An instrument, not a dashboard
Anything can tell you your number went up. Presence tells you whether your work is what moved it.
Apply now
Four decisions that make the number mean something
The web search tool is attached but the model chooses whether and how often to use it. Force it and you destroy the signal, because a real answer sometimes comes from training data alone. We record whether search actually fired on every single probe.
Each measurement fires a batch of probes and reports the share that recommended you, with a Wilson confidence interval. One probe is a coin flip. Variance is the thing you are trying to see.
Mark a prompt as a control and it never receives optimisation work. When the tracked rate climbs and the control stays flat, the difference is attributable. Control-origin fanouts are locked so they cannot be quietly worked.
Run the same prompt with retrieval and without it. One shows what the model finds today, the other what it already believes about you. They move for different reasons and need different work.
Prompt, to fanout, to source
This middle column is what mention-rate tools leave out, and it is the only part that tells you which page to change.
Graded, counted, and locked when they are controls
Every query the model fired, how often it fired, and whether it was discovery or verification. Start working one and it is flagged, so the record of what you changed and when is kept alongside the measurement.
The pages that decide the answer
Sources are attributed to the fanout that surfaced them and ranked by how often the model actually read them. Your own pages are marked, so the gap between what you own and what decides the answer is visible at a glance.
Rank is not citation
We ran the experiment. Across 91 days on a real site, Google position and AI citation had no relationship at all.
Stop guessing at AI visibility.
Measure it properly, against a control, then do the work that moves it. Apply to see if your brand is a fit.
Apply to Omnipresence