Coulomb counting is not inherently any more accurate than any other mean, either. It requires good algorithm combined with the user following the expected usage pattern of that algorithm (which sometimes isn't the case), and still cannot properly predict the capacity fade if the battery is never (or seldom) fully discharged (simple voltage based can do that to some degree).
I have seen coulomb counters totally fail and go nuts, but never seen a voltage based estimation more than about 20-30% off.
For simple li-ion gadgets, I often strongly recommend using simple voltage based estimation, unless utmost "best case" accuracy is needed. And if the best case accuracy is needed, then the designer needs to realize that coulomb counting worst case accuracy is, depending on the algorithm, even worse than voltage based, unless a lot of R&D is put in on the system level to improve it through algorithms combining data sources. A nice looking techno demo showing a steadily moving percentage reading like 61.7% looks nice, until you find out that the battery is actually dying out at 10% and all that accuracy was fake; a stateless 5-step approximate bar based on voltage is sometimes better and more reliable, even if it flickers between two bars and one bar when the device consumption fluctuates. At least it gives the user a signal: "if I keep doing this, it will soon die".
NiCd, NiMH, lead acid and some niche sorts of li-ion (typically LFP) require coulomb counting since the voltage curve is too flat.
Cell DC ESR (which is temperature dependant) is the biggest issue of the voltage-based measurements. It can be compensated by measuring current, but then the complexity goes up; it can still be stateless which is a plus compared to a coulomb counter.