That's a library prep thing.
Ideally, you want a small number of cell barcodes having a ton of reads, and if a bunch more cell barcodes have just a few reads each, you figure that's ambient cruft that didn't come from cells, and you ignore it.
Low % of reads in cells means that lots of your reads are from barcodes that don't have a whole lot of reads, which means it's all ambient cruft, and you shouldn't be counting it.
I suppose if you messed around with the white list of barcodes,or used a too-low "force-cells" argument, you could make the software draw that conclusion, but if you ran everything on default, and you got that message, it's not you, it's the library prep.
However, Cellranger 3.1.0 has a smarter algorithm for deciding what is or isn't a good cell, so you might get a better result if you upgraded.