I’d love to see an optional conversation mode that makes ChatGPT behave more like an actual messaging app rather than a strict request → response interface.
Currently, the interaction is essentially:
User → Assistant → User → Assistant → User → Assistant
I think it could be much more natural if both sides were able to communicate asynchronously, in smaller conversational turns.
1. Allow users to send multiple messages before ChatGPT responds
For example:
User:
I just thought of something
User:
What if we did it this way?
User:
And actually, there’s another thing
User:
WAIT, I think I just figured out the important part
Instead of forcing the user to wait for the assistant’s response after every message, ChatGPT could collect these consecutive messages as part of the same conversational turn.
This would be especially useful when someone is thinking out loud, brainstorming, troubleshooting, or simply typing naturally in several smaller messages.
2. Allow ChatGPT to respond in multiple smaller messages
The other half of the idea would be allowing the assistant to occasionally split a response into several messages instead of always producing one large block of text.
For example:
Assistant:
Yeah, I think there’s something interesting here.
Assistant:
The main advantage would be…
Assistant:
Actually, wait — there’s an even better way to look at it.
This could make the conversation feel much more natural.
More importantly, the user should be able to interrupt the assistant while it is responding, just like they can interrupt another person in a real conversation.
3. Make interruption a first-class part of the conversation
Voice Mode already demonstrates how valuable natural interruption and turn-taking can be.
I think the same concept could be brought to text.
For example:
Assistant:
The main reason this would work is that—
User:
WAIT WAIT, I DON’T MEAN THAT
Assistant:
Got it.
Or:
Assistant:
There’s another important consideration—
User:
Hold on, let me explain first.
Assistant:
Sure, go ahead.
This isn’t necessarily about making the AI “more human.” It is about making the interaction model more conversational.
4. A “hold / lock” state for users who are still composing their thought
I think there could also be a simple mechanism for temporarily telling ChatGPT:
“I’m still talking — don’t respond yet.”
For example:
User:
Okay, so
User:
I was thinking about this
User:
and I want to build ______
User:
I already tried a couple of things
User:
BUT—
At this point, ChatGPT could either wait because the user has explicitly enabled a “hold” state, or simply recognize that these messages are likely part of one unfinished thought.
Something as simple as a “Hold response” / “Continue sending” control could make this work extremely well.
5. Let ChatGPT recognize when it has been interrupted too early
I think the interaction could work in both directions.
Imagine the assistant is explaining something important:
Assistant:
The reason this happens is actually related to—
User:
Okay but what about X?
If ChatGPT knows that it hasn’t finished its current thought, it could respond:
Wait
I wasn’t finished — there’s one important part about X that I was about to explain.
And then continue.
Voice conversations already have situations where the assistant can effectively signal that it was interrupted before finishing its thought. Bringing a similar concept to text could make the interaction feel much less rigid.
6. This could potentially be more efficient, not just more expensive
One concern I can imagine with this feature is that generating multiple messages could potentially increase token usage.
However, I think the opposite could sometimes happen.
If the user can interrupt the assistant early, there is no reason for the model to continue generating another few hundred or thousand tokens that the user no longer needs.
For example:
Assistant:
Here are the five reasons why—
User:
STOP, I GET IT
Instead of finishing the entire response, the generation could stop immediately.
So depending on how the system is implemented, asynchronous conversation and interruption could potentially reduce unnecessary generation in some situations rather than simply increasing costs.
7. Make it an optional mode
I don’t think this should replace the current interaction model.
Instead, it could be an optional mode:
Efficient / Standard
- One user message → one complete assistant respon
- Optimized for information-dense answe
- Minimal overhead
Conversational / Realistic Chat
- Multiple consecutive user messageMore granular assistant response
- Natural interruption
- More conversational pacing
- Optional typing indicators / short acknowledgements
- Ability to temporarily hold responses while the user is still composing a thought
This would let users choose between maximum efficiency and a more immersive conversational experience.
Why I think this would be interesting
The important part isn’t simply:
“Make ChatGPT look like whatsapp/telegram.”
It’s about changing the rhythm of the conversation.
Real conversations aren’t always:
“Person A sends one perfectly formulated message → Person B writes one complete response → repeat.”
People send thoughts in pieces, they correct themselves, they interrupt each other, they react to individual messages!
They sometimes send five messages before the other person has even had a chance to respond
ChatGPT already feels remarkably conversational in terms of language. Giving the interface a more asynchronous conversational model could make that feeling significantly stronger!
|
And with message reactions now being introduced, this kind of interaction model could fit particularly well with the direction the interface is moving toward.
P.S. And while I’m here — the new message reactions are a fantastic addition!
It’s a small feature, but it makes the conversation feel noticeably more natural. Nice one, OpenAI!