AI Visibility Forecasting
Source class weight: what is gaining and what is losing
Not which specific sites are favoured, which is a private decision, but which kinds of source appear to be carrying more or less weight over time.
What a source class is
A category of origin rather than a specific site: independent editorial, reference records, professional and regulatory registers, user-generated platforms, aggregators, and business-owned pages.
Individual sites move in and out of favour for reasons nobody outside can see. Classes move more slowly and for reasons that are frequently visible.
Why classes are forecastable and sites are not
A specific site being favoured is a product decision. A class gaining weight is an emergent property of many decisions, and it tends to move in a direction for a while because the reasoning behind it applies broadly.
So a directional statement about a class is defensible where the same statement about a site would be a guess.
What tends to move which way
Classes gain weight when they are hard to fabricate, carry an accountable author, and have a record of accuracy. They lose weight when they become cheap to produce at scale, which is the mechanism that has repeatedly degraded whole categories of source over the last two decades.
The practical read: anything a business can generate about itself in volume trends downward in weight over time, and anything gated by a third party trends up.
What this implies for effort
It is the main reason independent corroboration is weighted so heavily in evaluation and is the slowest thing to build. The classes that are gaining weight are precisely the ones a business cannot produce for itself, which is inconvenient and is the whole point.
The failure to avoid
Chasing whichever class currently appears strongest. By the time a shift is clearly observable it is well underway, and a strategy built on a lagging read of it arrives late. The durable position is to be present in the classes that are structurally hard to fake, because that is the property doing the work rather than any particular ranking of them today.
Confidence on this kind of statement
Moderate at best, and stated as such. Class-level movement is inferred from citation patterns rather than observed directly, and the inference is the part most likely to be wrong.