Are sweeteners bad for you?

A new umbrella review pooled data from 50,034,327 people and concluded that sweeteners are linked to higher risk of overweight, metabolic syndrome, type 2 diabetes and hypertension. The number is real. The headline is accurate. And it is still the least interesting thing about that paper.

Because the same review also pooled the randomised trials. There, sweeteners reduced body weight, at the highest certainty rating the system awards. One paper, two opposite findings, the same substances.

That is not a contradiction in the data. It is a difference in study design. Once you understand it, you can sort out almost every sweetener headline of the past twelve years yourself.

What the new umbrella review actually says

The paper is an umbrella review, a review of reviews. It collects no raw data. It gathers meta-analyses that other people already published, studies that pool the results of many earlier studies into one number, and summarises those. It appeared in June 2026 in Nutrition Research and Practice, led by Jinyoung Jeong and Dong Keon Yon.

It included 35 papers. Most of them pool observational studies; a smaller share pools randomised controlled trials, meaning real experiments with a control group. In total the review checked 135 possible links between sweeteners and health, 110 of them from observational data. The included work is mostly well made.

From the observational data, highest versus lowest intake:

The column next to the percentages matters more than the percentages. The paper rated every link itself. Of the 110 links it checked, only 30 were clear enough to be rated at all, and 20 of those 30 got the weakest rating. That is where the line you will read everywhere comes from: two thirds of the findings are weak. And everything in this table is observational data. It shows that two things occur together, not that one causes the other.

And look at the last row. The best-supported finding in the whole table is, of all things, a reassurance: less breast cancer. Nobody quotes that row. Naming the risks while dropping the reassurance is sorting by result rather than by method.

What the randomised trials in the same review found

And then there is something that made no headline at all. From the randomised trials in the same review:

The paper does not hide these results. It reports them cleanly. It simply draws no consequence from them for its own summary.

Why observational studies get sweeteners systematically wrong

Picture two people. One weighs 70 kilos, drinks water and has no reason to think about sugar. The other weighs 105 kilos, has just been diagnosed with diabetes and switches to diet cola.

Five years later a cohort study comes along. It watches a group of people over time, asks both how much sweetener they consume and checks who got sick. The result is predictable. It says nothing about whether the sweetener caused the disease.

This mechanism has a name: reverse causation. People reach for sweeteners because they have a weight or metabolic problem, not the other way round. Adjusting the statistics for body mass index, activity or education only helps so far. It captures where someone stood on the day they were surveyed, not why they switched.

One example shows how much of this is simply body weight. In 2015 Fumiaki Imamura's group pooled 17 cohorts covering 38,253 diabetes cases in the BMJ. Drinks sweetened with sweeteners were associated with 25 percent more type 2 diabetes. Then the authors adjusted for body weight. The 25 percent became 8 percent. One confounder ate two thirds of the effect.

The authors of the new review know this. It is written plainly in the paper's own section on its weaknesses:

People at higher risk of obesity, diabetes or cardiovascular disease may simply be more likely to reach for these sweeteners to compensate, rather than the sweeteners causing anything.

And it notes that study after study has said the same.

That is the most important sentence in the paper, and it is the authors saying it about their own data.

The clearest demonstration is erythritol

If you want to see how badly reverse causation can distort a nutrition study, erythritol is the teaching case.

In 2023 Nature Medicine published work by Marco Witkowski and Stanley Hazen. Among people in the top quartile of blood erythritol, the risk of heart attack, stroke or death within three years was 80 percent higher than in the bottom quartile, in a US cohort of 2,149 people. In a European cohort of 833 people it was 121 percent higher. The authors also showed that erythritol activates platelets and promotes thrombus formation in animal models.

The story travelled worldwide. Erythritol has been the dangerous sugar alcohol ever since.

Except erythritol is not just an additive. Your body makes it itself, out of glucose, in an ordinary side branch of sugar metabolism. That is not a guess. A 2017 Cornell study showed it directly, with labelled sugar molecules it could follow through the body. The same study also followed 264 young adults through their first year at university, nine months in all. Those who put on belly fat over that year had 15 times more erythritol in their blood. Those who had started with a slightly raised HbA1c, the marker for your average blood sugar, above 5.05 percent, had 21 times more.

Nobody eats 15 times more erythritol than the person next to them. So that extra erythritol did not come from food. The body made it. A high erythritol level in the blood is usually a sign that sugar metabolism is going wrong, not proof that someone ate erythritol. And that turns the scary headline around. A sick metabolism can push erythritol up and cause heart attacks at the same time. The erythritol would then be the smoke, not the fire.

The Cornell authors put it more carefully than the numbers sound. They write that the body's own production may contribute to the levels they measured, and the 15-fold and 21-fold figures come from blood samples pooled by group, not from single people. A 2023 review blames exactly this broken branch of sugar metabolism rather than erythritol from food. Worth knowing before you lean on it: its submission was approved by Amyris Inc., a company that makes erythritol.

Where the erythritol defence stops working

That is the case for the defence. Two things cut against it. The first: the platelet work in the Hazen paper cannot be explained away by reverse causation. In 2024 the same group found the same pattern for xylitol, a 57 percent higher risk in the top tertile, and in a small intervention study with ten healthy volunteers, platelet reactivity rose after a xylitol-sweetened drink in every single participant. Ten people are not an answer to a safety question, but they are not nothing either.

And the second is where the elegant erythritol argument stops working. It does not transfer to xylitol. That is stated in the Hazen group's own xylitol paper: sugar alcohols are indeed produced endogenously, but at levels more than a thousandfold below what appears in blood after consumption as a sweetener. For erythritol, endogenous production carries the explanation. For xylitol, it does not. Clearing both compounds with the same argument is too easy.

One limitation applies to both Hazen papers: the cohorts were patients attending hospital for elective cardiac evaluation, not a general population.

A 2025 paper in the European Journal of Preventive Cardiology shows particularly cleanly how far that explanation carries. In the Nurses' Health Study, 762 women who later developed coronary heart disease were compared with 762 controls. In the quarter with the highest blood erythritol, the risk was at first 55 percent higher. Then the authors adjusted for diabetes. What remained was 21 percent, a remainder that could just as well be chance. For mannitol and sorbitol the association survived that adjustment. This is exactly what it looks like when a signal is carried mostly by metabolic state rather than by the compound.

Same compound, same apparent risk, and the signal all but disappears once you account for who was already ill. That is this entire subject compressed into one paragraph.

What happens when you deliberately replace sugar?

The question you actually care about is not "are sweeteners good?" but "are sweeteners better than the sugar they replace?" Answering it takes a study that does exactly that.

They exist.

The DRINK trial, published in the New England Journal of Medicine in 2012. For 18 months, 641 mostly normal-weight children received 250 millilitres of a drink per day through their schools. Half got a sugar-free version, half a sugar-containing one providing 104 kilocalories. Neither children nor parents nor researchers knew who got which. After 18 months the children on the sugar-free drink had gained 6.35 kilograms on average, the children on the sugar drink 7.37 kilograms. Skinfold thickness, waist-to-height ratio and fat mass also rose less in the sugar-free group.

One kilogram of difference over a year and a half, in a masked trial, in children who did not know they were in a sweetener study. That is as close to a clean experiment as nutrition research gets.

One caveat comes with it, and it sits in the trial's own abstract: 26 percent of the children had stopped drinking the beverages by the end, and rather than drop them, the researchers estimated their missing measurements statistically. In a secondary analysis that folded 136 of those quitters back in with the 477 who finished, the difference between the groups roughly halved, to 0.06 versus 0.12, P = 0.06, a hair on the wrong side of the line researchers conventionally treat as a real finding. The main effect stands, but the trial is less untouchable than the secondary literature usually makes it look.

The 2022 network meta-analysis in JAMA Network Open pooled 17 randomised trials with 1,733 adults and compared three substitution strategies. This matters for generalisability: these were not adults in general but adults with overweight or obesity who either had diabetes or were at risk for it. Sweetened beverage instead of sugary beverage gave 1.06 kilograms less weight, plus less body fat and less liver fat. The other two comparisons are the interesting ones. Water instead of sugary beverage showed no significant effect on any outcome, which reflects how few trials exist. And sweetened beverage instead of water showed essentially nothing, apart from a slightly higher HbA1c and a slightly lower systolic blood pressure.

The trials that cut the other way

Now the counterweight, because substitution is no miracle cure.

Ebbeling and colleagues, also 2012 in the New England Journal of Medicine, delivered calorie-free drinks to the homes of 224 adolescents with overweight and obesity for one year. After one year, BMI was 0.57 points lower and weight 1.9 kilograms lower. After two years, one year after the intervention ended, the difference had vanished. The primary endpoint was not significant.

Two things need saying about this trial before it can be filed correctly. It was not masked, so the adolescents knew which group they were in. And strictly speaking it did not test sweetener against sugar; it tested reducing sugary drinks against habitual consumption. What it says about sweeteners is therefore indirect. What it shows very well is something else: the effect lasted exactly as long as somebody kept delivering the drinks.

And the large systematic review in the BMJ in 2019 reached a sober verdict: 56 studies, 35 of them observational. Certainty of evidence for benefit was mostly low to very low. Mostly, not throughout: for body weight in children the review graded certainty as moderate. The adult BMI effect came from two studies totalling 174 people.

The largest intervention trial currently available is the European SWEET project, which published results in Nature Metabolism in 2025 from 341 adults with overweight who, after a two-month diet, spent a year eating either sweetener products or their sugar-containing equivalents. The sweetener group held on to 1.6 kilograms more weight loss, P = 0.029. There were no differences in blood lipids, blood pressure or blood glucose, and no weight difference among the 38 children included.

And what about water?

At this point nearly every article on the subject, including earlier drafts of this one, says "water is better anyway". It sounds sensible. The direct comparisons do not support it.

Four randomised trials have put sweetened beverages and water head to head. They reached three different answers.

Two trials find no difference, one favours water, one favours sweeteners. The longest and largest, running two years with 493 participants, was explicitly designed as an equivalence trial and returned exactly that: both work.

The two that disagree started from the same place, which is easy to miss. In both trials everyone was already drinking diet drinks, and then half of them switched to water. After that the trials part ways. Madjd tested 89 overweight women in Iran on a calorie-reduced diet, and the diet-drink group got one drink after lunch, five days a week. Peters tested 303 US adults with obesity in a year-long weight-loss programme, with at least 710 millilitres of the assigned drink every day. Madjd analysed only the 62 women who finished, Peters counted everyone. And the Peters trial was paid for by the American Beverage Association, with two authors taking consulting fees from Coca-Cola. In both trials one half had to give up a drink they were used to. So part of what got measured is how hard that is. Two small trials, two answers, and neither one settles it.

The meta-analyses match that picture. The 2022 network meta-analysis found essentially nothing for sweetener instead of water. Rogers and Appleton pooled 88 intervention studies in 2021 and found 0.10 kilograms in water's favour, which is statistically indistinguishable from zero. And then there is the largest cohort analysis on the question, published in 2026 in the American Journal of Clinical Nutrition across 143,409 people.

Swap three sugary drinks a week for sweetened ones, and you gain 1.39 kilograms less over four years. Swap them for water instead: also 1.39 kilograms. Identical to two decimal places.

What follows from this. For your weight, the choice between water and a sweetened drink appears to matter considerably less than the choice between either one and sugar. Water is still the sensible default, but for different reasons: it costs nothing, it contains nothing that will still be argued about in ten years, and it needs no safety assessment. Those are good reasons. A weight advantage is not among them. Replacing sugar with sweeteners, by contrast, buys a small, real, limited advantage, for as long as the sugar genuinely stays gone. And when the switch ends, the advantage usually ends with it.

This is not a new argument. It has been running for twelve years.

If you got the impression this debate started in 2026, that is the coverage talking, not the evidence. The contradiction between observational data and randomised trials has been documented since 2014, it shows up in every major review since, and always with the same shape. The paper under discussion here is the third umbrella review in four years, not the first.

Twelve years, fifteen major papers, one single pattern. Observational data find risk, randomised trials find a small benefit against sugar. Not one of these papers resolved the contradiction.

To be fair to the new paper: it does not pretend its predecessors are not there. Its discussion names earlier syntheses and positions itself against them. So the charge is not that the predecessors go unmentioned. The charge is that a pattern which has shown up in every single one of these papers for twelve years is still handled as a side finding rather than as the actual result.

There is a paper that resolves the contradiction. It is missing.

In 2025, Applied Physiology, Nutrition, and Metabolism published a review that took on precisely the question of why cohorts and controlled trials diverge on sweeteners.

The group around Sabrina Ayoub-Charette and John Sievenpiper searched specifically for meta-analyses that had analysed their data two ways. Naive means sweetener versus everything else, without asking what it replaces. Bias-adjusted means sweetener versus a defined comparator, that is, against what it actually displaces, or against change in intake over time.

Six trial analyses and five cohort analyses qualified. The result is strikingly clean. In the trials, people who used sweeteners took in fewer calories, weighed less and carried less body fat, and that held whichever way the numbers were analysed. In the cohorts, the comparison decided everything. Analysed the naive way, sweeteners looked bad: more obesity, more diabetes, more strokes, more deaths. Analysed the fair way, sweetener drinkers are compared with the people drinking sugar instead, not with everyone else. Then the same cohorts flipped: less weight, smaller waists, less obesity, less coronary heart disease, fewer deaths.

The PubMed record for that paper is dated 22 December 2025, a good two months before the new paper was submitted and barely four months before it was revised.

Same source material. Different comparator. Different sign.

The authors rate their own certainty, incidentally: generally moderate for the trials, very low for the cohorts. That is more honest than the summary suggests.

Who funds this work

One disclosure usually goes missing in the argument about this paper, so here it is.

John Sievenpiper, Toronto, shows up twice in this article. He co-wrote the 2025 review just described, and he co-wrote the 2022 network meta-analysis quoted earlier in favour of swapping sugar for sweetener. He is also one of the most productive researchers on the topic. His own disclosure lists speaker fees and honoraria from the International Sweeteners Association, from the Calorie Control Council, a sweetener industry group, and from Nestlé. It lists research support from the Tate and Lyle Nutritional Research Fund at the University of Toronto, and Tate and Lyle makes sweeteners. And it lists food donations from Danone and other producers for running trials.

The fund people argue about most is the Nutrition Trialists Network Fund at the University of Toronto. According to the statement, it was set up with money from three donors: the Calorie Control Council, the sweetener industry group again; the Physicians Committee for Responsible Medicine, which campaigns for plant-based diets and usually fights the food industry; and the Login5 Foundation. So calling this fund pure industry money is too simple.

So what do you do with that? The method itself is sound. Comparing a sweetener against what it actually replaces, instead of against everything at once, is an accepted principle, not a trick. But the group that keeps clearing sweeteners is partly paid by the sweetener industry, and that belongs in the picture. Otherwise you have only swapped one bias for another. And the money flows in both directions.

A 2016 analysis in PLoS One sorted 31 reviews of sweetened beverages and body weight by who paid for them. Of the four funded by the sweetener industry, three reached a favourable result; of the 23 without industry funding, exactly one did. All four funded by competing industries reached an unfavourable result. In 42 percent of the reviews, the authors' financial conflicts of interest were not disclosed at all.

The fair reading: take the methodological argument seriously, and demand the same scepticism you apply to the new umbrella review from the start of this article.

How well made is the new paper?

Two things stand out from the paper's own header data.

First, the literature search ended on 20 December 2024. The paper was submitted on 28 February 2026. Fourteen months of further publication sit in between, including the naive-versus-bias-adjusted review described above.

For fairness: a December 2024 cutoff is not a protocol violation. A systematic search has to stop somewhere or it never finishes. The search is not the problem. The discussion is. If you submit fourteen months later, the discussion should account for what appeared in the meantime, especially when it bears on the central interpretive question of your own paper.

Second, the discussion speculates instead of settling the design question. Why would people eat less sugar and still weigh more? The paper's answer is that a sweet taste with no calories behind it may leave the body expecting an energy hit that never arrives. For the opposing weight results it suggests something more technical: the observational studies sorted people into obese or not obese, while the trials tracked weight as a sliding scale.

Both are legitimate hypotheses. Both are second-choice explanations as long as the most obvious one, that sick people reach for sweeteners, sits only in the fine print.

What remains is a competently executed review with a currency gap and a conclusion that does not match its own evidence grading. That is a real criticism. It is not a scandal.

The sweeteners one by one

"Sweetener" is not a substance, it is a category. These compounds are chemically unrelated and their evidence bases differ considerably.

Aspartame

The best studied and worst regarded. In July 2023, IARC classified aspartame as possibly carcinogenic to humans, group 2B, based on limited evidence for hepatocellular carcinoma, that is, liver cancer. On the same day, the joint FAO and WHO expert committee reaffirmed the acceptable daily intake of up to 40 milligrams per kilogram of body weight. Acceptable daily intake means the amount you could consume every day for life without expected harm, set with a large safety margin built in.

Reading both at once is the point. Group 2B describes the strength of the evidence, not the size of the risk. To exceed the daily intake, a 70 kilogram adult would need, per WHO, more than 9 to 14 cans of a diet soft drink per day, assuming 200 to 300 milligrams of aspartame per can and no other source.

The French NutriNet-Santé cohort of 102,865 adults reported in 2022 a 15 percent higher overall cancer risk with higher aspartame intake and a 22 percent higher breast cancer risk. The same cohort reported a 17 percent higher risk of cerebrovascular events. Both are observational, with exactly the limitation this article is about. The authors themselves name reverse causation as a weakness of their work.

Acesulfame K

In the same French cohort, acesulfame K was linked to a 13 percent higher cancer risk and a 40 percent higher risk of coronary heart disease, and sucralose to a 31 percent higher risk of coronary heart disease, though that last figure is statistically borderline.

For acesulfame K there is a 2025 development almost nobody noticed: EFSA re-evaluated the compound and raised the acceptable daily intake from 9 to 15 milligrams per kilogram of body weight, because rat studies showed no adverse effect even at the highest dose tested. Estimated EU intake sits below that in every population group. An agency loosening a limit on new data is just as much a data point as one tightening it.

Sucralose

Start with the reassuring part. EFSA re-evaluated sucralose from scratch in February 2026 and found no safety concern at current uses. The only open question is heat. Beyond that there are two signals that do not come from cohorts, and both are small and early. The first comes from a 2022 randomised trial in 120 healthy adults. All four sweeteners tested changed the oral and gut microbiome and the metabolic products in the blood, but only saccharin and sucralose measurably worsened the glucose response. Keep the size in mind: that was only about twenty people per sweetener, for two weeks, at doses below the daily limit. Germ-free mice given the microbiomes of these participants then showed the same glucose responses as their donors.

The second signal is a 2023 toxicology review of sucralose-6-acetate, an impurity and breakdown product of sucralose. In cell tests it damaged DNA, pushed gene activity towards inflammation and oxidative stress, and together with sucralose it weakened the barrier that seals off the human colon. Those are cells in a dish, not people. A signal to watch, not a reason to panic.

The third point has nothing to do with cohorts and everything to do with your oven. Germany's Federal Institute for Risk Assessment assessed heated sucralose in 2019 and said it again in February 2026: above 120 degrees Celsius the molecule breaks apart and loses chlorine, and chlorinated compounds can form, some of them harmful or carcinogenic. Baking, frying and deep-frying reach 120 to 150 degrees. This is about your oven, not about your drink.

EFSA re-evaluated sucralose in the same month, and on eating it the verdict was clear: acceptable daily intake unchanged at 15 milligrams per kilogram of body weight, no genotoxicity concern, intake below that limit in every population group. The one thing the agency could not sign off was the makers' application to extend sucralose to more fine bakery wares, because the heat questions are still open. Factories heat briefly, which limits the chlorination. In a home kitchen, EFSA writes, the risk cannot be excluded. So the advice has not changed since 2019 and now comes from two agencies: do not heat sucralose to baking or frying temperature, or sweeten after heating. A precaution, not a proven harm.

In practice the distinction is simple. Sucralose-sweetened syrup over finished quark is fine. A sucralose-sweetened powder that goes into the batter and then into a 180 degree oven is exactly the case the institute means. This applies in particular to products that are sweetened with sucralose and sold for baking in the same breath: baking mixes, syrups and protein powders.

Stevia and steviol glycosides

The friendliest evidence base. In the new umbrella review, stevia clearly lowered blood glucose in randomised trials, with medium certainty. In the 2022 microbiome trial, stevia altered bacterial composition but, unlike saccharin and sucralose, did not impair the glucose response.

Erythritol and xylitol

Sugar alcohols, not sweeteners in the strict sense. The evidence is described in detail above. Short version: for erythritol the observational signals are highly vulnerable to distortion, because your body produces the compound itself in relevant amounts. For xylitol that does not hold, endogenous production is a thousandfold below intake, and the cohort signals remain unexplained. For both, the platelet and intervention data are weaker but cannot be explained away. If you consume large amounts daily, through sugar-free baked goods or protein bars, restraint is defensible.

Note that the WHO sweetener guideline explicitly does not cover sugar alcohols. Anyone reading "the WHO warns against erythritol" is reading something false.

Sorbitol

The one compound in the review with a clear benefit beyond weight: slightly better dental health, with medium certainty.

What does the WHO actually say?

In May 2023 the WHO published a guideline that the press read almost universally as a warning. What it actually contains:

• The WHO recommends against using non-sugar sweeteners to control body weight or reduce the risk of noncommunicable diseases.
• The recommendation is conditional. In WHO terminology that means the evidence does not support a strong recommendation and policy decisions require country-specific discussion.
• It covers acesulfame K, aspartame, advantame, cyclamates, neotame, saccharin, sucralose, stevia and steviol glycosides.
• It does not cover sugar alcohols, nor sweeteners in personal care products and medicines; the guideline names toothpaste, skin cream and medication explicitly.
• It applies to everyone except people with pre-existing diabetes.

So the guideline is not a safety warning. It is a statement about benefit: sweeteners do not work as a long-term weight strategy. On whether they cause harm, it is considerably more cautious than the coverage implied.

What this means in practice

If you are replacing sugar, sweeteners are the better choice. That is the one comparison with masked trials and hard endpoints behind it, and it comes out in favour of the sweetener consistently. One kilogram over 18 months in children, about one kilogram over a few months in adults.

If you are choosing between water and a sweetener, the decision is smaller than you think. Four randomised trials compared them directly. Two found no difference, one favoured water, one favoured the sweetener. Across the meta-analyses the difference averages out to zero. Pick water because it costs nothing and carries no open questions, not because it makes you thinner.

Sweeteners are not a free pass. The effects are small and disappear when the switch ends. Anyone who takes the saved calories back elsewhere has gained nothing. That is a property of the calorie balance, not of the sweetener. More on that in our overview of how many calories you need per day.

The permitted amounts are generous. Nine to fourteen cans of diet soft drink a day to reach the acceptable daily intake for aspartame. If your consumption sits anywhere near one or two drinks, dose is not your issue.

Do not bake with sucralose. Above 120 degrees Celsius it decomposes. The German risk assessment institute advises against heating sucralose-containing foods to baking and frying temperatures, and in February 2026 EFSA could not rule on extending sucralose to more baked goods precisely because those heat questions are open. Sweetening after heating solves it. Everything else about sucralose is considered safe at normal amounts.

Switch compounds if one bothers you. These substances are chemically unrelated. If one of them bothers you, say the lab signals around sucralose, then switch. Stevia currently has the friendliest evidence base, and switching costs nothing.

With diabetes, the WHO recommendation explicitly does not apply. So it is not a rule for you. If you track your blood glucose you will see the difference between sugar and sweetener directly anyway. What that looks like in practice is in our guide to glucose and blood sugar and our overview of continuous glucose monitoring without diabetes.

The real subject is still sugar. This entire debate concerns a substitute for something you are better off reducing. What an eating pattern that addresses the root looks like is in our guide to the longevity diet.

Five signals that tell you whether a study is worth anything

You do not need to be a methodologist. Five questions handle most nutrition headlines.

1. When did the literature search end, and when was the paper submitted? Both sit in the methods section. More than a year apart, check whether the discussion closes the gap.
2. Cohort or randomised trial? Only the second can show a cause. When one paper contains both and they disagree, the experiment wins.
3. Is reverse causation taken seriously, or does it appear only in the fine print while the conclusion still sounds causal?
4. Does the conclusion match the grades the paper gives its own findings? Weak findings under a confident summary do not fit together.
5. Who funds it, on both sides? Check the study you like as carefully as the one you do not. That is the harder check.

To keep that from being a platitude, here it is applied to this article. The 2022 network meta-analysis cited above in favour of substitution comes from the same Toronto group whose conflicts of interest are set out at length in this text. That does not invalidate the work. But if rule 5 is worth anything, it has to apply to the evidence supporting your own thesis too. That is what "the harder check" means.

These five questions clear most viral nutrition news in five minutes. More of them, including the recurring patterns in the longevity scene, are in our guide to longevity and pseudoscience.

The bottom line

The new umbrella review is neither a revelation nor a scandal. It is a competently assembled summary whose summary does not match its own data.

Its observational data show what observational data on sweeteners have shown for twelve years: people who consume sweeteners are sicker. Its randomised data show what randomised data have shown for twelve years: people who replace sugar with sweeteners gain slightly less weight.

Both are true. Only the second answers the question you have.