I do know how Zillow's tool was developed. It is in the public domain. Zillow's tool is built to minimize the percentage error in its estimate. This has triggered a lawsuit by those who feel they have been harmed by the estimate. The tool is an unbiased estimator in logarithmic space and not in the price space. In the price space, the best estimator would minimize quadratic loss.
If you are not mathematically inclined, it implies that errors are multiplicative instead of additive in appraising your home. Property appraisals are additive when done by commercial appraisers. The short version of this is that Zillow is as likely to be off by $10,000 on a $100,000 house as it is off $100,000 on a $1,000,000 house. A commercial appraiser is as likely to be off by $10,000 on a $100,000 house as they are to be off by $10,000 on a $1,000,000 house.
The second issue is that Zillow cannot see the condition of your property. If your home were struck by a meteor, the Zillow price would not change until the neighborhood had reduced prices due to the catastrophe. It cannot know the actual state of affairs.
Zillow does worst with two types of homes. The first are homes built before 1920 as these homes are generally not uniform with the surrounding environment and are not similar to anything else. The second is high-value homes. This is due to the multiplicative scaling of errors combined with a lack of comparability.
This is what has triggered the suit. A commercial appraiser may say the home is worth $4,000,000 and Zillow $3,600,000 or $4,400,000. The difference is that the appraiser is required to know the territory in which the home sits and which structures are actually comparable and so the data is restricted to reasonably linked properties. Additionally, the appraiser physically observes the property and the neighborhood, is aware of local financing issues, and is very experienced in the local market. Zillow cannot approach that level of modeling difference.
People are upset because when Zillow is low, then no one will make an offer comparable to the appraiser's written estimate, but when it is high the sale can still go through if the buyer is willing to add cash out of their own pocket.
There may also be an endogeneity issue present in their regression model, but that is unclear. It is also unclear as to whether it would have an impact.
Depending on what you are paying for the estimate and how important it is for you to know the value, I would say that the Zillow estimate is likely reasonable. It is important to understand that the method used averages over the "sample space," so that it is a great tool on average for what it is trying to do, but not a substitute for a commercial appraisal.
An additional potential source of information is your local property assessor if you live in a property tax state. Most states require assessors to value properties in roughly the same manner that commercial appraiser value properties. If you can reverse engineer the assessor's method, you can get a pretty good idea of your property's value.
per the request in the comments. Information on the suit and its dismissal can be found at:
I can no longer find enough data to link to the specific mechanism Zillow uses. You can find some loose validation information at: