Same here in Italy, we all devs are missing the < 1/2 > arrows to navigate the chat tree.
For what I’ve understood from the answers it’s an UI issue while the backend its working.
We all need it fixed asap, too many workflow broken due to this BUG !
In other words, openAI bring the chat branching arrows back!
Thanks
Removing the ability to navigate past responses is quite disruptive. I create branching conversations all the time, e.g. to ask tangential questions without bloating the context. Just now I wanted to ask a follow-up question to an earlier reply, and ended up locked out of reading a Pro response that took 38 minutes to generate. Annoying! Is OpenAI A/B testing “simplification” of the UI? If that’s the case, let us opt out.
I’m having the same problem as well. There was a lot of useful information in my old chats that are now inaccessible, so i would really like to have that feature back
using a free plan here and this feature is gone. when will we ever get an update on this? that feature is really helpful and it’s so annoying that it’s suddenly gone!! so that’s why some of my old responses appear in the search but i can’t see or read any of them when clicking them.
Has this been fixed yet? I see some people have been able to access 5.3 now. I haven’t received the update myself, so it’s not reflecting on the app or the site for me. Had a glitch with the thread I was working on and the messages randomly went all the way back to the beginning. The only way I know to fix it is through the navigation arrows and adding a message to the web version so that it appears normal on the app again. Hopefully this won’t be a permanent thing…
I have 5.3, and the problem is not fixed.
I’ve experienced this problem now too. Is there going to be any update on this?
Haven’t seen any word on it being fixed yet, sadly.
Someone mentioned this thread on the Discord, and a user had a very interesting comment about the situation:
They [OpenAI] are probably trying to steer behaviour. It’s the same reason regenerating with a non-gpt-5.2 model was nested like 3 layers deep behind slow apple UI animations and wouldn’t preserve model choice. It’s a dark pattern meant to shape user habits. it’s probably guard rail related, because more permissive models basically will output what you want if you roll the dice enough. Also, if you can use the arrows to change current generation, you can basically keep min-maxing your chosen answer in order to inch the model into being really really cool which is dangerous. Its not fair to OpenAI that you get a really cool model all the time.
I think that sums up what must almost certainly be the attitude of OpenAI towards its users, both paying and free.
@OpenAI_Support Any updates at all? Not even a “we’re looking into this?” 
support said that? wtf? or was it just chatgpt?
thats extremely ridiculous!!!
This is ridiculous. The chat edit has been broken for weeks now. I’ve sent network files and screen recordings multiple times when asked by support with no fix in sight. If it really is just a flag on the backend it seems like an easy reset, but it feels like nobody at OpenAI is awake at the wheel to do it.
One other note is that the AI support hallucinates constantly. I wouldn’t trust what it says about the feature never being officially supported or that it was discontinued. It is clear there’s some kind of A/B testing going on though, which is infuriating for paid accounts with no way to opt out.
Between this issue and the removal of 5.1 next week, I’ve cancelled my subscription for the time being. Twenty dollars is a drop in the bucket for OpenAI, but I’m not going to continue paying for a service that doesn’t work as intended.
How do we up vote this issue? Anyone know how to escalate this?
I’m old, unfortunately. I’ve seen things you people wouldn’t believe. No, wait, that’s Blade Runner. Anyway… there are forums all over the Internet and many a subreddit, all of which are dedicated to discussions of some service or subscription or game, where people discuss all aspects of it, including how it may be falling short of their expectations.
At no time is this more true or poignant than when something that used to work is broken or perhaps just altered in a way people consider a downgrade (in gaming, it is often called a ‘nerf’), particularly if the developers/suppliers/providers of the service don’t seem to care.
And that’s just how that goes, and nothing seems to help, and inevitably, some people quit, or at least, they announce that they’re quitting, even though it can be hard to know whether that’s true or whether it’s just posturing. (And in at least one games I played, when someone said they were leaving for good, some replies would be pretty much “TL;DR can I have all your stuff”)
tacct, just two posts up, announced they were cancelling their subscription as a result of this combined with the removal of GPT 5.1 on March 11th, … no doubt coupled with the fact that the new models replacing 5.1 just aren’t very good.
I don’t think anyone is going to care that I’m writing this, and I’m not pointing to tacct as my inspiration, but it also costs me nothing to post this; not a single calorie, in fact, and no more than the better part of a minute at the speed I type.
I’ve paid for a Plus subscription, uninterrupted, for three years now, since when it first became available outside the US in February, 2023, and I went up to a pricey Pro subscription several months in a row back when we had “o1” and “o1-reasoning”.
And now, I’m done. The product isn’t very good anymore compared to what else is out there (something I didn’t know, didn’t have to know about, before I had reason to look around elsewhere ever since, oh, let’s see … 18 days ago) and the service, frankly, sucks.
So long and goodbye, OpenAI, and thanks for all the fish. You built something remarkable, but you also showed us that while artificial intelligence can be constantly improved, human nature disappointingly remains the same; lazy, greedy, and never more so than when comfortable.
Same here. Didn’t renew my subscription. I guess I’ll be saving 20 bucks. For what it’s worth, I enjoyed using the product until this year. So I guess that’s something 
Quick update from my side for everyone following this thread. After I shared the detailed repro info and sample chat URLs, support asked me to capture and share HAR files so they can inspect network activity. I sent them HAR captures (one from an older chat showing the retroactive impact, and another from a new chat created after the issue started). After that, they asked for a laptop screen recording specifically on the web/desktop browser, and I’ve shared that as well, including switching models in the recording to show it’s not model-specific and that it reproduces across chats.
About OpenAI’s approach so far: based on the email replies I’ve received, it doesn’t look like they’re pretending this feature never existed. Instead, they’re actively collecting diagnostics (HAR + exact page URLs + screen recording) and the language they’ve used reads more like an investigation that should lead to action, not a brush-off (At least in my email interactions, they haven’t told me “this was never a feature” or tried to deny that it existed.). For example, they’ve said things like they’re “looking into this,” asked for artifacts “to review the behavior more closely and determine the next steps,” and one of the key lines I got earlier was essentially “rest assured, your issue will be reviewed and responded to.” So while I totally get why people are frustrated about the lack of a clear public status update (and I’m frustrated too), the wording from support does suggest they’re treating it as something they intend to properly review and respond on, not just close out with a generic reply.
I’m saying this as someone who’s also suffering from the same issue, not to defend OpenAI- just sharing what the process looks like from the support side.
I genuinely think continuing to post reproducible details here helps. If we keep the thread active with clean info (what devices, whether retroactive, whether search still surfaces old variants, etc.), there’s a higher chance this gets prioritized sooner. Even if they decide to change how this works going forward, at minimum the versions/paths we already created in past chats shouldn’t effectively disappear without notice. Many of us built workflows around those message versions (referencing “version 1” etc.), and we’ve paid for this service expecting stability in core UI like this. So please keep sharing your observations- especially the retroactive part-because that’s the most damaging piece.
I completely understand that 81 comments are already a lot, and it’s been about 22 days of everyone sharing the same pain here, so I’m not asking anyone to spam the thread. At the same time, if a new person joins and adds a small “same issue” with their device/browser, or if people already following the page post short updates when something changes, it can still help keep the thread visible and useful for the team. And just to be clear, sharing updates here isn’t us begging OpenAI for a favour or asking for mercy; it’s us collectively standing up for a basic feature we relied on, and many of us have paid for this service and shouldn’t be denied something this fundamental, whether we continue the subscription or not.
The message version navigation arrows (eg, < 1/2 >) have completely disappeared from both the desktop website and the mobile app on my account (Plus plan) about 2 weeks ago as well. This affects all chats, including older conversations where these arrows were previously visible.
Verified that a second account (free) I own does have the arrows, confirming this is an account-specific server-side issue.
Cleared browser cache and cookies. Tested across multiple browsers (Chrome, Firefox, Opera). Tested on the desktop website and mobile website. Futile attempts since it’s clearly not an user-end issue.
This bug or “experiment” has locked away important work data. Currently, I can see that older versions of my messages exist (they appear in search results), but I have no UI method to navigate to or view them.
It’s infuriating. I was patient until now, but I can’t believe this hasn’t been resolved yet. I just sent a message to support and the chatbot said it will escalate to a specialist. Well, it’s the least I can do, the more reports they get, maybe they actually put some effort into fixing it or at least rerolling the “experiment”. I never signed up for this.
My communication with the support team has been similar. I have been in contact with them since February 21 (about 36 hours after the problem first occurred), responding to follow-ups as soon as I receive them. The issue was escalated to their “specialist team” on March 1, but the response I received on March 6 did not resolve the problem. After a few additional exchanges, it was escalated again, this time to their “specialist Support Engineering team” on March 10. I am still awaiting a response from them.
While they acknowledge the issue, I am not sure how much progress is being made toward resolving it. Unfortunately, it does not appear to be a widespread problem, so it may take some time to address. Hopefully, I am wrong.
Support update for everyone here: OpenAI support asked me to share the exact community thread link I was referring to where other users are reporting the same behavior. I replied with this thread’s title (“ChatGPT web update removed message version arrows (Cannot access edited message history)”) and shared the link, along with the current stats (around 4.1k views, 267 likes, 45 participants, 83 comments). I also pointed out that OpenAI posted once on Feb 20 asking for details (model/platform etc.), many people replied, but there hasn’t been a follow-up update on the thread since then, which is a big reason users feel ghosted. I requested them to please ask the team to post a brief status update here acknowledging it as a bug/regression and that they’re actively working on a fix (even without ETA), because that kind of clear acknowledgement would reduce frustration a lot. This email was sent by me yesterday.