AI share of voice is not a universal metric published by answer engines. It is a measurement model you define over a prompt set, competitor set, engines and sampling cadence.
A defensible definition
Start with a fixed universe of prompts and competitors. Decide whether a mention, citation, top citation or shopping/brand object counts. Document the weighting before looking at the result.
Store enough raw data
Keep engine, prompt, geography, timestamp, answer text, cited URLs and extracted entities. That lets you revise scoring later without recollecting historical data.
Separate channel views
Report each engine separately before blending them. A single aggregate can hide a strong result on one surface and weak visibility on another.