I started a new chat session and sent the first message.
After seeing GPT’s initial reply, I made a slight edit to that first message.
Then, GPT responded with a phrase like:
“Alright, let me summarize it again for you.”
This implies GPT was reacting as if it had seen a previous message or a prior version of my input — even though, from the user’s perspective, there was only one message (the edited one) in this brand-new chat.
Why this is a serious issue:
- Screenshot 1 shows the edited message, which was the first and only user input in the chat.
You can confirm it’s the first message by using a simple behavior in the ChatGPT mobile app:
when you scroll to the very top of the conversation and pull further, the text stretches visually, instead of revealing earlier messages.
This behavior is visible in the screenshot, proving that there were no messages above the one shown.
- Screenshot 2 shows GPT’s reply, which starts with:
“Alright, let me summarize it again for you.”
This language suggests GPT thought it had already responded to a similar message before —
strongly implying it was still referencing the original, pre-edited version of my message.
- The two screenshots clearly belong to the same chat, which you can verify by comparing the small visible text fragments outside the blue-masked areas — they match between both images.
Core issue:
Edited messages should fully replace the original input in GPT’s processing pipeline.
But in this case, GPT seems to have retained the content of the pre-edited message,
which resulted in:
a response based on invalid or outdated input,
a potential violation of user intent, and
a risk to context reliability and user trust.
Please review the attached screenshots and investigate whether this is a case of prompt caching, state retention, or improper handling of user edits.
