On one specific alternate account, I cannot go back to edit or even reroll old messages. This is back when it was using 4.0 for messages, but on other accounts where I had chats with 4.0 or older models I can still edit. It’s a very very weird bug. I don’t even know what to look up for help.
It has been over a week, and I still have not heard back from the “specialist Support Engineering team”. I sent a follow-up a few hours ago asking if any progress has been made on resolving the issue.
Following up on this, I did just get a message back from support and the engineering team. Not the AI support, the actual customer service.
“We confirmed that the version selector or arrow controls previously used to move between regenerated or edited responses are not available in the current ChatGPT web experience for affected accounts. This is expected behavior at this time.
We understand this may change a workflow you previously relied on. Support cannot restore the previous 1/2 or arrow-based version history UI on an individual account, and we do not have a timeline to share for any future changes.”
In what world is this acceptable for any kind of service? I’m done. Absolutely infuriated. If this is a legitimate UI update they’re testing, it’s the worst one yet.
Hi everyone, I’m starting a fresh thread because the existing one (“ChatGPT web update removed message version arrows (Cannot access edited message history)”) has a lot of reports, but we still don’t have a clear product-level answer.
Support also has now sent a flat confirmation(to me as well on personal email) that changes the framing from “bug” to “expected behavior.” Quoting their email:
“We’ve confirmed that the version selector or arrow controls previously used to move between regenerated or edited responses are not available in the current ChatGPT web experience for affected accounts. This is expected behavior at this time. Support cannot restore the previous 1/2 or arrow-based version history UI on an individual account, and we do not have a timeline to share for any future changes.”
This raises serious questions that need an OpenAI response:
-
Why is this feature still available for some accounts but not others? If this is a feature flag/cohort rollout/A-B test, what are the criteria for who is “affected”?
-
Is this a permanent deprecation of version history navigation, or a temporary rollout state?
-
Retroactive impact: older chats that previously had 1/2 arrows/edit history navigation are now missing it, breaking workflows where people referenced “version 1” etc. How do users recover/access those past variants?
-
Export only includes the currently visible version, so “copy before regenerate” isn’t a fix for retroactive loss. Will OpenAI provide an export/recovery option that includes stored variants, or an alternative UI?
I’m not posting this to spam or attack anyone. I’m posting it because thousands of users have relied on this workflow, and removing it unevenly (some users still have it) without a clear statement or recovery plan is not acceptable.
If you reply, please include:
-
Do you still see 1/2 arrows? (yes/no)
-
Is it retroactive for you (old chats too)?
-
Web/desktop/mobile + browser
-
Does search surface old hidden text that you can’t navigate to?
OpenAI team: please answer directly here whether this is intentional or a regression, why access is unequal across accounts, and what the remediation plan is.
I’ll add the link to the older thread here for reference: “”
And I’ll post the link to this new thread once created: “”
Above is the link to the new thread I posted: “Support Can’t Restore 1/2 Version Arrows — Why Do Some Accounts Still Have the Version Selector (Paid Users Affected)? Paid Users Losing the Message response Selector”. If you’re facing the same issue, please drop your experience and device/browser details there too, so it stays visible and OpenAI is pushed to give a clear answer and response. We’re not wasting money here, this isn’t free, and the work we built around version history isn’t useless—so the retroactive removal and unequal access needs a proper explanation and resolution.
I experienced this problem specifically after they started phasing out their legacy models, and I can absolutely confirm that I no longer see the arrows or the history of my prompts since they removed 5.1.
I work on several writing projects and lore bibles, and this change HAS DESTROYED MONTHS OF WORK for me. I can’t maintain the same workflow as before—it forces me either to branch the chats (which is completely inconvenient, as it would flood my projects with hundreds of separate chats) or to revert to the old method of copying different versions of my prompts, answers, and model changes into a document file.
I was already considering canceling my subscription after they removed the legacy models (a decision I frankly found absurd), but now I see very few benefits to continuing my subscription.
Well, I have gotten the same response to my ticket as well. Sent them the screencaps, then the HAR file, and they finally replied this today:
“We understand you are reporting missing version arrows for regenerated or edited messages on your Plus account, while the feature is still available on free accounts.
We can confirm this is expected behavior, and our team is actively rolling back this feature. We apologize for any inconvenience this may have caused.
If you have any other questions, please feel free to reach out. We are here to help.”
I don’t know how many of you are also aware of the silent “update” that rolled out yesterday/today (depending on your timezone, March 19) that has the web version of ChatGPT (android app is unaffected as of now) to be able to only edit your most recent message in the conversation, and you can only regenerate the last assistant response. All the messages before the last can’t be edited or regenerated, pretty much how Gemini works.
I think this is connected to why they removed the navigation arrows. If you can’t regenerate/edit older messages, then there’s no risk of losing your work and needing the arrows, so it feels like a safety net for their arrow removal.
Anyway, absurd reasoning on their part, it’s a downgrade of functionality without an apparent motive, except the fact that maybe they don’t want people to waste their compute power on regeneration?
So Paying customers are being used as guinea pigs for a downgraded experience, nice, I always liked to pay more to get less.
If it’s about usage, just lock it behind paid plans, say “Plus plan gets access to navigation arrows” or “Pro plan gets access to 1000 pointless regenerations”.
Nevertheless, OpenAI, please publish change logs or deprecation notices before removing core workflow features. Even a banner a week in advance would let us adjust. Or better yet, give us time to cancel our subscription and move to a company that respects its users.
This is atrocious.
Even Grok, or Deepseek have those arrows, and this AI can’t even keep them? Absurd.
Yeah, I can’t edit older messages anymore. And It’s even more detrimental to my work and workflow.
Supposedly (but not officially declared) the new edit restrictions are because OpenAI summarizes / truncates long conversations on the backend, and editing previous messages requires it to rollback the summary.
But this makes ChatGPT as an app effectively useless. The ability to edit and pick between responses is a basic fundamental part of AI interaction.
What happens if I need to update code requirements mid-conversation, or want to choose from several creative writing responses? It’s all broken now. I won’t consider resubscribing until they offer full edit capabilities again. Why would I keep paying for this?
I have opened a thread requesting this change to be reverted – Restore the Ability to Edit All Messages . I will submit a complaint to the support team in a few hours, including links to various resources showing that people are unhappy with the change. Hopefully, this will encourage them to reconsider.
Exactly. As a writer, I use it to help me to organize my ideas and lore bible, but without this feature, it will be useless…what’s the difference between me doing it the traditional way on a document anyway? I’m pretty bummed.
Thanks for your initiative, I left my humble view and like to boost it (hopefully).
I also have a Plus account and have been affected by this issue of missing version arrows. its very frustrating. i was willing to put up with it, but with the most recent update i now can no longer go back and edit older messages (yet remain able to edit most recent message at least) at all, which is highly disruptive to my workflow where i am expected to use this for my job. very confused at why this happened and why they cannot partially restore a previous build while working out the error. very confusing! very unfortunate! my free account does not contain these bugs, as experienced by other users as well. hope this gets resolved soon.
(idly wondering if this is something like… integer overflow? or… heavy use users only? is everyone on here a very frequent chatter? I have millions of words in a year, most of which contained behind version history of edits of chats, so perhaps this density of use caused the problem… but that is wild speculating).
Yeah honestly, if that theory is even remotely true (that heavy usage / lots of versions caused this), then that actually makes the situation worse, not better.
Like okay, removing a UI feature… fine, companies do experiments, rollouts, whatever - that part we can understand. But if the reason is that we have more data, more edits, more branches… then you’re basically saying the data exists on your servers, but we’re being restricted from accessing it. That’s the real problem.
Even now, export doesn’t solve anything - it just gives a flat version of the chat. All those branches, edits, regenerated paths we actually worked on are just… not there. If they can’t restore the UI immediately, at least give us some way to access our own data - like a proper JSON/HTML export that includes version history. Honestly, if that existed, many of us wouldn’t even push this hard for the UI to come back.
Because at the end of the day, this isn’t just “usage data” - this is our actual work. The more we’ve used it, the more input we’ve given, the more intellectual effort is in those branches. And saying “we can’t fix it” while still holding that data on the backend just doesn’t sit right.
Openai please atleast show some courtesy and professionalism, it’s already more than a month and we cannot access our own data, which is breaking our workflow alot, above all no medium or time provided us to store these branches on our side… and the biggest sour is it is being discriminatory, available for some while not for others, is even more frustrating. It is not as as is it is retroactive. We are just asking you for work around, a pdf or a downloadable and readable file, that all we are asking for, it already more than a month please at minimum give some way to us so that we can access our data.
This is insane. So they’re just deprecating such a core feature without ever telling us?
It’s going to be really funny if in that new movie/documentary about ai companies like this the CEO of GPT talks about how they’re super communicative with their customers and consults with us or at least let’s us know ahead of time before they change big things or even things that seem mundane but are big to at least someone. I can’t remember the name of the documentary and won’t be watching the full one but certainly will look up the GPT CEO parts for a good laugh. One of the only reasons I don’t use Claude instead is because I like the voices for GPT and that you can have the voice read it without going through all the bells and whistles. It is kind of charming though with Gemini. One of my favorite voice options on Gemini sounds JUST like one of my favorite characters from a game called Fire Emblem Three Houses/Hopes.
This is actually insane! Haha wow, what a way to get rid of actual HELPFUL features!!! I’m astonished, and i’ll admit, slightly amused. Wow.
to everyone the response navigator feature is now restored.
Can confirm. Navigation arrows are back and all previous history is intect. (Models 5.3 and 5.4 are still horrible, but at least we can once more paginate between each iteration of their generated horribleness.)