AI for Advisors newsletter
In mid-2025, we showed you how to use ChatGPT’s image tools to build simple infographics for client conversations. Since then, the underlying models changed more than almost anything else in this space—and the gap between “AI-generated” and “client-ready” has nearly closed.
Last July, in “How to Use AI to Create Infographics That Make Your Advice Clearer,” we walked through a four-element prompt formula—visual style, content focus, layout, color scheme—and three sample prompts for retirement conversations. That formula still works.
But it was written for a generation of tools that struggled with basic things: rendering readable text, organizing information cleanly, holding up under a follow-up edit. If you tried it and came away unimpressed, that wasn’t a prompting problem. That was a 2024 problem.
It’s worth a fresh look, because the tools underneath that formula have changed more than the formula itself.
The same prompt, a year apart
Here’s the simplest way to see it. Type this exact prompt into ChatGPT: “Create a picture of a financial advisor sitting at his desk.” No camera details, no style direction, nothing fancy.
In 2024, that plain prompt got you something that looked like AI—competent, but with the slightly unreal quality that gave the technology its “uncanny-valley” reputation—something that is human but also slightly off.
Better tools make it easier to produce something polished. They don’t make it easier to skip making sure it’s right.
Today, the same plain prompt produces something close to stock-photo quality. No 200-word camera-and-lighting prompt required to get there anymore.
The editing tells the same story. Run that same image through a few follow-up instructions—“make his hair salt and pepper,” “remove the tie,” “now show him after working 120 hours straight, sustained only by coffee”—and the model holds the likeness, the setting, and the intent across every edit.
A year ago, conversational editing like that was hit or miss. Now it’s reliable enough to build a workflow around.
What changed? In short, dramatically better text handling, native infographic and chart generation, layout that holds together without your help, and editing that actually listens to what you ask for next.
The old four-element formula—style, content, layout, color—still describes what to specify. It just describes a much higher ceiling than it did last year.
A sharper vocabulary for sharper tools
Last year’s formula will get you a reasonable result. If you want to push past reasonable, the more useful upgrade isn’t a new formula—it’s a small design vocabulary that lets you give the AI more precise instructions, the same way you’d direct a junior designer.
A few terms worth adding to your prompting toolkit:
Visual hierarchy—the order in which a viewer notices things. Clients should see the main point first, then supporting points, then detail. If a graphic feels flat or hard to scan, try: “Create a stronger hierarchy with a large headline, clear section titles, and smaller supporting text.”
White space—the empty space that lets a design breathe. Crowded graphics feel less trustworthy, not just less attractive. Try: “Add more white space and reduce visual clutter.”
Grid—the invisible structure organizing a page into rows, columns, or panels. This matters more for AI-generated work than hand-designed work, because a grid is what keeps the AI from drifting into randomness across a layout. Try: “Use a clean four-panel grid with equal spacing.”
Visual clutter—too many elements competing for attention, and probably the single most common failure mode in AI-generated infographics. The fix is almost always the same instruction: “Simplify the graphic and remove unnecessary decorative elements.”
Four terms won’t make you a designer. But they will get you a noticeably more polished result on the second pass than vague feedback like “make it better” ever will—because the AI now has a much better track record of doing exactly what specific design language asks for.
What the tools still get wrong
Better output quality creates a new risk that’s easy to miss: Confident-looking AI graphics are more persuasive than confident-looking AI graphics used to be, which means a wrong one is more dangerous, not less.
Here’s a demonstration worth running yourself. Prompt ChatGPT to build a line chart comparing two retirement savers—one starting at 25, saving $500 a month, one starting at 35 saving the same amount—projected to age 65. Don’t specify a rate of return. You’ll get back a clean, professional-looking chart in seconds, complete with a title, axis labels, and a dollar figure at age 65 that looks entirely authoritative.
Now ask it directly: “What annual rate of return is this compounding assuming?”
In our test, the AI had quietly assumed an 8% return—and hadn’t disclosed that anywhere on the chart. Worse, when we checked the math, the two lines weren’t even internally consistent with each other at that assumed rate. The chart looked finished, but it wasn’t accurate.
That’s the habit worth building: Never accept the first version as the deliverable.
Ask what’s baked into the numbers before you ask for anything else. One follow-up prompt—asking the AI to rebuild the chart at a stated 8% return and label the assumption directly on the image—turned the same flawed chart into something that actually held up: a labeled assumption, a table of balances at every age, and a footnote noting it’s a hypothetical example. That’s the difference between a draft and something you can put in front of a client.
The lesson isn’t that AI graphics are unreliable. It’s that AI is confident whether or not it’s right, so the verification step doesn’t get smaller as the tools get better—your job as the expert in the room gets more important, not less.
A few prompts to try this week
If you want updated versions of the ground we covered last year, three are worth testing:
The drag of taxes prompt: “Create a side-by-side bar chart showing growth of $100,000 invested for 20 years at 7% return. Left bar: taxable account with 25% annual tax drag. Right bar: tax-deferred account. Label the final dollar values on each bar. Title: ‘The Drag of Taxes on Growth.’ Professional, clean look.”
Social Security claiming options prompt: “Create a timeline infographic showing Social Security claiming ages at 62, 67, and 70. At each marker, display a sample monthly benefit. Use arrows to show growth over time. Keep the layout clear, numbers bold, professional style.”
Roth conversion timing prompt: “Create a side-by-side bar chart comparing Roth conversion outcomes: converting during low-income years versus higher-income years, both growing for 20 years. Title: ‘Why Timing Roth Conversions Matters.’ Professional financial chart style.”
Run each one, then push back on it the way you’d push back on a junior associate’s first draft. Ask what’s assumed. Ask for the hierarchy to be stronger, the clutter reduced, the grid cleaner. That back-and-forth—not the first prompt—is where the result actually becomes something you’d use with a client.
You’re still the editor
The update from a year ago is this: The ceiling on what AI can produce moved up substantially, and the floor on what you should accept without checking didn’t move at all. Better tools make it easier to produce something polished. They don’t make it easier to skip making sure it’s right.
That combination of real capability paired with real oversight is what actually builds client trust. Not the graphic itself, but your judgment about what goes in front of a client and what doesn’t.