It would be very surprising if the AI training corpus includes a lot of project estimation documentation, since most of those are confidential and not publicly available.