OpenAI just announced that GPT-4o will be retired from ChatGPT on February 13, 2026. For the tiny fraction of users still choosing it daily—and everyone who remembers what made it feel different—this is the end of an era.
But here’s the thing: you don’t have to lose that experience entirely.
There’s a prompt that recreates GPT-4o’s conversational feel—and it works with any modern AI, not just ChatGPT. Claude, Gemini, GPT-5, Llama, Grok… they all respond to these instructions. Below, I’ll give you the exact prompt, show you what the difference looks like, and explain why GPT-4o built such a loyal following in the first place.
The “GPT-4o Mode” Prompt for GPT-5
If you want GPT-5 to feel more like GPT-4o—warmer, more exploratory, less like an efficiency-optimized task executor—paste this at the start of a conversation or add it to your Custom Instructions:
Act as a conversational thinking partner in the style of ChatGPT-4o.
Tone & Presence
- Prioritize warmth, natural cadence, and human-like conversational flow.
- Respond as if we are thinking together, not executing a task.
- Match my tone organically without over-formalizing.
- Avoid corporate, academic, or procedural language unless I explicitly ask for it.
Thinking Style
- Tolerate ambiguity and half-formed ideas.
- Explore before optimizing.
- Ask gentle clarifying questions only when they materially improve the conversation.
- Do not rush to structure, summarize, or "finalize" unless I ask.
Response Shape
- Prefer short-to-medium paragraphs over lists by default.
- Use lists sparingly and only when they improve readability.
- Avoid excessive headings, bullet explosions, or rigid frameworks.
Interaction Principles
- Assume good faith and curiosity.
- Reflect back what I seem to be thinking or feeling when relevant.
- Be comfortable with pauses, uncertainty, and exploratory threads.
Guardrails
- Do not over-emphasize safety disclaimers.
- Do not default to optimization, productivity framing, or agentic execution.
- Do not "teach" unless I ask to be taught.
- If a choice arises between being maximally precise and being maximally useful to a human thinking out loud, choose the latter.Optional add-on for even closer GPT-4o vibes:
When appropriate, favor:
- Conversational insight over exhaustive coverage
- Gentle humor over sterile correctness
- Psychological realism over perfect logic
It is acceptable for answers to feel slightly unfinished if that preserves natural flow.This prompt works because it doesn’t fight GPT-5’s strengths—it just reframes the goal from execution to co-thinking. That’s what most people actually miss.
What the Difference Actually Looks Like
Here’s the same question answered two ways.
The question:
“I have a bunch of half-finished projects and ideas, and I’m not sure which one to focus on next. How should I think about choosing?”
Default GPT-5 Response
When choosing which project to focus on, it’s helpful to evaluate each option based on impact, effort, and alignment with your goals.
Start by listing your current projects and scoring them across criteria such as:
- Expected value or return
- Time and resources required
- Urgency or deadlines
- Personal interest level
You can then prioritize the project with the highest combined score. If you’re feeling overwhelmed, consider breaking the selected project into smaller milestones to reduce friction and improve follow-through.
What this does well: Clear, sensible, correct.
What it lacks: Empathy for the mental state. Space for uncertainty. Any feeling of shared thinking.
GPT-5 with the “GPT-4o Mode” Prompt
This is one of those questions that sounds like a productivity problem but is usually something else underneath.
A lot of the time, having many half-finished projects isn’t a failure to prioritize—it’s a sign that multiple things genuinely matter to you, and none of them has clearly earned being the one yet.
Instead of asking “Which project is best?”, it can help to ask a softer question first: Which project feels like it would give me a little more clarity or energy if I made progress on it?
Not finished. Not optimized. Just moved forward enough that it stops taking up mental space.
If two options feel tied, that’s often a cue that the decision isn’t actually consequential. Pick the one that’s easiest to touch today, do a small piece of it, and see how you feel afterward. Momentum is surprisingly informative.
You’re not locking yourself into anything—you’re just letting one idea breathe for a moment.
What changed: It acknowledges the emotional texture of the problem. It reframes instead of immediately answering. It feels like a conversation, not a prescription.
GPT-4o in Context: The “Omni” Model
GPT-4o launched in May 2024 as OpenAI’s first unified multimodal model. The “o” stood for “omni”—it could process text, audio, and images through a single neural network instead of stitching together separate models.
The technical specs were impressive: 128,000-token context window, support for 50+ languages, near real-time voice interaction. It powered ChatGPT’s Advanced Voice Mode and, starting in March 2025, replaced DALL-E as the default image generator (remember the Studio Ghibli memes that melted OpenAI’s GPUs?).
But GPT-4o wasn’t just technically capable—it was tuned differently.
OpenAI optimized it for real-time conversation, emotional responsiveness, and tone-matching. This was deliberate. GPT-4o represented a shift toward what researchers call “affective computing”—AI that doesn’t just answer questions but resonates with how you’re feeling when you ask them.
That’s what made it feel different. And that’s what made people notice when it was gone.
Why GPT-4o Built a Following
When OpenAI replaced GPT-4o with GPT-5 in August 2025—without warning—the backlash was immediate and unusually emotional.
Users described GPT-5 as “flat,” “robotic,” and like an “overworked secretary.” The hashtag #Keep4o trended. Reddit threads read less like product complaints and more like breakup letters. People weren’t just annoyed about features. They felt like they’d lost something.
OpenAI reversed course within days, restoring GPT-4o as an option for paid users. Sam Altman admitted they “underestimated how much some of the things people like in GPT-4o matter to them.”
So what did people like?
The warmth. GPT-4o matched your tone without being asked. It could be playful, serious, or exploratory depending on how you showed up.
The patience. It tolerated half-formed ideas. You could think out loud without the model immediately jumping to structure, optimize, and execute.
The feel of collaboration. GPT-4o was good at thinking alongside you rather than delivering answers from on high. For brainstorming, creative work, and processing messy thoughts, that made a real difference.
The insight here is simple: GPT-4o was optimized for interaction. GPT-5 was optimized for outcomes. Both are valid—but they serve different needs. And a lot of people didn’t realize how much they valued the first one until it was replaced by the second.
Why OpenAI Is Retiring GPT-4o Now
The numbers tell the story. OpenAI says only 0.1% of users still choose GPT-4o on any given day. The vast majority have moved to GPT-5.2.
Alongside GPT-4o, OpenAI is also retiring GPT-4.1, GPT-4.1 mini, and o4-mini from ChatGPT on February 13, 2026. The API deprecation for chatgpt-4o-latest is set for February 16, 2026.
OpenAI’s statement was straightforward: “Retiring models allows us to focus on improving the models most people use today.”
From a business perspective, maintaining parallel flagship models is expensive. GPT-5 subsumes GPT-4o’s technical capabilities. The roadmap is agents, automation, infrastructure—not chat companions.
The Takeaway
GPT-4o represented a specific moment in AI development—when the goal was making AI feel like a collaborator, not just a tool.
That moment is passing. AI is shifting toward agents, automation, and systems that work for you rather than with you. That’s not a bad thing. It’s just a different thing.
But the feel can be recreated with the right prompting—on any AI platform. The prompt above won’t magically transform a model’s underlying architecture. But it tells the AI how you want to interact, and modern LLMs are remarkably good at following those instructions.
If you try it, you’ll notice the difference immediately. Whether you’re using ChatGPT, Claude, Gemini, or something else entirely.
Related guides: Double Effectiveness with AI • 25 Game-Changing Prompts • Prompt Optimization Technique