Why AI referral share is a floor, not your AI visibility

Being cited and being clicked are different events

An assistant that names a business in an answer has done most of the work a search result used to do. If the answer was complete, the reader has no reason to click, and the business appears in no referral log at all. It was visible, it was recommended, and its own analytics record nothing.

So referral share counts the subset of citations that failed to satisfy the reader enough to keep them. It is a real number and it moves for real reasons. It is a floor under AI visibility rather than a measure of it, and AI Visibility Forecasting treats it that way everywhere it appears.

Three ways the floor sits far below the ceiling

The answer was enough. Opening hours, a phone number, a price range, whether a service covers a county: all citations, none of them clicks. Businesses whose value is easy to state suffer this most, which is the opposite of who you would expect.

The referrer was stripped. Some assistants pass no referrer, some pass one that analytics files as direct, and the same visit lands in a different bucket depending on the surface and the month. A fall in measured referral share can be a reporting change with no change in behaviour underneath it.

The mention was not a link. Being named in prose without a citation link produces a searcher who arrives later by typing the brand, and lands in organic or direct having been sent by an assistant nobody can see.

What follows, and it is not despair

The useful consequence is that referral share is worth watching as a direction and is close to worthless as a level. A business comparing its AI referral percentage against a competitor's is comparing two floors of unknown depth. The same business watching its own floor move over eight weeks is looking at something real.

It also means the honest answer to "how visible are we in AI" is often unknown rather than a number, and unknown is published as unknown. A percentage invented to fill that gap would be quoted back for years.

What we hold, and what we do not

Referral share is reported as a floor, labelled as one, and never presented as total AI usage. Where a change in it is more likely to be a measurement artefact than a behaviour change, that is said in the forecast rather than left for the reader to work out.

What we do not do is estimate the gap between the floor and the ceiling. It cannot be measured from outside, an assumed multiplier would be the least defensible number on the site, and a forecast built on one would inherit the error without showing it.

Reading the probability covers how a call is scaled, why a forecast misses covers the record, and both sit under AI Visibility Forecasting.