I’m a professional appraiser and all I can say is that people should stop looking at Zillow as being anything other than a source of amusement. It might be suitable for satisfying your idle curiosity but that’s about the extent of it.
In our OPs case, if they live in an area in the OC where there hasn’t been any REO activity and then all of a sudden a couple REO sales occurred that might explain the big jump. Zillow can’t tell the difference between typical vs. atypical sale conditions and would weight those recent sales just because they’re recent.
In general, an Artificial Valuation Model (AVM) will work best in an area that is comprised of very similar subdivision homes that don’t have any dramatic site influences like views or traffic or such, and in where there is a lot of sales activity. It’s all about quality and quantity of data, the primary element of the “quality” aspect being how similar the data are to each other.
An AVM would work okay in a big conforming tract; it won’t work well in an area that doesn’t have that sort of homogeneity or which has a relatively low number of sales.