How to improve the accurate for gpt-4-vision in detail message?

Hello everyone,
I’ve been using the gpt 4-vision for some WhatsApp chat message.I required a screenshot of the chat to gpt4-vision assistant and want get the key chat infomation in the screenshot。

Basically, logical information and chat message can be accurately identified.
But detail infomation especially for WhatsApp’s checkmark (been sent, received, and read) is very inaccurate.It’s only 20% correct in my test (over hundreds of screenshot )

The message checkmark is very important for me and I want to improve the accurate to 80% at least.
Does anyone have any suggestions or idea?