LLM perception drift vs AI visibility
Visibility measures whether a brand appears. LLM perception drift measures how its portrait changes. A brand can improve on the first axis and deteriorate on the second.
| Visibility | Fidelity/stability | Reading |
|---|---|---|
| Low | Low | Entity absent or badly structured |
| Low | High | Correct portrait when present, limited presence |
| High | Low | Frequently visible but misunderstood or unstable |
| High | High | Strong presence and representation in the tested protocol |
Different metrics answer different questions
Mentions and share of voice measure comparative presence. Citations expose visible sources. List order measures position. Sentiment measures tone. Category stability measures framing. Scope preservation measures boundaries. Canon-output gap measures fidelity. Cross-model repetition measures stability.
A dashboard that collapses these into “visibility” can hide drift. More historical mentions may reinforce an outdated position.
Example
After a PR campaign, a company is cited twice as often. Answers describe it through the campaign topic, which is only a secondary activity. Visibility rises while brand hierarchy drifts.
Decision rule
Use visibility to know whether the entity enters the answer. Use perception and fidelity to know what comes out. Presence may require citability and sources; drift may require canon, disambiguation, time or authority.
Success therefore needs two separate verdicts, never one score.