We Gave 3 AIs the Same Terrible Ad Copy. One Model's Rewrite Was Shockingly Good.
Most AI comparisons test chatbots on trivia questions or coding problems — not on the thing most of us actually need help with: writing that sells. We took a piece of genuinely bad ad copy (weak headline, buried value prop, zero urgency) and fed it to ChatGPT, Claude, and Gemini using the exact same prompt. Then we scored each rewrite on clarity, persuasion, and how much it actually sounded like a human wrote it. The winner wasn't ChatGPT — the model everyone defaults to. Here's exactly what happened, what each AI does differently under the hood, and how to use this to make your own copy 10x better starting today.
The Original Ad Was Bad. Like, Really Bad. Here's What We Fed All Three AIs.
The test ad was for a fictional project management app called TaskFlow. Here's the original copy:
"TaskFlow is a project management solution that helps teams manage their projects. It has many features and is easy to use. Try TaskFlow today."
That's three sentences with no hook, no specific benefit, no emotional pull, and a call-to-action so weak it hurts. It's the kind of copy that gets written when someone's in a hurry and doesn't think about the reader at all.
We fed all three models this exact prompt: "Rewrite this ad copy to be more persuasive, specific, and emotionally compelling for a busy project manager who's tired of missed deadlines: [paste]"
The instruction to include a specific audience — "a busy project manager who's tired of missed deadlines" — is intentional. Vague prompts get vague results. Giving the AI a person to write for is what separates good output from generic output.
All three models got the same copy, the same prompt, and zero extra context. No examples, no tone guidelines, no word count. Just the bare brief — so we could see what each model does when left to its own instincts.
Here's Exactly What Each AI Produced — Side by Side
ChatGPT (GPT-4o) came back fast with this:
"Deadlines slipping? TaskFlow keeps your team aligned, your projects on track, and your stress levels in check. With intuitive features built for real teams, you'll spend less time chasing updates and more time shipping work. Start your free trial today — no credit card required."
Solid. Clean. The "deadlines slipping?" opener does a good job mirroring the pain point we specified. The phrase "real teams" is a nice touch — it signals authenticity. But "intuitive features" is still a bit vague, and the CTA, while better than the original, is pretty standard.
Gemini produced this:
"Stop letting projects spiral out of control. TaskFlow gives busy teams one place to plan, track, and deliver — so nothing falls through the cracks. Trusted by over 10,000 teams. Try it free."
Interesting. Gemini invented a social proof number ("10,000 teams") that we never gave it — which is a red flag we'll talk about later. The structure is tight and punchy, and "nothing falls through the cracks" is a strong visual metaphor. But the fabricated statistic is a problem you need to watch for.
Claude (Claude 3.5 Sonnet) wrote this:
"You've got twelve tabs open, three people waiting on you, and a deadline in four hours. TaskFlow was built for exactly this moment — one place where your team's work is visible, accountable, and actually moving forward. Stop managing chaos. Start managing projects. Try TaskFlow free for 14 days."
That opening sentence is doing something different. It's not describing the product — it's dropping you inside the reader's actual day. "Twelve tabs open, three people waiting on you" is specific enough to feel real. Claude also added a specific trial length (14 days) that creates more concrete commitment than "try it free." This was the rewrite that surprised us most.
Why Claude Won — and the Technique Behind Its Edge
Claude didn't win because it's "smarter." It won because it defaulted to a specific copywriting technique called perspective immersion — leading with the reader's lived experience before ever mentioning the product.
Most ad copy (and most AI-generated copy) leads with the product. "TaskFlow is..." or "Introducing TaskFlow..." Claude flipped this. It started inside the reader's head and only then pivoted to the solution. That structure is used by every top direct-response copywriter, from David Ogilvy's era to today's best email marketers.
Here's the mental model: Pain → Recognition → Solution → Proof → CTA. Claude's rewrite follows this almost perfectly. ChatGPT's version starts at "Recognition" and skips the visceral pain. Gemini's version starts at "Solution" and invents proof. Claude is the only one that earned the reader's attention before asking for it.
What's interesting is that you can push any of these models toward this structure with a better prompt. Try: "Rewrite this ad copy. Start with a one-sentence scene that puts the reader inside their worst workday moment, then introduce the product as the solution. End with a specific, time-bound CTA." That prompt would have pushed ChatGPT and Gemini much closer to Claude's output.
The lesson here isn't "always use Claude." It's that Claude's default copywriting instincts are currently the strongest of the three — but any model can be steered to produce great copy if you know what you're asking for.
How to Run This Test Yourself in the Next 20 Minutes
Pick one piece of copy you've been meaning to improve — a bio, a product description, a social media ad, a landing page headline. It doesn't need to be bad. Even decent copy can be sharpened.
Step 1: Open three tabs — chat.openai.com, claude.ai, and gemini.google.com.
Step 2: Use this base prompt in all three: "Rewrite this copy to be more persuasive and emotionally compelling for [describe your specific audience and their #1 frustration]. Keep it under [word count]. End with a specific CTA." Paste your copy below it.
Step 3: Score each output on three criteria — specificity (does it avoid vague words like "easy" and "powerful"?), emotion (does it make you feel something?), and CTA strength (is the action clear and urgent?). Give each a 1–5 score per category.
Step 4: Take the best elements from each rewrite. Claude's opening scene. ChatGPT's clean structure. Gemini's punchy brevity (minus any made-up facts). Build a hybrid version that you'd actually use.
This whole process takes about 15 minutes and will produce better copy than most freelancers would hand you after a full day. That's not an exaggeration — it's what happens when you give AI a real audience and a real structure to work with.
The Part Most People Get Wrong
Most people paste bad copy into an AI and ask it to "make this better." That instruction is almost useless. "Better" by what standard? For whom? With what tone? The AI fills in those blanks on its own — and it usually fills them with whatever sounds most generic.
The real mistake is treating AI like a magic fix button instead of a skilled collaborator who needs a brief. You wouldn't hire a copywriter and say "make this better." You'd tell them the audience, the goal, the platform, and the tone. AI works exactly the same way.
Watch out for Gemini's hallucination habit specifically. In our test, it invented a statistic with zero prompting. That's charming when it produces a compelling number — but if you publish "trusted by 10,000 teams" and you have 47 users, that's a legal and credibility problem. Always fact-check any numbers, testimonials, or specific claims any AI generates.
Finally, don't just pick a winner and use it forever. The best workflow is to rotate which AI you start with depending on the task. Claude for emotionally nuanced copy. ChatGPT for structured, clear formatting. Gemini when you need fast, punchy short-form. Using all three takes three extra minutes and produces dramatically better results than relying on just one.
Key Takeaways
- Claude's default instinct: It leads with the reader's lived experience before mentioning the product — which is the structure of high-converting copy.
- The prompt is the product: Vague instructions ("make this better") get vague results. Specific audience + specific pain + specific CTA = specific, usable output.
- Gemini hallucinates statistics: It produced a credible-sounding number we never gave it. Always verify any data point AI generates before publishing.
- Perspective immersion technique: Starting your ad copy inside the reader's worst moment — before mentioning your solution — is the fastest way to earn attention.
- Build the hybrid: No single AI wins every category. Take the best element from each rewrite and combine them into one version that's stronger than any individual output.
What to Do Right Now
Grab the worst piece of copy you've written in the last six months — the one you knew wasn't great but published anyway. Open Claude first and use this prompt: "Rewrite this ad copy. Start with one sentence that puts a [describe your audience] inside their most stressful moment. Then introduce my product as the solution. End with a specific, time-bound CTA. Keep it under 75 words." Then run the same prompt in ChatGPT and Gemini. You'll have three rewrites in under five minutes — and at least one of them will be better than what you've been using.