To be able to compare strings / Words or documents the data needs to be converted to a format the computer understand, vectors.
Google has a nice guide on Universal Sentence Encoder for sentence similarity that you can follow which explains how to generate a vector from the neural network that they have already trained.
What they do and what you can try is ...
They might or might not be similar, the embeddings extracted by mean pooling the BERT output usually have high cosine similarity even though the input sentences are completely different.
Bert embeddings are not meant for sentence similarity task(SST), but there is some research combining Bert and SST.
Here are those resources,
SBERT paper: https://arxiv.org/...