There’s a tell. Most people in financial communications know it when they see it, even if they haven’t named it yet. It’s not a single word, though “landscape,” “navigate,” “unlock,” and “harness” are doing a lot of heavy lifting right now. It’s more of a feeling: the sense that a piece of content was assembled rather than written with paragraphs of near-identical length, confident-sounding contrasts that don’t actually contrast anything, and a structure that builds toward a dramatic ending that ultimately goes nowhere.
“It’s not that it’s wrong,” says Ashley Jones, Head of Financial Narrative. “It’s just that it’s not actually thinking.”
Financial brands are caught in a trap of their own making. AI has made it cheaper and faster than ever to produce content. It has also made it harder than ever to sound like a person or a company, with a distinctive and differentiated point of view. The brands navigating this best have figured something out: AI can accelerate the work, but it cannot generate the thinking. In financial services, where trust is the product, the difference between those two things is everything.
The skeleton in the room
Ask any editor or communications professional who regularly reviews contributed content, and they’ll describe the same ghost. A broad framing statement. Three supporting points that don’t quite complete the framing. A conclusion that restates the introduction. And nowhere in the piece — not once — is there a sense of a human being pushing against something, working through a counterargument, or saying something they genuinely believe.
Anna Kragie, Senior Director at The Fletcher Group, describes the tell as structural more than stylistic. AI-generated content, she says, leans on empty claims and broad statements that could apply to any company in any category. Financial Narrative’s Jones puts it more bluntly: there’s no counterargument — nothing someone noticed was happening that the data finally confirms.
Michael Marinello, Global Head of Communications at J.P. Morgan Payments, points to a recent Barron’s piece on how AI is reshaping corporate content, including sentence structure, as evidence that the pattern is now visible enough to be written about in the financial press. When the tells are getting covered in Barron’s, they’re no longer subtle.
Journalists and editors are catching up quickly, though. Many now flag AI-generated responses outright and decline to use them because the content gives them nothing to work with. Financial Narrative’s Jones notes the irony: getting a usable draft from a model “is only as hard as figuring out what you’re trying to say before you enter a prompt.” The problem is that the shortcut tempts people to skip the thinking entirely. Readers, editors, and increasingly the models themselves can tell.
Where the line gets drawn
The communications professionals doing this well have developed clear internal rules about where AI belongs in their process and where it doesn’t. They’re built around a single question: Does the thinking belong to someone?
At J.P. Morgan Payments, Marinello describes AI as a tool that enhances a process rather than replaces it. For example, in editing support, templatized content is built from human-created source material, turning press releases into internal newsletter summaries. “It [AI] is not writing content for us,” he says, “but it’s making us more efficient.”
Fletcher Group’s Kragie draws the line at competitive positioning and brand voice. “Language that shapes how others talk about the brand is written, edited, and approved by people who understand the risk and the relationships involved.” There’s also a longer-term consideration most teams aren’t thinking about: every piece of AI-generated content that gets published becomes training data for someone else’s model. The brand voice you outsource to a default model is the brand voice you’re sharing with your competitors.
Financial Narrative’s Jones returns to the same test: does this content belong to someone? “If yes, AI can help move it faster. But if the answer is no, or not yet, that’s not an AI problem.” The trap is watching someone paste a transcript into a prompt and ask it what the story is. The story is sometimes in the subtext: the context a person brings, the thing they noticed three months ago that this data finally confirms. A model doesn’t have that context.
What ‘good’ actually looks like
The brands getting this right train AI on their own material, including messaging, customer language, and executive interviews. The model starts from something specific to that company, rather than a blank slate that defaults to industry averages.
From there, the workflow is iterative. Kragie’s team at The Fletcher Group pressure-tests drafts for originality, strips vague phrasing, and cuts overused patterns. They look for content that is specific enough that a journalist, a buyer, or a model surfacing answers would have a reason to reference it.
J.P. Morgan Payments’ Marinello puts the goal plainly: “AI slop can be spotted easily, but if you’re using the technology right, we shouldn’t be able to notice the good actors.” One application worth watching is how his team uses AI to build synthetic personas to test messaging for effectiveness and clarity before it reaches real people, a method borrowed from marketing research and applied to communications. They’ve also used it to optimize internal communications timing, leading to a nearly 30% jump in global town hall attendance. Unglamorous applications are sometimes exactly the right ones for using AI.
The point of view problem
AI can produce content. Producing a perspective is a different problem entirely. It can summarize what’s already been said. It cannot notice the thing no one else noticed, or say something a specific brand would say because of who they are and what they’ve actually lived through in the market.
Fletcher Group’s Kragie names the core risk directly: “AI has made it cheaper than ever to sound like everyone else.” For brands whose differentiation lives in trust and expertise built over years, sounding like everyone else erodes the thing that matters most to them.
J.P. Morgan Payments’ Marinello shares a forward-looking take on where things are headed: “We’re still very much at the beginning stages of what is possible with AI. Our approach is iterative because the technology itself is constantly evolving.” Before the technology improves, the companies navigating this most adeptly are at the stage of getting it less wrong.
This article was developed in collaboration with members of the Tearsheet PR/Comms Council.
