Parametric Memory
Knowledge baked into a model's weights during training, which it can recall to answer without searching the web.
What is parametric memory?
Parametric memory is knowledge encoded in a model's weights — answers given without any web search. SkuLift separates parametric SOV from web-grounded SOV per engine (except Perplexity).
Parametric memory is what a model already knows: the answer it gives before it ever looks anything up.
During training a model absorbs vast text into its parameters. At answer time it can respond purely from that internal memory, with no live retrieval. The reply reflects what the model learned about your brand and category up to its training cut-off, frozen and unverified against the current web.
This matters because parametric answers reveal how deeply a brand is embedded in the model itself. Strong parametric presence means an engine names you even without grounding; weak presence means you only appear when retrieval surfaces your content. The two are very different competitive positions.
That is why SkuLift measures them separately, splitting parametric SOV from web-grounded SOV per engine — except Perplexity, which is always web-grounded. The split shows whether visibility is durable in the model's memory or contingent on fresh retrieval.