This is almost certainly not due to the accuracy of the sensor, but due to the WAY people use, and wear, their watches. Some like watches loose, some don't mind their straps snug. Some wash the backs of their smart watches regularly, some don't. Sensors blocked by schmutz, or sensors that are continually being jostled and lifted away from contact with the underlying skin, will have compounded rates of errors of all kinds.
I use a Samsung Galaxy watch since my phone preference is Android. I have two versions, one a nicer/dressier/newer Galaxy 6 that didn't need any updating, and an older Galaxy 4 that did need a firmware update before it could measure ECG, blood pressure, and blood oxygenation saturation. I wear the older 4 most days. It detected a recurrence of AF when the missus and I were visiting our youngest and her two young toddlers. I felt the AF return (this was post ablation), and made the app record my current heart rhythm. I uploaded that file to my PC and attached it to an email to my electrophysiologist. He accepted that record without question, and it did clearly show the erratic R-R intervals, and my watch did show no P-waves, both of which are the definitive indications of AF.
I haven't looked for formal assessments of my Galaxy's accuracy, but it has to be much better than 34%, and I'm sure the vaunted Apple watches are much better than that as well. It's the wide variance in their wearer's use of them that should account for this low assessment This is strictly my opinion, of course, and it is without the benefit of seeing the report and what methods were used to control for confounds like wearer variations in the use of their watches.