Does Coffee Help You Lose Weight? Ten Trials Poured Real Coffee. No.

Not as a drink. Ten randomised trials have handed people actual coffee and weighed them afterwards; pooled together, coffee did not beat a placebo [5]. The supplements made out of it are a separate question, and a thinner one.

And the most famous version of this claim — the one that sold green coffee bean extract to the world — ended with a $3.5 million cheque to the Federal Trade Commission and a retracted paper [2][3].

Let’s do the real evidence first, then the crime.

What 13 trials actually found

A 2019 review pooled thirteen randomised trials covering 606 people and asked a narrow question: does taking caffeine change body weight [1]?

All three measures moved in the right direction — weight, BMI and fat mass — and none of the results were compatible with “nothing at all”.

Better still, there was a dose-response relationship, which is one of the things that separates a real effect from a coincidence. More caffeine, more reduction. Specifically, every doubling of caffeine intake corresponded to about 22% more weight reduction, 17% more BMI reduction and 28% more fat-mass reduction [1].

That sounds terrific. Here is why it settles nothing.

The trials differ by more than chance can explain

Every meta-analysis reports a number called I-squared, and it is often the most informative figure in the paper. None of the nine first-page web results we read for “does coffee help you lose weight” on 7 October 2026 quotes it, though three of them cite this analysis.

I-squared asks: how much of the spread between these trials’ results is more than the ordinary noise you would expect by chance? Zero percent means they differed no more than luck allows. It does not mean they agreed: trials with wide enough margins of error can score zero and still point different ways. Above about 75% is usually described as considerable.

This analysis came in at 91.2%, 93.0% and 94.0% for the three outcomes [1].

That is about as high as it goes. It means nearly all of the spread between the trials’ results is real rather than luck. What it does not tell you is which way they pointed. Trials that all found weight coming off, some a lot and some a little, can score this high; so can trials that split, some finding a loss and some a gain. The per-trial results that would show which sit in the paper’s full text [1], behind the paywall we describe below, and we have not seen them. Either way, an average laid over differences that large tells you little about what the next trial would find.

There is a second problem, quieter but just as limiting. The pooled figure is a regression coefficient on a double-log scale [1]. In plain terms: it describes the shape of the relationship, not an amount. It says the reduction gets about 22% bigger each time the caffeine dose doubles, and it never says what the reduction was to begin with. So there is no pound-count to read off it, and a label that quotes you one did not get it from here.

And the caffeine came mostly in mixtures, not in coffee

Worth saying plainly, because the whole internet slides between the two.

The trials above dosed caffeine, which is the question that paper asked — its answer is a curve running across caffeine doses [1]. In what form the caffeine arrived is set out in the paper’s own table of its thirteen trials, and that table sits inside the full text, behind the journal’s paywall. The publisher’s page gives up the abstract and nothing more. Its PDF, PubMed Central, Europe PMC, Semantic Scholar and the open-access indexes all refused in September; an archived copy of the page, four university repositories and a fresh search for a free copy turned up no table on 23 October 2026. We still have not seen it.

What is open is the paper’s reference list, and we went through it. Of its 38 entries, thirteen are human trials that gave caffeine and then weighed people or measured their body fat: the same number of trials the review says it pooled. We read each one’s own abstract [7]. Seven gave the caffeine together with ephedra, the herb also called ma huang, or with ephedrine, the stimulant it contains: as herbal supplements, as ephedrine-and-caffeine combinations, or in a supplement with vitamins, minerals and omega-3 fats. Two paired it with green tea catechins, and one with bitter orange and St John’s wort. Two were coffee: one compared an instant coffee enriched with chlorogenic acid against ordinary instant coffee, and the other had 33 volunteers drink a special brew for four weeks, with no separate control group. One, in 60 women on a diet, gave caffeine on its own, dissolved in water [7].

That is our reading of its reference list, not of its table, and the two need not match. A review can cite a trial it did not pool, and these thirteen enrolled 807 people between them against the review’s 606 (our arithmetic from their abstracts), a gap that counting only those who finished, or only some of the arms, could account for. Only the table can settle it. If the lists are the same, one trial in thirteen tested caffeine by itself.

Why does that matter? Because in seven of them the caffeine arrived with a second stimulant. Whatever weight came off in those trials belongs to the pair, and a curve drawn across their caffeine doses cannot say which of the two did the work. That is our reasoning, not the review’s. Nor is the pair something you can buy as a supplement in the US: in 2004 the Food and Drug Administration ruled that dietary supplements containing ephedrine alkaloids “present an unreasonable risk of illness or injury”, a finding that, in its words, “results in a prohibition on the distribution and sale” of them, and the rule is still in the federal code [8].

Coffee, for its part, is caffeine plus several hundred other compounds, in a dose that varies with the bean, the roast, the grind and how heavy-handed you were this morning.

So the honest translation is: the caffeine in your coffee has thirteen trials behind it, whose results differ by more than chance can explain, and by our reading of the paper’s references most of them gave it in mixtures, seven with a stimulant that US supplements may no longer contain.

So has anybody just given people coffee?

Yes. And until today this page did not tell you so.

In 2024 a network meta-analysis — a pooling that compares several options at once, including ones nobody ever tested head to head — went looking for randomised trials of tea and coffee drinking. Twenty-two trials, 1,710 people, ten weeks on average. Extract supplements were thrown out on purpose: this was the cup, not the capsule [5].

Ten of those twenty-two poured actual coffee — instant, boiled, filtered, decaffeinated — at two to six cups a day. The shortest ran two weeks. The longest gave 126 adults with overweight four cups of instant coffee a day for twenty-four weeks [5].

Coffee against a placebo drink came out at 1.4 lb (0.64 kg) — and the range of plausible answers ran from 4.2 lb lost to 1.5 lb gained (−1.92 to +0.66 kg) [5]. When a gain is still inside the range, “does nothing at all” has not been ruled out.

The authors put it in one line. Coffee, decaffeinated coffee, green coffee, black tea and sour tea “were not effective for weight loss” [5]. Green tea was the only drink in the whole network that was, and only just.

So your actual cup of coffee is not short of evidence. It has been handed to people, it has been pooled, and the pooling could not tell it apart from a placebo drink.

Now the crime

In 2010 a Texas company called Applied Food Sciences paid researchers in India to run a trial on its green coffee extract [2].

Announcing the settlement, the FTC said it charges that the study’s lead investigator “repeatedly altered the weights and other key measurements of the subjects, changed the length of the trial, and misstated which subjects were taking the placebo or GCA during the trial” — GCA being the company’s green coffee product [2]. That wording is the press release’s. The complaint itself says the investigator “confused which subjects took either the placebo or GCA at various points during the trial” [2].

Read that again. The weights were altered. In a weight-loss study.

When the investigator could not get it published, the FTC says the company hired two researchers at the University of Scranton to rewrite it [2]. It was published in 2012 [3].

Off the back of that paper, the company claimed its product caused users to lose 17.7 pounds, 10.5% of body weight and 16% of body fat — with or without diet and exercise — in 22 weeks [2].

The FTC’s position was that the study was, in its words, so “seriously flawed” that no reliable conclusion could be drawn from it at all [2].

The extract was then featured on a daytime television show, and here the FTC is scrupulous in a way worth repeating: the company “played no part” in that happening. What it did do was issue a press release riding the publicity afterwards — one claiming subjects had lost weight “without diet or exercise”, when the subjects had in fact been instructed to restrict their diet and exercise more [2].

September 2014: $3.5 million, and an order requiring two adequate, well-controlled human trials before it makes another weight-loss claim [2]. The Commission voted 5–0.

The paper was retracted. PubMed now files it under the publication type “Retracted Publication” [3], which is the permanent version of an asterisk.

Does that mean green coffee does nothing?

No — and this is the part where it would be easy, and wrong, to let the fraud do our arguing for us.

A 2023 meta-analysis looked at green coffee extract supplying at least 500 mg a day of chlorogenic acid, the compound green coffee is actually sold on. Three trials, 103 people total. Average weight difference: down 2.9 lb (1.30 kg), with the plausible range running from about 1.1 lb to 4.6 lb [4].

Those three trials scored an I-squared of 0% [4], meaning their results differed no more than chance allows. The review’s own chart shows what sits behind that. All three put the extract ahead, two of them by more than chance. Those two carry 99% of the pooled weight and landed close together, at 3.3 lb and 2.6 lb (1.48 and 1.16 kg). The third, the one-week trial, has a range running from 18.8 lb lost to 16.8 lb gained (−8.53 to +7.61 kg), so wide that almost any result would have sat inside it [4]. Now look at what those trials were. They ran one week, eight weeks and eight weeks. The one-week trial had sixteen people in it. Two of the three were in patients rather than in ordinary volunteers — one group with fatty liver disease, one with metabolic syndrome, both in Iran. One trial was rated high risk of bias. And the reviewers graded their own certainty Low [4].

Step outside those three and the picture changes. The same paper walks through the other poolings of green coffee extract, and their I-squared figures come in at 83%, 92.4%, 97% and 99% — which is to say that in every one of those poolings, most of the spread between the trials’ results is more than chance would produce. The largest of those poolings found no significant change in body weight [4].

So: a small effect, from a very small and very short evidence base, graded low certainty by the people who pooled it. About three pounds. A 2024 umbrella review — a pooling of the poolings — put green coffee extract at 2.7 lb (1.22 kg) across five meta-analyses, so the neighbourhood holds up [6].

And note what this is not. It is not a finding about coffee. Green coffee extract is a capsule of compounds from the unroasted bean, and the 2024 network analysis that pooled coffee as a drink deliberately left extract supplements out of it [5]. Three pounds from a capsule is still a long way from 17.7 pounds without diet or exercise, and that gap is the whole business model.

A fabricated study does not make an ingredient useless. What it establishes is narrower, and bad enough: at the time the company was making those claims, it had no honest evidence for them.

What to actually do

Drink coffee if you like coffee. It is not a weight-loss intervention — and that is no longer an inference from the caffeine trials. It is the direct result: twenty-two randomised trials of tea and coffee drinking, pooled, and coffee was not effective [5].

If a product puts a specific pound-count on a coffee compound, ask which paper it came from. The caffeine pooling contains no pound-count at all [1]. The green coffee extract pooling contains about three pounds, over eight weeks at the outside, in 103 people [4]. Neither of them contains 17.7.

And “with or without diet and exercise” is the phrase to walk away from. It cost one company $3.5 million, and the subjects in the study behind it had been told to diet and exercise [2].