When ChatGPT Goes Flat—The 10 Second Fix

AI for Advisors newsletter

Lately, I’ve heard a version of the same complaint from people: “ChatGPT just isn’t as good as it used to be.”

If that sounds familiar—if your AI’s tone feels choppy, the output bland, or the structure confusing—you’re not imagining it. You’re likely dealing with something no one told you about: You’re using a different model than before. And the good news? You can fix it in less than 10 seconds.

What changed and why it matters

Back in early August 2025, OpenAI rolled out a variety of new models under the GPT-5 banner, while moving GPT-4 into its Legacy models. That rollout included new defaults in tools like Projects and custom GPTs.

If you’ve been using AI regularly, you may not have realized that your model switched. One day you’re getting crisp, clear responses—and the next, everything feels...off.

‘The Berkshire Test: Buffett vs. Munger’

Here’s how I think of it: Using GPT-4o feels like working with Warren Buffett—clear, conversational, and always able to explain complex ideas in plain English. Using GPT-5? That’s more like Charlie Munger—all sharp logic and disciplined frameworks, but less warmth. Informative? Absolutely. Inspiring? Not always.

And when you’re writing for clients or creating marketing, that difference in tone matters.

It’s not that GPT-5 is wrong. It’s just not what most advisors need when writing for clients, building marketing, or doing anything that requires tone, polish, or persuasion.

Here’s how to fix it

When ChatGPT isn’t delivering like it used to, do these two things first:

1. Switch back to GPT-4o: In the model selector dropdown, scroll to the bottom and select GPT-4o under Legacy models. This instantly restores the tone and fluency you were used to.
2. Embed model awareness in your tools: I now include this line at the very top of all my Project and GPT instructions:

That simple reminder saves time, reduces friction, and helps keep your outputs consistent.

One prompt, two different worlds

Here’s a quick experiment that shows just how different these models can be—even when given the exact same prompt.

I asked both GPT-5 and GPT-4o to write a single paragraph about Charlie Munger, aimed at a financial advisor audience. Same instructions. Same tone. Same objective. The goal was to blend storytelling with professional insight—something advisors do all the time in client communications.

Here’s what happened:

The Charlie Munger prompt:

“Write a single, narrative-style paragraph (no more than five sentences) about the life and times of Charlie Munger. Your audience is financial advisors, so emphasize both his partnership with Warren Buffett and the practical lessons advisors can draw from his disciplined thinking. The tone should balance storytelling with professional insight, weaving his biography into a clear takeaway that advisors might apply in their own work.”

GPT-5 response:

Charlie Munger’s life was a study in disciplined thinking, sharp judgment, and the power of partnership. As Warren Buffett’s longtime collaborator at Berkshire Hathaway, Munger pushed Buffett to move beyond “cigar-butt” investing and toward owning great businesses for the long haul—a shift that created staggering wealth. His trademark mental models, drawn from psychology, economics, and history, reinforced the value of seeing the world as it really is, not as we wish it to be. For financial advisors, the practical lesson is clear: long-term success comes not from chasing fads but from cultivating patience, clarity of thought, and the courage to ignore market noise. Munger’s legacy is a reminder that disciplined frameworks, applied consistently, can transform both portfolios and professional practice.

GPT-4o response:

Charlie Munger, the irascibly brilliant vice chairman of Berkshire Hathaway, spent decades as Warren Buffett’s closest partner and intellectual equal, shaping one of the most successful investment philosophies of the modern era. A former lawyer with a razor-sharp mind and deep respect for multidisciplinary thinking, Munger brought a fierce clarity to decision-making—eschewing emotion, embracing patience, and insisting on understanding one’s “circle of competence.” His influence helped move Buffett beyond cigar-butt investing to a focus on high-quality businesses bought at fair prices, a shift that advisors can study for its disciplined balance between value and quality. Munger famously quipped, “Invert, always invert,” a reminder that avoiding stupidity is often more valuable than chasing brilliance—a lesson in risk management that applies as much to client portfolios as it does to life. For financial advisors, Munger’s legacy is less about stock picks and more about cultivating a mindset: rational, curious, independent, and utterly devoted to long-term thinking.

The comparison

• GPT-5 delivers streamlined efficiency—direct sentences, accessible vocabulary, and clear action items that busy professionals can absorb quickly. It reads like an executive briefing.
• GPT-4o offers richer texture—complex syntax, literary flourishes like “irascibly brilliant,” and detailed storytelling that builds character before delivering insights. It reads like a thoughtful magazine feature.

Neither is “better” in all cases. But if you’re writing for clarity, storytelling, or readability—especially for clients—GPT-4o often gives you a smoother experience.

And that’s the point: The model you choose will shape the tone and trajectory of everything you write.

Sometimes it’s the prompt, not the model

To be fair, poor results aren’t always about model choice. The second most common issue I see is vague prompting. Even GPT-4o struggles when the prompt lacks clarity. That’s why I always recommend prompting with structure using this six-part formula, RTF-CQE, which stands for Role, Task, Format, Context, Questions, Examples.

That one shift in prompt language—from “make this better” to “act as a marketing coach, rewrite this client letter for clarity and urgency, etc.”—can instantly elevate your output.

Generative AI isn’t deterministic like traditional software (that either works or doesn’t work). There’s no locked script. These tools use probability, not certainty. That means two people can enter the same prompt at the same time and get similar, but not identical results.

So, when things go sideways and you feel like your AI chatbot isn’t writing as well as it has in the recent past, be sure to:

• Check your model. It makes a difference in writing style and tone. (I recommend ChatGPT’s Legacy mode 4o.)
• Check your prompt and add more context. It also makes a difference in style and tone.

Good results come from model awareness and prompt precision. Master those, and you’ll always find your way back to solid cowriting.

Ready to make the leap? Horsesmouth’s AI for Advisors Pro training programs provide the structured, advisor-specific approach that transforms occasional users into confident practitioners. Learn more at www.horsesmouth.com/aipro.