That's a neat idea. I worked for a client who wanted to verify writing was human-generated. There was a way around it, but overall it worked pretty well.
The only real flaw I see is that it could be mimicked very easily. What may be better would be to have the hash be uploaded to a server or use some cloud provider's free tier and serverless functions (within free tier) to verify the transaction hash is correct (e.g., right email sender, time). I wouldn't store the body or title of the email.