This is from an ai answer so ymmv ... What I heard before is they know linked addresses so they can cluster owners. But here's the ai

Entity-Adjusted Filtering

Glassnode uses proprietary clustering heuristics to consolidate activity across addresses controlled by the same entity, such as exchanges or ETFs6. This process helps filter out internal transfers and provides a clearer picture of actual economic activity occurring on-chain.

Identifying Non-Economic Transactions

Relay Address Adjustment

Glassnode identifies and removes "relay addresses" from their volume calculations1. These are addresses that simply forward funds to another address, often resulting in double-counting of volume. They define relay addresses as those with:

A mean spent output lifespan of less than 1 hour

A current balance of zero (excluding UTXOs created within the last hour)

Change Volume Filtering

The company recognizes the issue of change volume in UTXO-based blockchains like Bitcoin3. They have implemented change-adjusted volume metrics to avoid overestimating true market behavior. This involves:

Identifying UTXOs controlled by the same entity

Calculating "virtual quantities" for the complete set of UTXOs belonging to one entity

Analyzing balance changes per account on the block level

Conservative Approach

Glassnode opts for a conservative approach in their data cleaning process4:

They aim to limit the reporting of false positives

Individual data points, such as large inflows or outflows, are considered preliminary until verified

Data points become increasingly reliable over time as exchange wallets transact and interact, allowing their heuristics and clustering algorithms to enhance labeling accuracy