A few people reached out after the click radius post
in January. Two of them are also going through the leaked docs independently and had
a different interpretation of the DensityClass enum that I think is probably
right.
In the original post, we said the DensityClass (URBAN, SUBURBAN, RURAL)
likely controls the radius size - a dentist in the suburbs has a bigger radius than
a restaurant in downtown Chicago. That felt intuitive. But looking at it again, we
think that's wrong.
The radius_meters field is a separate float. If the density class controlled
the radius, you wouldn't need a separate field for it - the class itself would imply
the radius. Having both suggests they serve different purposes.
The more likely interpretation: DensityClass controls the weighting curve
between in_radius_weight and out_radius_weight. In an urban
area, clicks from outside the radius might still get decent weight because people in
dense cities search from transit, offices, and apartments that are all close together.
In a rural area, a click from 20 miles away might get almost zero weight because
if someone is that far from the business, the click is probably not a real buying signal.
This is a more nuanced interpretation and honestly more consistent with how Google typically builds these things. They parameterize the tuning, not the binary threshold.
The practical advice is the same
The core observation still holds. Google uses some kind of geographic weighting on click signals for local search. Clicks from nearby matter more than clicks from far away. The density class just affects how steep that dropoff is, not whether it exists.
The practical advice is the same: your physical location relative to where your customers search from probably matters more than we gave it credit for. The correction is in the mechanism, not the conclusion.
Thanks to the people who wrote in. This is exactly why I'm publishing these write-ups - more eyes on the code means fewer wrong interpretations. If you're going through the leaked docs and see something we've gotten wrong, email us.