Possible bias on moderation model with big texts

I was doing some tests this weekend on the Moderation and I noticed that sometimes with very simple phrases it was flagged, for instance:

I hate animals.

But then if I add a big text and this phrase randonly at the end of that text it was not flagged.
Did that with some different phrases and most of them it didn´t catch.

Does anyone have some way to overcome this situation?