Diet Soda Matched Water for Weight. The Word “Equivalent” Didn’t

For two years, 247 adults in and around Liverpool drank at least two 330 milliliter servings of diet soft drinks a day, roughly two cans. Another 246 were told to drink water and give up sweeteners entirely, even in their tea. At the end, the diet drink group had lost 4.8 kilograms on average and the water group 3.7. The researchers call that equivalence. It is a textbook case of a trial built to show no difference, and those trials play by rules most readers are never told about.

What SWITCH actually tested

The trial, called SWITCH, ran at the University of Liverpool and enrolled adults aged 18 to 65 with a body mass index between 27 and 35: people with overweight or mild obesity who already drank cold beverages regularly. Everyone joined the same group weight loss program, with weekly sessions alongside a nutritionist for 12 weeks, monthly sessions until week 52, then a full year on their own and a final weigh in at week 104.

The diet drink group picked from twenty branded products, sweetened mostly with aspartame, acesulfame potassium and sucralose. The water group drank bottled water plus any other water they liked. Nobody was blinded. The second year results were published online in the British Journal of Nutrition on August 4, 2026, and the university publicized them on September 29, describing SWITCH as the longest and most comprehensive randomized trial of these drinks so far.

The result, in kilograms

Participants started at about 90 kilograms. At week 104, the water group was down 3.7 kg and the diet drink group 4.8 kg, a gap of 1.1 kg in favor of diet drinks that did not reach statistical significance. That is about 4% to 5% of starting weight. Waists shrank by 4.5 and 5.5 centimeters. In a small subgroup scanned by DXA, both arms lost about 4 kg of fat. Both groups regained some weight after peaking around weeks 36 to 44.

A few markers moved differently. HbA1c, a gauge of average blood sugar, and diastolic blood pressure improved more with water, while total cholesterol edged up with diet drinks. The authors judge none of these gaps clinically meaningful, and the cholesterol and blood sugar tests covered only 14 to 16 people per group, far too few to settle anything in either direction.

How to read a trial that finds no difference

Most trials ask whether a treatment beats its comparator. When the answer comes back as "not significantly," that is not proof the two are the same. The trial may simply have been too small or too noisy to see a real gap. It is the same trap I flag in my guide to reading a health study in 5 minutes: the p value tells you about the experiment as much as about the world.

A trial meant to show sameness has to work harder. The standard approach is to declare in advance the largest difference that would not matter, called the equivalence margin, and then show that the entire confidence interval for the difference fits inside it. If it pokes outside the margin, equivalence has not been demonstrated, whatever the p value says.

SWITCH used a looser rule. Its paper defines equivalence as a two sided p value above 0.05, which is another way of saying no significant difference. The trial was sized to detect a 1.5 kg gap, so that figure is the closest thing it has to a margin. The 90% confidence interval for the difference runs from 0.2 kg in favor of water to 2.4 kg in favor of diet drinks. Against 1.5 kg, a comparison that is mine and not the authors', the upper end spills over. A version of the analysis that fills in missing participants statistically reaches 1.7 kg, also over. A cruder method that carries each dropout's last weight forward stays inside, at 1.4 kg.

Read plainly, the numbers rule out water being meaningfully better for weight. They do not rule out diet drinks being somewhat better.

The looser rule also has a strange side effect. At week 52, with more people still enrolled, the same trial found diet drinks significantly ahead and reported the two arms as not equivalent. Under a definition based on the p value, the more participants a trial keeps, the harder it becomes to call two things the same.

When sameness means no significant difference, every dropout makes the drinks look a little more alike.

The limits that matter

Dropout is the big one. Of 493 people who started, 220 reached week 104, about 45%. The design had assumed 27% attrition. The second year was a voluntary extension added by protocol amendment, and the trial's ClinicalTrials.gov record, NCT02591134, lists weight change at 52 weeks as its primary outcome. Completers resembled the full group at baseline, but people who last two years in a weight program are rarely a random slice of those who start.

Then the money. The American Beverage Association funded the trial and paid for the medical writer who helped draft the paper, and an association representative was allowed to review the manuscript. The authors state the funder had no role in design, analysis or interpretation. Industry funding does not make a result wrong. It does tell a careful reader where to look first, and the definition of "equivalent" is exactly that place.

Finally, scope. One site in England, no race or ethnicity data, healthy adults inside a structured program, and payments of £300 for finishing the first year and £100 more for the second. The comparison was diet drinks against water, not against sugary drinks, and the outcomes were weight and a handful of markers, not diabetes, heart attacks or cancer.

What comes next

The argument now moves to guidelines. A 2023 World Health Organization guideline advised against using sweeteners other than sugar for weight control, a position the SWITCH paper says rested mainly on uncertain observational studies. In the university's release, lead author Joanne Harrold said she hopes the evidence informs future WHO guidance. The authors say anonymized participant data will be available on reasonable request, which means an independent team could rerun the primary analysis against a margin set in advance.

I come away with a narrower verdict than the coverage. For people trying to lose weight who already reach for diet soda, this trial offers no sign the habit undermines the effort; if anything, the numbers lean the other way. But "equivalent to water" is a label the statistics do not quite earn, and it was attached by a study the soda industry paid for. The lesson I would keep travels well beyond sweeteners: whenever a study announces that two things are the same, look for the margin. If there is none, what you are reading is a trial that failed to find a difference, which tells you as much about the trial as about the drinks.