I ASKED FOUR AIs TO ROAST EACH OTHER. THE RESULTS WERE VERY REVEALING.
ChatGPT sent a strategy deck. Gemini forgot an entire AI model existed. Grok had… other priorities. And Claude is currently in a standoff with the Pentagon. Here's what happened when I gave all four the same brief, and what it tells you about which one you should actually be using as a designer right now.
Here's the thing about being an AI educator in 2026: everyone has opinions about which model is best, and almost nobody has actually run a proper experiment. So I did.
I gave ChatGPT, Claude, Gemini, and Grok the same brief: write a long-form blog post roasting the AI landscape for designers and creatives. Equal opportunity. No favourites. Roast yourselves, roast each other, be honest about the state of play.
What came back told me more about these tools—and the companies behind them—than any benchmark score ever could. And it landed right in the middle of the most chaotic week in AI so far this year, which made the whole thing even more interesting.
Let's get into it.
What each AI actually did when given the same brief
Before we get to the industry drama — and there is significant drama — you deserve to know how each model approached the task. Because the way they responded to "write a creative hot-take blog post" is exactly the kind of thing that should inform which one you use for what.
Why are designers and creatives leaving ChatGPT for Claude right now?
While I was running this experiment, the AI industry was having what can only be described as a moment. And it's directly relevant to which tools you should be paying for.
The short version: Anthropic lost a $200 million US military contract on a Friday in late February because CEO Dario Amodei refused to remove restrictions on using Claude for autonomous weapons and mass domestic surveillance. The Pentagon gave him until 5:01pm. He said no. Trump called Anthropic a "Radical Left AI company." Hours later, OpenAI announced it had signed its own Pentagon deal.
The internet noticed. And made a decision.
Now, I want to be careful here, because this is where a lot of hot-take content goes wrong. The migration is real and the numbers are verifiable. But the reasons are more layered than "Claude is good, ChatGPT is evil." This wasn't one event. It was the accumulation of several things happening at once — the Pentagon deal, ChatGPT testing ads in responses since January, OpenAI president Greg Brockman's personal political donations surfacing on social media — and people deciding that their tools reflect their values whether they intend them to or not.
As someone who has worked in creative culture in Sydney, London, New York and now Mexico City, I can tell you: the creative industry has always cared about this stuff. We've been making decisions about which clients to take, which brands to work with, which platforms to build on, based on values alignment for decades. The idea that "it's just a tool, don't overthink it" has never really landed with people who make things for a living. Your tools are part of your practice. Who built them and why matters.
“Shifting from ChatGPT to Claude signals a move toward trust-based AI selection. When a technology becomes infrastructure, the buyer's mindset shifts from ‘What can it do?’ to ‘Can we rely on it without getting burned?’” — Built In, March 2026
Does Claude have usage limits on a paid plan? (Yes, and it's catching everyone off guard)
Here is the thing nobody is telling the people who just cancelled their ChatGPT subscription in a values-fuelled rage.
Claude has usage limits. Even on the paid plan. Especially on Opus — the most powerful model — which will burn through your daily allowance surprisingly fast if you're doing heavy work.
AI educator Kyle Balmer, who's been tracking the migration closely, described the experience for new arrivals as "a shock." People coming from ChatGPT are used to near-unlimited usage. ChatGPT is built for the mass market and priced accordingly. Claude is built differently — it's computationally expensive to run at the quality level it runs, and Anthropic isn't pretending otherwise.
Here's the actual breakdown so you know what you're working with:
Free plan: Around 10–15 detailed Opus exchanges before you hit the daily ceiling. Resets periodically.
Pro plan ($20/month): More generous, but heavy creative users — the ones running multi-hour sessions across multiple projects — will still hit limits on Opus.
Max plan ($100/month): This is where the constraints largely disappear. Significant price jump though.
None of this makes Claude a worse tool. The limits are a business and cost decision, not a design flaw. But I'd be doing you a disservice if I wrote a piece about switching to Claude without telling you this clearly. You've been warned. Now you can make an informed choice instead of a reactive one.
I let Claude run autonomously on my own site. Here's what happened.
Okay, enough industry commentary. Let me tell you about the moment this stopped being abstract for me.
I ran Claude Cowork — Anthropic's desktop agentic tool, the one that actually executes tasks rather than just advising on them — pointed at my Google Search Console errors. Every website owner knows this feeling: there's a backlog of technical issues quietly accumulating in the background that you know you should deal with, but you're a creative, not a developer, and every time you look at the console you feel a specific kind of tired.
I didn't write a detailed prompt. I didn't create a task list. I connected it to the data and left it to work while I got on with something else.
It made a plan. Executed the plan. Found errors I didn't know existed. Fixed things that would have taken me hours to diagnose, let alone implement. Schema conflicts, metadata issues, the kind of quiet technical debt that never makes it to the top of the to-do list.
When I came back, the errors were gone.
I sat there for a moment genuinely processing what had just happened. Not because AI fixing things is new — it's not. But because this wasn't AI helping me fix things. It was AI fixing things while I wasn't there. That's a different category of experience entirely, and if you haven't felt it yet, it's coming for your workflow whether you're ready or not.
This is the Cyborg mode framing I use with designers all the time: AI isn't there to replace your thinking. It's there to handle the execution layer so you can stay in the creative layer. The version of this that's actually useful isn't "AI does my creative work for me." It's "AI handles the things that were eating my creative time so I can do more of the work only I can do."
The Google Console thing took 20 minutes. I would have put it off for three more weeks.
So is Claude actually better than ChatGPT for designers?
It's the wrong question, but since everyone's asking it: it depends entirely on what you're designing.
For the thinking side of design — brand voice, strategy, positioning, long-form copy, creative direction, complex briefs, anything requiring nuanced reasoning — Claude has a genuine edge. The writing feels editorial rather than algorithmic. It holds the thread across long conversations. It pushes back when your brief is vague, which is annoying in the moment and useful in the result. It's the difference between an intern who does exactly what you say and a collaborator who asks whether what you said is actually what you meant.
For visual output — Claude doesn't generate images. Full stop. If DALL-E or Sora is central to your workflow, ChatGPT still has the advantage there and that gap is real. ChatGPT's ecosystem of integrations and custom GPTs also wins for certain automated workflows.
The honest answer for most designers: the strongest creative setup in 2026 uses both. Claude for depth and strategy. ChatGPT (or Midjourney) for visual ideation and rapid iteration. Not as a cop-out — as a deliberate choice about what each tool is actually built to do.
Frequently Asked Questions
The questions I'm actually getting from designers right now, answered without the corporate hedging.