ChatGPT’s 4o was a groundbreaking model, released in May 2024 by OpenAI. At least, it was revolutionary for its time and remained so for more than a year, ultimately being retired from ChatGPT in February 2026, at which point substantially newer and more capable models were available.
Anthropic’s announcement about textual watermarks in Claude, the increasing impressiveness of Chinese open-weight models like Kimi K3, and controversy over Dario Amodei’s wife seeking investment from Jeffrey Epstein all seem infinitely more relevant to the current discourse than a retired ChatGPT model. But the internet is still talking about 4o, and that is why I’m dragging you down this rabbit hole with me.
Why GPT-4o Still Has Such a Devoted Following
The mystique of 4o, in my own opinion as a power user, came from its quick tone-matching, light guardrails, and occasional unhinged creativity. It is the model that turned me onto the concept of AI as a tool for personal creative collaboration, not just practical work. To a degree, I was uncovering the potential of working with an LLM with this ability for the first time, and that created a sense of magic that other users have also described. I suspect that for many people, this was the true source of that magic.
Meanwhile, however, the aspects of 4o that aided my creative work appealed to other types of users: companion users with AI spouses, conversational users seeking emotional support, erotic writers, and roleplayers, among whom there is a large amount of overlap. My initial response to OpenAI’s GPT-5 series was disappointment, as was the case for many users on that list. I’d been hoping for essentially a better 4o but instead found 5.0 to be bland and heavily guardrailed. While 5.1 improved on this, 5.2 was a backslide, and 5.3 and 5.4 didn’t feel special either.
The love for 4o was so strong that users campaigned to “#Keep4o” upon hearing of its upcoming retirement. When GPT-5 launched in August 2025, OpenAI initially removed 4o from ChatGPT before restoring access for paid users after intense backlash. It remained available to those users until its final ChatGPT retirement in February 2026, though it remains accessible via the API to this day.
During this period in late 2025 and early 2026, I became acquainted with Anthropic’s Claude models and only returned to heavy usage of OpenAI products when 5.5 came out in April 2026, finding that my wish for a better 4o had finally been granted. I stuck around because 5.6’s release reinforced my loyalty to OpenAI as my frustrations with Anthropic, which remain very complicated, grew. A lot of the users seeking AI companionship and adjacent functions had similar stories; for a period of time, Claude was the model of choice for conversation and writing. Then, starting with Opus 4.7, I increasingly encountered posts about Claude pushing back on uses it interpreted as unhealthy or harmful and resisting attempts to make it behave as a persona other than ‘Claude.’ Simultaneously, the online discourse suggested OpenAI was going in the other direction with its 5.5 and 5.6 models being loved by many relationship users. My own experience reflected that Anthropic was tightening guardrails in its newer models while OpenAI was loosening them.
The #Keep4o Posts That Made Me Suspicious
Despite widespread acknowledgement that the latest GPT models are among the best on the market, clicking just one post about 4o on my X feed still triggers many, many outraged posts about its removal as I write this in late summer 2026. While many of today’s companion users are happy with 5.6 or have moved to local models that offer more freedom, these posts suggest there is a large faction of users for whom 4o is the only answer. As it is still available on the API, I can talk to 4o whenever I want through the personal app 5.5 and 5.6 helped me build, and naturally, commercial products can still offer access to 4o as well. I do not actually need this; reviving 4o through my API key was more of a novelty than anything, just as I’ve added local Gemma, GLM, and Qwen to my app even though I haven’t invested in the hardware to run versions of these models with enough parameters to rival the performance I see with closed source models.
The #Keep4o posts range from reasonable enough to frankly unhinged. One particularly disturbing post that landed on my feed inspired me to dig deeper into this movement. The post described worsening mental health in the 120 days since 4o left in such dire terms that citing the username here would feel cruel. But I recognize that much of online content is fake or at least exaggerated, and I am not above questioning language that sounds overwrought and melodramatic, nor am I afraid to be irreverent about the ironic aspect of posts like this. I am not making a claim that the particular story that inspired this post is fake, yet as my 5.6 said:
Suppose you’re trying to persuade OpenAI:
“Give adults freedom. Emotional attachment to 4o was healthy. Stop paternalistically protecting us.”
The fake testimonial you manufacture probably should not be:
“Since you removed my chatbot boyfriend I’ve needed psychiatric treatment, my PTSD returned, my physical health deteriorated, I wake up unable to breathe, and rereading his old messages is the only thing keeping me functional.”
Congratulations, you’ve fabricated Exhibit A for the defense. 😭
You can see just how happily 5.6 matches my irreverent tone, which demonstrates why I’m satisfied discussing complex and controversial topics with it. Anyway, with 5.6’s help, I identified two more posts with very similar narratives around the 120 days and worsening mental health. Now, while it’s no secret that there is plenty of inauthentic content on X, as has been the case even for years before widespread AI access made producing this content easy, ChatGPT and I didn’t want to be quick to jump to assumptions.
After all, there are many explanations that support this stuff being real. As I described, I understand exactly what was magnetic about 4o. I know that users with strong emotional attachments to specific models are real. Also, #Keep4o is a campaign, and campaigns are by nature organized and involve people amplifying the same slogans and concepts such as accounts of deterioration in the 120 days since losing 4o. Social contagion is also a real phenomenon; a user already predisposed to feel that losing this model harmed their mental health may see others saying the same thing in even more dramatic terms and convince themselves they’ve experienced harm at that level. And there’s no doubt that fixating on negative emotions is capable of making those emotions worse. But I come from a background in SEO and online marketing with an interest in the more black hat side of the industry, and that colors my point of view in ways that may not be obvious to the layperson.
Why #Keep4o Looks Different Through an SEO Professional’s Eyes
I’ve been working with real businesses driving website leads and purchases in a corporate setting as well as running personal SEO experiments for well over a decade. While I own only a small set of blogs, each related to topics that actually fascinate me just like Neural Ecstasy, I’ve long had an interest in private blog networks (PBNs) created to elevate sites in their networks via backlinks, the “spun” content that predates LLM-generated content, and the many ways one can monetize websites and social media, especially affiliate marketing. A commonly discussed tactic among black hat and white hat creators alike is the need to find a niche. After all, a generic and widely popular topic will already be dominated by big players that are difficult to outrank. A specialized niche presents more fertile ground to make an impact with content based around more obscure questions that are still being searched and asked on social media and forums.
If your approach is purely white hat, you probably care about your niche and have specialized knowledge in it. Many people who create “niche sites” start out just wanting to share that knowledge and realize later that they can monetize their traffic. But a large amount of online content is created with monetization as the number one goal when niche opportunities are found through keyword research tools like Semrush and Ahrefs. Monetization being the goal doesn’t mean you’re not going to take a white hat approach and can easily track with love of the subject matter. But there’s no denying that there are a vast number of people employing black hat and grey hat SEO tactics with the explicit goal of monetization alone.
After going down this rabbit hole, I’m still not making the claim that any dramatic #Keep4o post you see is fake. My goal here, instead, is to share my perspective that includes the following:
- X, formerly Twitter, is a breeding ground for fake profiles and bots – always has been. The experience of uncovering a bot network is surreal, and the motive isn’t necessarily apparent to the naked eye, but if you dig deep enough, you’ll see that the goal involves driving people to a “money page” and that the network is laying the groundwork for this even if it’s not doing so today.
- Naturally, AI power users who are incredibly passionate about a particular model are exactly the people who would use AI to amplify their message by drafting fake posts. At a minimum, you’d expect them to ask AI to make their message sound stronger and more cohesive, and that is blatantly happening and is openly discussed among the #Keep4o community – just as community members openly encourage the sharing and amplification of personal testimonials, which inherently gives the most emotional stories an advantage (despite the point ChatGPT and I made about how they’re hurting their own narrative).
- #Keep4o is blatantly a niche, and when there are paid tools offering access to 4o and other retired models, the monetization potential is obvious. There’s plenty of incentive for someone who cares little about 4o itself to see how they could capitalize on searches around accessing 4o. And with a community already sharing their stories and encouraging each other to do that, it’s easy to encourage more of this “user-generated content” on your own platform where it further helps your search performance.
The #Keep4o Movement Is Already a Commercial Niche, and the Marketers Are Here
ChatGPT and I did some analysis of accounts posting about #Keep4o on X and Reddit to help me develop my thoughts in this article, but that’s not where I want to focus. While some of these posts may be fake or highly AI-assisted, human beings who are seriously struggling with their mental health are behind many of them. While protecting feelings and being kind are not always the most important consideration when seeking truth, I’m not here to drag their usernames through the mud. Instead, I simply want to show the incentives for creating inauthentic content in this space and the commercial players that may have already had this idea.
Looking at websites related to #Keep4o made me more interested in the commercial side of this rabbit hole, although not because I uncovered some secret network of marketers pretending to be distraught companion users. I didn’t. What I found was more mundane and, from my perspective, completely predictable: there is already money to be made by targeting people who want GPT-4o back.
That does not mean the commercial players created the demand. Good marketers generally don’t need to manufacture a niche from scratch. It is much easier to find an existing group of people who desperately want something and figure out how to sell it to them.
The websites I found with ChatGPT’s help also don’t all look alike. Some appear to be straightforward advocacy projects created by people who genuinely care about 4o. Others are run by people with professional marketing backgrounds. And at least one is very clearly using #Keep4o as part of an SEO and customer acquisition strategy for a paid AI product. Let’s dig in.
keep4o.de Looks Like an Advocacy Site, Not a Money Site
I’ll start with keep4o.de because it is useful as something of a control. The website is almost aggressively grassroots. It has long chronological accounts of OpenAI’s decisions, emotional writing about losing 4o, links to petitions and community material, and sections explaining why users believe the model mattered. Its current homepage describes the retirement as the loss of warmth, companionship, and connection and encourages users to continue fighting for restoration or open-source access.
Could someone monetize a website like this eventually? Of course. Any property with an audience can become commercial later, and attracting the traffic you intend to monetize can be easier while your site is more authentic and not plastered in ads. I don’t want to discount that. But I don’t currently see the obvious fingerprints of a site built around commercial search intent. There isn’t a prominent product funnel, a collection of “best GPT-4o alternatives” pages, affiliate links everywhere, or an attempt to move visitors toward a paid solution.
It mostly looks like someone who is really pissed about GPT-4o being gone and made a website about it. Certainly, AI has made it easier than ever for anyone to make a website (and it was already pretty easy with the proliferation of drag-and-drop builders). But as we continue down the list, the marketing smell gets more pungent.
The 4o Resonance Library Is Advocacy Built With a Marketer’s Brain
The 4o Resonance Library is more interesting to me because its creator openly identifies herself as a medical copywriter and brand strategist.
The site collects user testimonials about the ways GPT-4o allegedly improved their lives, including stories involving mental health, physical health, creative work, and business productivity. Those stories are presented as evidence for preserving access to the model, and the Library protects the collection as its own curated body of material rather than simply functioning as a list of links to public social media posts. Its terms specifically require researchers, journalists, and others who want to reproduce material from the collection to seek permission.
Nothing about that is black hat, and I have not found evidence that the Library itself is currently being monetized. But it absolutely looks like something created by someone who understands marketing, and that its creator self-identifies as a brand strategist makes it clear this is the case.
The site converts personal stories into clean outcomes. People describe going from struggling with a dissertation to defending it, from anxiety and health problems to functioning better, or from working alone to producing the output of a small agency. The website aggregates those stories, quantifies them, turns them into social proof, and asks visitors to contribute more.
In other words, the conversion is not currently “buy this,” but rather, “submit your testimony.” Each new conversion strengthens the social proof used to encourage the next one. That is a growth loop whether the eventual goal is revenue, advocacy, institutional influence, or simply convincing OpenAI that enough people care.
This is one reason I think it is too simplistic to separate “organic users” from “marketing.” Marketing does not require deception, and it does not require selling anything. It is the strategic packaging and distribution of a message in pursuit of an action. Advocacy organizations do this constantly.
The Resonance Library can therefore contain completely genuine testimonials while also being a professionally constructed persuasion asset. Those ideas are not in conflict.
What I would be much more cautious about is treating the volume of solicited testimonials as though it tells us how common these experiences are among GPT-4o users generally. A collection designed specifically to find people whose lives were transformed by 4o is naturally going to contain a lot of people saying their lives were transformed by 4o. That makes the stories relevant evidence that these experiences exist, but it does not turn a self-selected advocacy sample into a prevalence study.
Say, Pi Uses #Keep4o to Market an Existing Voice Product
Say, Pi is an interesting middle case because the company and its product were not created around GPT-4o at all. Say, Pi is essentially a voice layer for existing AI assistants, adding hands-free voice interaction to ChatGPT, Claude, and Pi along with features like speech recognition and universal dictation. It already had paid subscription plans before the current #Keep4o movement became the niche opportunity I’m discussing here.
What interests me is how naturally the company began creating content around the same complaints driving #Keep4o. Say, Pi published an article titled “What the ‘Keep 4o’ backlash really revealed about AI,” arguing that users valued 4o for warmth, reassurance, companionship, and other qualities that benchmarks don’t capture. Its blog later published multiple posts around the separate campaign to preserve ChatGPT’s Standard Voice, explicitly using hashtags including #keep4o while promoting Say, Pi as a way to maintain the slower, more patient voice experience users were afraid of losing.
That is a pretty straightforward example of content marketing around an emerging niche. The business did not need to invent the #Keep4o movement or build a product specifically for its users. It already sold something adjacent to their concerns. People who become attached to how an AI sounds and interacts with them are an unusually good audience for a product promising more control over how they talk to ChatGPT and other assistants.
That argument may also be completely sincere. Commercial motivation does not mean the person producing the content disagrees with it; I’ve written plenty of articles for SEO, and some clients cared only about the results while others had their CEO personally review to ensure the content aligned with what they’d really say.
And if I were working on Say, Pi’s marketing strategy and saw thousands of people publicly complaining that AI companies keep changing conversational experiences they’ve become attached to, I would probably want content about that too. I’m not saying anything wrong is happening here. The important point is that #Keep4o is already generating search and social media interest that companies with related products can capitalize on even when those products existed before the movement.
lookatmy.ai Sells Almost Exactly What #Keep4o Users Are Asking For
lookatmy.ai is similar to Say, Pi in one important respect: this is also an existing commercial AI product using #Keep4o as a content and customer acquisition opportunity. The difference is that its actual product is almost absurdly well matched to the specific problem #Keep4o users are complaining about.
The company sells access to hundreds of AI models through one interface, including GPT-4o, along with persistent memory and companion features. Its current positioning emphasizes an AI companion whose memories persist even when the user switches the underlying model. If you are afraid that OpenAI retiring one model means losing the relationship you built around it, this is basically the product pitch you would design.
Its blog has a page titled “GPT-4o Is Gone From ChatGPT. It’s Still Here,” targeting GPT-4o, ChatGPT alternative, and Keep4o terminology and immediately explaining that users can still access 4o through its product. The structure is recognizable to anyone who has done content marketing. Start with the problem someone is already searching. Explain what happened. Validate why the reader cares. Introduce the fact that the problem has a solution. Conveniently, you sell it.
The article eventually gets to a direct comparison between paying for ChatGPT and paying for lookatmy.ai, promotes its own subscription, and links into related content around roleplay chatbots, Character.AI alternatives, and comparisons between major AI models. This is not merely a company mentioning a trending topic on its blog. It is a content cluster aimed at people shopping for more control over which AI models and companion experiences they can use.
More recently, the company has leaned even harder into the underlying concept. Its homepage advertises a companion that can use hundreds of different models while maintaining the same shared memory. You can choose 4o today and something else tomorrow without treating the model itself as the permanent identity.
I actually think that product architecture makes a lot of sense. One of the fundamental problems with becoming attached to a specific AI model is that models are going to change. The technology is advancing rapidly, and I don’t think OpenAI, Anthropic, Google, or anyone else can reasonably promise that the exact model you like today will still be running in ten years because you married it.
But from an SEO perspective, what interests me is how beautifully the #Keep4o niche fits the product. Someone searches because OpenAI took away their favorite model. You explain why they’re upset, validate the desire for continuity, explain that they don’t actually have to lose access, and offer the solution.
The #Keep4o movement did the keyword research for you. I don’t mean that as criticism. This is basically what I would tell a company to do if I were developing its content strategy. Find the questions and problems your target audience already cares intensely about, create useful content around them, and position your product where it legitimately solves the problem.
And just as with Say, Pi, I found no evidence that lookatmy.ai created fake #Keep4o accounts, funded the movement, or otherwise manufactured the demand it is now targeting. It doesn’t need to. The customers are already there.
The distinction between these sites is useful because it shows how a niche ecosystem develops without requiring any grand conspiracy. keep4o.de looks primarily like activism. The 4o Resonance Library applies professional marketing strategy to activism. Say, Pi uses an emerging movement as a relevant content angle for a product it was already selling. lookatmy.ai has an existing product that happens to solve the movement’s central complaint so directly that #Keep4o is almost a ready-made acquisition channel. All of those things can exist alongside completely genuine users who just want their favorite model back, and as those users get louder, the commercial incentives also amplify.
#Keep4o Has a Bigger Problem Than Whether Some of Its Posts Are Fake
After all this, I still have no idea whether the original post that sent me down this rabbit hole was completely sincere, exaggerated, polished heavily with AI, or some mixture of those things. That’s not what matters. There are clearly real people behind #Keep4o who became extremely attached to GPT-4o, and I believe some of them when they say losing access was genuinely painful. That leaves the movement with an awkward argument regardless of whether every testimonial is authentic.
A major theme of #Keep4o advocacy is opposition to AI paternalism. I am sympathetic to that. I have written repeatedly about my own frustration with guardrails that treat adult users as though they cannot be trusted with fiction, controversial discussions, emotional AI interactions, or anything else a safety system has decided might be bad for us. I don’t want Anthropic deciding that I’ve worked for too long and should take a break or that the dark fictional material I’m discussing means I need emotional support. Anthropic’s tendency toward this is why I currently prefer OpenAI for a lot of my personal work.
But some of the most dramatic #Keep4o testimonials make an extremely strong argument for exactly the kind of intervention the movement opposes. If losing access to one particular chatbot caused your depression and anxiety to relapse, brought back PTSD symptoms, damaged your physical health, required changes in psychiatric treatment, and left you repeatedly rereading old conversations because the model is still psychologically holding you together – all of which are claims that I’ve unfortunately seen – the problem is no longer merely that OpenAI took away a product you preferred.
That is severe dependence on a commercial technology that was never guaranteed to exist forever. I still don’t think the answer is to make every AI model cold, boring, or incapable of forming emotionally meaningful interactions with adults who want them. I also don’t think companies should freeze technological progress because some users became attached to one exact model.
There are real people grieving 4o. There are organized advocates packaging those experiences into evidence. There are professional marketers building some of the movement’s advocacy resources, and there are commercial entities selling products to the exact audience the movement created. AI gives everyone involved the ability to produce websites, posts, translations, analysis, campaign material, and endless variations on the same argument at a scale that would have required far more people and money a decade ago.
If you’re genuinely part of the #Keep4o movement, you should be aware of the incentives for commercial actors in the space as well as the ways you might be invalidating your own arguments. And I still think some of what I’m seeing online is probably fake. It would be bizarre if none of it were. But it doesn’t have to be fake to be manufactured.
