If there is workload on the replication source, the checksum will see a lot of differences due to the asynchronous nature of replication.

pt-table-checksum uses replication in STATEMENT base but I prefer to not use that binlog row format.

However to lower the differences, maybe you could try to use WAIT_FOR_EXECUTED_GTID_SET() (see https://dev.mysql.com/doc/refman/8.0/en/gtid-functions.html#function_wait-for-executed-gtid-set)

This will allow the checksum on the replica to wait for some missing transactions to be applied.

Of course GTID must be enabled.