver of faster cognitive decline may not be the sweetener but the health condition the person already had [source: Harvard Health Publishing, 2025]. Sweetener intake could be a "consequence and marker" of that condition rather than a "cause" acting on the brain — a possibility that cannot be ruled out.
The two explanations are related but distinct. Reverse causation runs the arrow backwards: the condition heading toward the outcome changes the exposure. A hidden common cause sits upstream of both, driving sweetener use and cognitive decline independently, so the link between them is real in the data and empty as an explanation. Neither can be settled by looking harder at the same dataset, which is why both belong in the reading of the result rather than in a footnote to it.
The diabetes subgroup shows the ambiguity
The diabetes subgroup result seen earlier illustrates exactly this ambiguity. The fact that the association was stronger in people with diabetes can be read two ways — the sweeteners may have been especially harmful to them, or the underlying diabetes itself may have driven the cognitive decline. That the same data permit both readings is the essential limit of an observational study.
Why the distinction survives the retelling
This distinction may look like a technicality. In practice it is decisive. A punchy line like "62% faster decline" travels much faster and farther than all the caveats around it. The moment an association is quietly reread as causation, an unverified conclusion starts circulating as a "scientifically proven fact."
The practical test is simple. Ask what would have to be true for the headline reading to hold, then ask whether this study tested it. For "sweeteners age your brain" to hold, the sweetener intake would have to be doing the work — not the diabetes, not the wider diet it sits inside, not the eight years of eating that a single questionnaire never saw. This study did not put that to the test, and by design could not.
Reading the numbers carefully: sample, confounding, and measurement
Relative speed, absolute years
Start with that famous 62%. It is a relative figure. It describes how much faster the slow, natural process of cognitive aging ran in relative terms. That is why the study paired it with "about 1.6 years over eight years." That framing lets you gauge the size of the effect honestly, without inflating it.
Relative figures behave differently from absolute ones. A 62% faster rate of decline sounds enormous because the percentage is doing the talking, but what it is 62% of is a slow process unfolding over eight years. The paired statement — about 1.6 additional years of brain aging across that period, and about 1.3 for the middle group — is what turns the percentage into something a reader can picture.
A single snapshot standing in for eight years
The measurement method has limits too. Diet information was collected only once, at the start of the study, by self-report. People may not accurately remember what they ate, and if their eating habits changed over the eight years, that change was not captured in the data [source: Harvard Health Publishing, 2025]. In effect, a single snapshot at baseline stood in for eight years of exposure.
Two distinct errors ride along with that design. One is recall: what a person remembers eating last year is not what they ate. The other is drift: someone who drank diet soda daily at baseline may have quit years before the final cognitive test, and the dataset would still count them among the high consumers.
What "adjusted for" actually means
The researchers did statistically adjust for a range of confounders — age, sex, income, physical activity, blood pressure, cardiovascular disease, body mass index (BMI), and diet quality [source: Harvard Health Publishing, 2025]. That procedure strengthens confidence, but it is not a cure-all. Variables that were not measured, or not fully controlled, could still have shaped the result. "Adjusted for" does not mean "removed."
One more covariate belongs on that list: the adjustment set also included education [source: Neurology, 2025]. Statistical adjustment estimates what the association would look like if the compared groups matched on those variables. It works only for variables that were measured, and only as well as they were measured — which is why an adjusted association beats an unadjusted one and still falls short of an experiment.
No imaging, no biomarkers, no mechanism
The study also had no brain imaging or biomarkers. In other words, it offered no biological evidence of the mechanism — of how sweeteners might act on the brain [source: Harvard Health Publishing, 2025]. The association was observed, but the mechanism that would explain it lies outside this study.
A mechanism is what turns an association into an explanation: a route by which the exposure could plausibly produce the outcome. With no imaging and no blood-based markers, this study has nothing to say about that route. It can report that the two things moved together, and there it has to stop.
The age paradox: absent, or undetected?
The most intriguing part is the "age paradox," in which the association vanished in those 60 and older. Andrew Budson, a neurologist at Harvard Medical School, called the result "curious" but cautioned that it is hard to read as evidence of safety after 60. Older adults show more individual variation in memory change, which makes small effects statistically harder to detect [source: Harvard Health Publishing, 2025]. The same result can be read as "harmless in older adults" or as "simply harder to detect in older adults" — and this study alone cannot tell you which.
The distinction that matters is between an effect that is absent and an effect that is not detected. A study finds nothing when there is nothing to find — and it also finds nothing when the signal is smaller than the noise around it. The second is a statement about the study's ability to detect, not about the substance. From the outside, the two results look identical.
One population, one starting point
Finally, this result came from a single population, in Brazil. Whether the same pattern would appear elsewhere, among people with different diets, genetics, and environments, has not yet been confirmed. One large cohort is a powerful starting point, but it is not, by itself, a universal conclusion.
Generalization is not automatic in either direction. A Brazilian cohort does not make the finding local to Brazil, and it does not make it universal either — it makes replication a real question rather than a formality. The caveat names diets, genetics, and environments, and any of those could change the size of an association, or whether it shows up at all.
What regulators say
ADI: the frame regulators use
Step back, and the official stance on sweetener safety sits on a wider footing than this one study. Regulators such as the U.S. Food and Drug Administration (FDA) and the European Food Safety Authority (EFSA) have judged approved sweeteners to be safe within an Acceptable Daily Intake (ADI), and this observational study does not change those approvals.
An Acceptable Daily Intake is the amount a regulator judges a person can consume every day without appreciable risk, set as part of approving the substance. It is a threshold: the question it answers is whether intake stays under a limit. That is a different question from the one this cohort asked, which was whether intake tracked with faster cognitive decline among people eating as they normally do. One can produce a finding without the other having to move.
The WHO's 2023 conditional recommendation
The World Health Organization (WHO), by contrast, issued a conditional recommendation in May 2023 advising against using non-sugar sweeteners (NSS) as a means of weight control or reducing the risk of noncommunicable diseases. The certainty of the evidence was rated "low," and the guideline noted that long-term use may be associated with an increased risk of type 2 diabetes and cardiovascular disease [source: World Health Organization, 2023]. It is worth being clear, though, that this recommendation concerns weight and metabolic health, not cognition directly.
Two qualifiers in that recommendation do real work. The recommendation is conditional rather than strong, and the certainty of the evidence behind it was rated low — those are the guideline's own labels for how far it is willing to go, and it was released on 15 May 2023 [source: World Health Organization, 2023]. Read with the labels attached, it is a cautious steer about weight and metabolic outcomes. It is not a safety ruling, and it is not a statement about the brain.
Two positions, two questions
Placed side by side, the picture sharpens. Regulators do not brand sweeteners as an immediate hazard, but neither is the assumption that "zero calories means zero consequences" a verified fact. This cognitive study simply adds one more question mark to that assumption — as grounds for caution, not for alarm.
It is tempting to net the two off against each other — regulators say safe, the WHO says avoid, so they cancel out. They do not. They answer different questions with different evidence, and neither was asked about cognition. The cohort result does not overturn an approval, and an approval does not dispose of the cohort result.
So what should you do?
What was measured
Let us sum up. The verified facts are these. In one large cohort, people who consumed more low- and no-calorie sweeteners declined faster cognitively, and that association was especially pronounced in those under 60 and in people with diabetes. These are values that were actually measured.
What is still open
What has not been established is just as clear. Whether sweeteners cause cognitive decline, by what mechanism they might act, whether the result replicates outside Brazil, and what it means over the long run in real life are all open questions. This study is less an answer than a well-designed question.
The case for a moderate default
Given that, the reasonable move right now is neither alarm nor dismissal, but moderation. Making water or unsweetened drinks the default, and keeping sweetener intake at a sensible level, is a stance this evidence comfortably supports on its own. At the same time, remember that the study does not say "so sugar is better." Sugar carries its own risks, and this result was not a contest to rank sweeteners against sugar.
Moderation is also the choice that costs least if the finding turns out to be wrong. If sweeteners are eventually cleared, whoever drank more water lost nothing. If the association turns out to reflect something real, that same person was already ahead. It is the asymmetry, not the 62%, that justifies adjusting a default.
Three things to watch
Three things are worth watching from here. First, whether cohorts in other populations and randomized controlled trials (RCTs) replicate this association. Second, whether brain-imaging and biomarker studies reveal a plausible mechanism. Third, whether regulators revisit their positions as this kind of cognition-related evidence accumulates. For now, one thing is clear — "harmless" is not a proven conclusion but a still-open question.
The three have something in common: each would convert an association into something firmer. A replication would show the pattern is not an artifact of one cohort. A mechanism would show how the effect could work. A regulatory review would show that the evidence had reached the threshold where official positions move. Until at least one of those lands, the honest summary is the one the researchers gave themselves — an association, carefully measured, and not yet an explanation.