What a forecast cannot see
Every forecasting method has a shape, and the shape has edges. Naming them is not modesty. It is the only way a reader can tell which silences mean nothing is happening and which mean nobody is looking.
Absent and unmeasured are different findings
A quiet report can mean two opposite things: conditions are stable, or the instrument does not reach where the change is. Collapsing those is how a monitoring service becomes reassuring rather than useful, and it is the failure this page exists to prevent.
AI Visibility Forecasting reports coverage separately from conditions for that reason. Where a surface is not covered, the report says unknown rather than calm.
Five edges, stated plainly
Inside a private conversation. What an assistant says in one person's session is not observable from outside. The method infers behaviour from patterns across many public probes; it never sees an individual answer to an individual user.
Personalisation and location. The same question answered for two people in different places, with different histories, can return different businesses. A forecast describes the central tendency, not the answer anyone in particular received.
Surfaces with no public probe. Some assistants expose no stable way to ask the same question repeatedly and compare. Those are recorded as uncovered rather than quietly dropped from the index, because a shrinking denominator that nobody announces makes every trend look better.
Why a change happened. A model update, a ranking change, a data source swap and a competitor's new content can produce the same observable movement. The method reports what moved and how confidently; attributing cause is frequently beyond what the evidence supports, and saying so is more useful than a confident guess.
Anything before we started watching. A baseline begins when measurement begins. Claims about how a business appeared last year are not available, and a reconstructed one would be an estimate wearing a measurement's clothes.
The edge that moves
Coverage is not fixed. Surfaces appear, change how they answer, and occasionally close. A method that quietly re-scopes to whatever it can still reach will show a smooth record through a genuine loss of visibility, which is why coverage changes are published as events rather than absorbed.
What we hold, and what we do not
Every report states what was covered and what was not, and an uncovered surface reads as unknown rather than as stable. The source list names limitations rather than only coverage, which is the same rule applied to inputs.
What we do not do is fill an edge with an estimate to keep a chart continuous, infer a private answer from a public probe, or assert a cause the evidence cannot carry. Unknown is a published result here, and it appears as often as it is true.
Why a forecast misses covers what happens when a covered call goes wrong, and both sit under AI Visibility Forecasting.