Look, I hear you. One minute collagen peptides are great for skin health. The next, they’re a complete waste of money.
Thankfully, two studies have given us a much clearer picture: a 2026 meta-analysis, and a trial from here in New Zealand that almost no one has talked about yet.
Table of Contents
- Why I Went Back Through the Trials
- How We Got Here
- What the New Paper Got Right
- Where the New Paper Went Too Far
- The Other 2026 Meta-Analysis
- Collagen or Just More Protein?
- The New Zealand CHAMP Trial
- Joints, Tendons and Muscle
- Where This Leaves Collagen
- References
Why I Went Back Through the Trials
My re-analysis of the collagen research started a couple of weeks ago, when a hornet’s nest erupted on social media when I saw this post:
“A new meta-analysis finds no good evidence for benefit of supplementing with collagen for skin aging.” [1]
But not all is as it seems, as I’ll explain shortly.
The analysis was blunt with its message: the big studies claiming collagen helps your skin were unreliable because of inaccurate reporting of the data [2].
That mattered to me personally, because I designed a supplement that contains collagen. So I went through the trials myself to see if I should indeed remove it.
Here’s what I found.
How We Got Here
For years, the individual collagen trials looked promising. Women took collagen powder or a placebo for two or three months, and their skin was measured with machines. Over and over, the collagen groups came out ahead on hydration, which is how much water the skin holds, and elasticity, which is how well it springs back.
Hydration is usually measured with a small probe pressed against the skin. Elasticity is often measured with a suction device that gently pulls the skin up and measures how far it springs back.
Then researchers combined those trials. A meta-analysis is a study that pools lots of smaller trials into one big result. A 2021 meta-analysis found a benefit [3]. A 2023 one did too, and it was confident about it [4].
Then, in 2025, a third meta-analysis split the trials by who paid for them. Shockingly, it found that only the company-funded trials showed a benefit [5].
The betrayal. Or so it seemed.
I made a video pushing back on that conclusion. I pointed out that the result for the independent trials only just missed statistical significance. And if we look at the individual studies that were graded as high quality, 7 out of 9 showed a positive impact for skin health from collagen peptides.
That’s where things stood when the new 2026 paper landed. It concluded that there was actually no evidence of benefit from collagen supplementation, because of “inaccurate reporting of data from primary studies and inappropriate combinations of measures likely to give rise to spurious results” [2].
What the New Paper Got Right
So were we indeed tricked into buying collagen supplements? Let’s take a deeper look.
The 2026 paper came from David Robert Grimes, a biostatistician at Trinity College Dublin, and Anjali Mahto, a dermatologist in London. They went back through the two positive meta-analyses, the 2021 and 2023 ones, and checked how the numbers had been entered.
They found real problems [2].
One of the errors is easiest to see with an example. Bear with me on this one, because it does illustrate the point perfectly.
Imagine a company selling a pill that claims to make men… bigger. Down there.
Researchers have actually measured this. A 2015 review found the average erect length was about 13 centimetres, and most men (68% of men, to be precise) land within about 1.7 centimetres either side of that [6]. That number has a name: the standard deviation. It tells you how much people naturally differ from each other.
Now the company runs a trial. A hundred men take the pill, a hundred take a placebo. At the end, the pill group is 2 millimetres longer on average.
Is that a big effect? Well, if we compare it with the standard deviation, which is 1.7 centimetres, a 2 millimetre effect is tiny. Nobody’s partner is going to notice.
But many studies report a second, much smaller number instead, called the standard error. It doesn’t tell you how different men are. It tells you how precisely you’ve pinned down the group’s average. With a hundred men, that’s about 1.7 millimetres.
Now make one mistake. Compare that same 2 millimetres with the standard error of 1.7 millimetres, instead of the standard deviation of 1.7 centimetres, and suddenly the pill looks ten times more powerful.
Same men, same tape measure, same 2 millimetres. All that changed is which number you compared it with.
That’s what happened when the earlier meta-analyses pooled some of the collagen trials. The standard error was entered where the standard deviation belonged, and small differences were blown up into big ones.
The second problem was mixing measurements that aren’t the same thing. Skin elasticity was measured with different machines, in different units, on different parts of the body, and the meta-analyses pooled them all together.
This is the problem that I pointed out in my 2025 video.
They also made two other points I agree with: most of these trials were paid for by collagen companies, and the comparison was against a placebo, so we couldn’t tell if this was just a benefit from extra protein intake or a specific benefit from collagen peptides. Hold on to that second one. We’ll come back to it.
But then I got to their conclusion.
Where the New Paper Went Too Far
To reach “no evidence of benefit”, Grimes and Mahto leaned on one specific test: do trials that used more collagen show bigger effects? They found no clear pattern [2].
But that’s a different question. Imagine testing whether umbrellas keep you dry by checking whether bigger umbrellas keep you drier. If a large umbrella works no better than a medium one, you haven’t shown that umbrellas don’t work. You’ve only shown that size doesn’t seem to matter much.
It’s not about the size of your equipment, it’s how you use it.
And across trials that used different products, in different countries, measured with different machines, a dose pattern is very hard to see even when an effect is real.
Then I found something more surprising. The authors published their full dataset, which is excellent practice, and they deserve real credit for it. So I checked each row I could against the original papers.
In eight trials, the collagen and comparison groups were the wrong way around.
One example: a trial in older adults in hospital in Japan. The people taking collagen finished with an elasticity score of 0.74. The comparison group, who got their usual hospital meals, finished at 0.66 [7].
In the new paper’s dataset though, those numbers are swapped. Collagen gets 0.66, and the comparison group gets 0.74.
Another, from Thailand. Four weeks after people stopped taking collagen, cheek elasticity was 0.56 in the collagen group and 0.48 in the placebo group [8]. The new dataset has those reversed too.
There were a few other errors as well.
So what happens when you fix them? The results are reported as effect sizes. As a rough guide, 0.2 is small, 0.5 is moderate and 0.8 is large.
For hydration, their own data, pooled the usual way, already showed a benefit of about 0.5, and correcting it barely changed that. For elasticity, the errors mattered.
Before correction, the result was 0.37, and the effect could have been zero. After correction, it was about 0.6, and it reached statistical significance. Even using only one type of elasticity measurement, from one kind of device, it was about 0.6.
So we see benefits for hydration and elasticity once the data errors are corrected. But don’t take my word for it.
The Other 2026 Meta-Analysis
And there’s another 2026 meta-analysis, published in July, a couple of months before the Grimes paper. It pooled 35 trials and more than 2,500 people, and it set out to deal with the problems in the earlier analyses, like a few extreme results pulling the averages around [9].
For hydration, it found 0.44. For elasticity, 0.62, after setting aside two trials with extreme negative results. That’s very close to what I got by correcting the Grimes data.
Collagen also reduced how much water escapes through the skin. And the benefit took time: elasticity only clearly improved in trials that lasted at least 12 weeks [9].
But it found no effect on wrinkles [9].
Why?
The analysis postulates that the discrepancy between the improvements in hydration and elasticity compared to the lack of structural changes in wrinkles is likely due to time.
While oral peptides can rapidly stimulate hyaluronic acid synthesis in the outer layer of the skin to quickly improve hydration and barrier function, the reduction of visible wrinkles requires extensive and long-term reorganization of the deeper structures of the skin [9].
Basically, it’s possible that these short-term studies didn’t go on for long enough to detect an improvement in wrinkles.
And aside from correcting for errors, why does this 2026 meta-analysis suggest benefits, whereas the 2025 meta-analysis didn’t?
First, it included several trials published after 2024, which the earlier 2025 analysis didn’t have. And the authors used a different statistical approach [9].
What about concerns regarding industry funding? Here’s what the 2026 analysis said:
“Concerns regarding the influence of industry funding have also been examined. Independently conducted trials reported effect sizes comparable to those observed in sponsored studies, and multiple approaches to detecting publication bias suggested that preferential reporting was unlikely to meaningfully alter the overall conclusions.” [9]
Collagen or Just More Protein?
OK, fine. Say there’s a benefit for skin hydration and elasticity. How do we know that we need collagen peptides, rather than regular protein, to get those benefits?
After all, collagen peptides are just short chains of amino acids, and amino acids make up protein.
Well, in cell experiments, we consistently see that collagen peptides stimulate fibroblast activity, augment type 1 collagen building, and reduce collagen breakdown [9]. And we can see that collagen peptides are absorbed from our gut and enter the bloodstream.
That’s all nice in theory, but what about human studies that directly compare collagen peptides to protein?
To answer that, you need trials that compare collagen with the same amount of another protein. There are very few. One, from Korea, compared collagen with the same amount of casein, a milk protein, and elasticity improved more with collagen [10]. But the collagen group also got astaxanthin, a pigment from algae, so we can’t be sure which ingredient did the work.
The second is a trial in people with serious burns. Men received either 36 grams of collagen or 35 grams of soy protein a day for four weeks. Of the 31 who finished, every patient in the collagen group had fully healed, compared with fewer than half in the soy group [11].
That’s a small pilot trial, and a burn is very different from normal skin aging. But it’s a hint that collagen might do something protein alone doesn’t.
The New Zealand CHAMP Trial
Which brings me to the New Zealand trial.
It’s called the CHAMP study, and it was done at Massey University in Palmerston North. Women aged 50 and over took 15 grams of collagen, 15 grams of milk protein, or a placebo powder with the same calories, every day for about four months. It was registered before it started and funded through a New Zealand government science programme. Two New Zealand food companies sponsored it and paid for the collagen to be made, but they had no role in designing or analysing the study [12].
Its main skin measure was dermal thickness, the depth of the layer of skin that holds collagen. On the inner forearm, skin got significantly thicker within the collagen group, but not within the milk protein or placebo groups [12]. That’s skin that rarely sees the sun, so it reflects aging itself rather than sun damage.
I went through every skin-thickness and hydration measurement in the thesis, ten in total, at different parts of the body. Collagen came out ahead of placebo on 9 of the 10. And it came out ahead of milk protein on 7 of the 10.
Now the caveats. Only one of those differences between groups was clearly statistically significant, on the upper inner forearm.
The trial was designed around knee pain, so it was too small to be sure about skin. In the measures where collagen did best, the collagen group happened to start lowest, which can make a group look like it improved more than it really did.
The author concluded: “Overall, the CH [collagen] supplement may have shown a superior effect of increasing DT [dermal thickness] in photoprotected skin regions, but more research is warranted” [12].
Overall, from the new 2026 meta-analysis, we have reasonable evidence that collagen peptides may improve skin hydration and elasticity. Skin wrinkles are a different story, and we likely need longer studies to see whether collagen peptides improve wrinkles or not.
And in the three trials I could find that directly compare collagen peptides with another protein, the results lean in favour of collagen, although none of them is conclusive on its own.
Joints, Tendons and Muscle
But skin isn’t the only reason people take collagen, so here’s the short version for the rest of the body. A 2026 umbrella review, which combined 16 earlier meta-analyses covering 113 trials, found a modest reduction in osteoarthritis pain with collagen, and a gain in lean mass [13].
But almost all of those trials compared collagen with a placebo that had no protein. Where trials did match the protein, collagen didn’t come out ahead. In CHAMP, knee pain improved no more with collagen than with milk protein [12]. And whey protein raises muscle-building signals more than collagen does [14].
Where This Leaves Collagen
So here’s where I’ve landed.
The supplement world is hard to navigate, because most of the trials are going to be funded by the companies selling the product. We have to keep that in mind with every result in this article.
But a 2026 meta-analysis found benefits for skin hydration and elasticity [9], and my own correction of the Grimes data points the same way. Then there’s the Korean trial, the burns trial and the CHAMP trial. None of them is conclusive on its own, but all three lean the same way, suggesting there may be benefits from collagen peptides beyond simply getting more protein.
In a perfect world, we’d have a large, independent trial, registered in advance, comparing collagen with the same amount of another protein for a year or more. We don’t. In the supplement world, we’re often feeling our way in the dark, trying to make the best decision we can from noisy data.
Which brings me back to my own conflict of interest. I sell a supplement that contains collagen, so you should weigh everything I’ve said with that in mind.
We’re on the eighth and ninth formula versions of my supplement, and based on the data we have today, I’ve elected not to remove collagen from the next version.
References
1. https://x.com/AviBittMD/status/2106272650377023587
2. https://doi.org/10.1111/jocd.71208
3. https://pubmed.ncbi.nlm.nih.gov/33742704/
4. https://doi.org/10.3390/nu15092080
5. https://doi.org/10.1016/j.amjmed.2025.04.034
6. https://doi.org/10.1111/bju.13010
7. https://doi.org/10.1097/01.ASW.0000655492.40898.55
8. https://doi.org/10.1080/09546634.2020.1725412
9. https://doi.org/10.1111/jocd.71041
10. https://doi.org/10.1089/jmf.2013.3060
11. https://doi.org/10.1016/j.burns.2019.02.015
12. https://mro.massey.ac.nz/items/358fc22e-e691-436a-baa4-196a62d9e5d8
