“Scientifically proven”: what those two words really mean

“Scientifically proven.”

Two words with a remarkable ability to put everyone at ease.

The problem is that they can cover almost anything: an experiment on cells, a trial in twenty people, a negative study with a positive secondary endpoint, or a meta-analysis of dozens of trials.

Technically, there is science in all four cases.

But not the same level of evidence.

First question: what was actually studied?

A cell in a laboratory can help us understand a mechanism. An animal can teach us how a whole organism responds. But when we want to talk about a benefit in humans, human studies become essential.

It is a simple principle that already prevents a lot of overambitious statements.

Second question: was there a comparison group?

Thirty people take a supplement for eight weeks. Twenty say they feel better.

Interesting. But we do not yet know why.

Between the beginning and the end, sleep, diet, stress, expectations or simply natural fluctuation can change.

That is why randomised controlled trials are so useful: they allow the product to be compared with a placebo or another intervention, with participants allocated at random.

Double blinding adds another layer of protection against certain biases. Not perfection. A better method for answering the question.

Third question: what did researchers decide to measure before they started?

A study can measure hydration, elasticity, wrinkles, satisfaction, biomarkers, sleep, fatigue and ten other things.

The more outcomes you measure, the greater the chance of obtaining a statistically positive result somewhere.

The primary endpoint therefore matters: it is the central question the trial was designed to answer.

If the primary endpoint is negative but a small secondary endpoint turns positive, we can discuss it. But we should not rewrite the aim of the study afterwards.

The CONSORT 2025 recommendations specifically strengthen transparency requirements for reporting randomised trials.

“Statistically significant” does not mean “impressive”

A result can cross a statistical threshold while representing a tiny difference in real life.

Conversely, a result that does not cross that threshold is not automatically proof that nothing is happening.

You need to look at effect size, confidence interval, sample size and context.

The p-value is one piece of information. It is not a “works” stamp.

Fourth question: was it really the same ingredient?

This is one of our favourite topics because it changes so much in supplements.

A study of a proprietary saffron extract at a specific dose does not become a study of “saffron” in general.

A study of a specific collagen hydrolysate does not validate every collagen powder.

A study of 200 mg of ExceptionHYAL® Star documents that material and that dose. It does not automatically document every hyaluronic acid on the market.

The same ingredient name does not mean the same material. And the same material does not mean the same dose.

Fifth question: does the product dose resemble the study dose?

It is the simplest check, yet one of the most frequently forgotten.

If a study uses 2.5 g and the product provides 250 mg, the publication may perhaps inform our understanding of a mechanism or the category.

It does not allow us to calmly announce that the product will reproduce the same result.

At Eclo, that is why we carefully distinguish the raw material from the finished product.

For example, our Hyaluronic Acid 200 mg formula contains 224.8 mg of ExceptionHYAL® Star to provide 200 mg of hyaluronic acid. The published trial on the material evaluated 200 mg per day of full-spectrum hyaluronan. It is still not a clinical study of the finished product.

Sixth question: who funded the study?

Industry funding does not automatically make a study bad.

Manufacturers are often the ones who fund research on their own raw materials. Without them, many ingredients would simply be less studied.

But funding remains important information.

Collagen illustrates this well: a 2025 meta-analysis found a favourable overall result, then less clear conclusions in some subgroups of studies that were not industry-funded or were of higher methodological quality.

The right reaction is not “industry = false”. It is: does the result hold up when we look at the studies in several different ways?

Seventh question: has the result been replicated?

One positive study is a signal.

Several teams, several populations and consistent results give us more confidence.

That is also why meta-analyses can be useful, provided we look at what they combine.

Pooling weak, very different studies on heterogeneous materials does not magically create certainty.

Our nine-question reading framework

“We do not know yet” can be an excellent conclusion

A brand does not need to choose between “miracle” and “no effect”.

It can say: well-established mechanism, encouraging trials, heterogeneous results, study on a different material, data still limited.

That vocabulary is less spectacular.

More importantly, it lets us distinguish between a formula supported by science and a formula merely decorated with science.

A publication is not an argument from authority. It is one piece of the dossier.