ChatGPT Wants to Be Your Financial Advisor

OpenAI announced this week that ChatGPT will soon let users link their bank and investment accounts directly to the platform in exchange for personalized financial guidance — expense breakdowns, portfolio reviews, goal planning, the works. Robinhood followed almost immediately with its own announcement that users would soon be able to execute trades entirely through AI agents. Two announcements in one week is the kind of drumbeat that makes people think something has shifted.

Maybe it has. Maybe it hasn't. The honest answer is that we're early enough that both things are true depending on which corner of this you're standing in.

Richard Coffin, a registered portfolio manager who runs the Plain Bagel YouTube channel, walked through the full landscape in a recent video — and he's a useful guide here precisely because he has skin in the game. He acknowledges upfront that he likes his job and would prefer to keep it, which at least makes his relatively measured assessment of AI advisors more credible than if it came from someone with nothing to lose.

The feature is currently being piloted for US users on ChatGPT's $200-per-month Pro tier. Coffin notes it supports more than 12,000 financial institutions, which means the connectivity isn't the obstacle — it's the everything-else that gets complicated.

What you're actually getting

The access piece is real. That's worth saying plainly.

The finance industry has a structural problem that's been there for decades: it doesn't have a good business model for serving smaller investors. A standalone financial plan from a qualified planner can run thousands of dollars. The advisors with the best credentials tend to work for firms that cap access at household minimums — a $250,000 floor is common, half a million isn't unusual. Everyone below that line gets routed to bank branch advisors, who may or may not be as qualified and who are almost certainly selling you something their employer wants sold.

AI doesn't fix that problem, but it does route around it. Anyone with a ChatGPT subscription and a linked account can get a financial plan in seconds, at any hour, on any question, without anyone on the other side of the screen raising an eyebrow about where your money actually is.

Coffin makes an observation I find genuinely interesting: people may be more honest with an AI than with a human advisor. Finance is a vulnerable subject. Debt, bad decisions, years of not saving — these things are easier to type into a chat window than to say out loud to someone who's going to write it down. It's the same reason people Google symptoms they wouldn't mention at a checkup.

That honesty gap is real, and it matters for the quality of the plan.

What you're not getting

Here's where it gets more complicated.

Large language models are, as Coffin puts it, "language prediction models. They're not judgment models." They're not researching earnings reports or reading footnotes or developing a thesis. They're predicting which words should follow which other words based on what they were trained on. That works surprisingly well for a lot of financial questions. It works less well for the specific, personalized, edge-case situations where people most need accurate guidance.

Coffin tested this in a small way during the video — asking ChatGPT for a list of stocks to hold for ten years, then rephrasing the request slightly. The list changed. Not dramatically, but it changed based on a difference in wording that shouldn't have mattered. For someone who knows how to prompt these systems, that's a manageable quirk. For someone who doesn't, it means they might be getting a different answer than they'd get with slightly different phrasing — with no way to know which one is more accurate.

The garbage-in problem compounds this. When a human advisor sits down with a client, they ask about health, family status, employment stability, things that don't show up on a balance sheet. An AI tool will build a plan out of whatever you tell it. If you forget to mention that you're supporting an aging parent or planning a major career change, the plan reflects a version of your life that doesn't exist.

OpenAI's own accuracy benchmarking, which Coffin references without a published citation we could independently verify, suggests the model handles financial tasks well — but "well" still leaves a meaningful gap. The question that matters isn't the average performance. It's what happens at the edges. As Coffin asks: is the model's failure mode a rounding error, or is it telling someone to pour their savings into a speculative asset? Those are very different error rates to live with.

The regulatory gap is not a technicality

This is the part that sounds like industry gatekeeping but isn't.

In the US and Canada, giving paid, personalized financial advice requires registration. The rules that come with registration — fiduciary duty, know-your-client requirements, know-your-product requirements, disclosure of conflicts of interest — exist because the history of unregulated financial advice is a long record of people losing money they couldn't afford to lose.

ChatGPT is not registered. OpenAI has not gone the route of Portfolio Pilot, an AI tool offered by Global Predictions Inc., which itself holds SEC registration as a Registered Investment Advisor. Coffin notes that even that registration required the CEO and chief compliance officer to pass the required examinations — the SEC, he says, still insists on the human knowledge element even for a platform arguing its advice is AI-generated.

Without registration, there is no fiduciary duty. There's no requirement to ensure the advice is appropriate for your specific situation before it's handed to you.

Coffin flags the Intuit partnership as a concrete example of how this plays out. OpenAI confirmed a partnership with Intuit — the tax planning software — to integrate tools into the ChatGPT Finance offering. What's confirmed is the partnership itself. Coffin reads the arrangement as one where Intuit is paying for that integration, framing it as a revenue model where OpenAI connects users who have revealed their financial needs with companies willing to pay for access to that audience. That reading — and it's Coffin's reading, not a confirmed public characterization from either company — points toward a potential conflict of interest: what happens when the platform has financial relationships with the products it might recommend to you?

To Coffin's credit, he notes this isn't unique to AI. In Canada, only around 3% of advisors are registered fiduciaries who are legally required to put client interests first. The difference is that when a human advisor crosses a line, there's a regulatory framework with consequences. With ChatGPT, Coffin says, "it really is use at your own risk."

The privacy question you should actually think about

OpenAI states that the platform can access your balances, transactions, investments, and liabilities, but cannot access account numbers or make changes to your accounts. If you disconnect, Coffin reads the terms as indicating your synced account data is deleted within 30 days — but under his reading of those same terms, your conversation history and what OpenAI calls "financial memories" persist until you manually delete them. That distinction — synced account data versus the record of what you told the platform about yourself — is Coffin's interpretation of the current terms, not a sourced citation from a specific OpenAI privacy policy document. Before acting on any of this, read OpenAI's actual current privacy documentation yourself, because Coffin's read and the platform's current policy may not match.

The point worth sitting with: you'd be telling a platform with 200 million monthly users exactly what your financial situation looks like, what your goals are, and where you feel vulnerable. That's a detailed profile. What gets done with it depends on terms of service that will change over time.

What this actually replaces — and what it doesn't

When robo-advisors launched, a lot of smart people called the death of the financial advisor. That didn't happen. A Vanguard poll found that "peace of mind" is the top reason people seek out human advisors — not asset allocation, not tax optimization, but the experience of talking to a person who knows your situation and can talk you down when the market drops fifteen percent in a week.

Coffin references the Dalbar studies, which have long argued that individual investors underperform the markets primarily because of emotional decision-making, not bad security selection. Some robo-advisor firms actually walked back their fully automated models and added human advisors when clients demanded it during volatility.

Think about March 2020. Or late 2022. The question isn't whether you had a good financial plan — it's whether you held to it when everything felt like it was on fire. An AI tool will tell you the plan; it cannot sit across from you and make you believe it's going to be okay. That's not a small thing.

Coffin's framing is useful here: treat ChatGPT like the insurance rep who shows up at the small business association meeting — confident, organized, probably knows the products — but verify everything before you act, ask plainly about what's in it for them, and do not mistake a polished pitch for independent counsel.

Used that way — skeptically, as a starting point rather than a finish line — Coffin thinks these tools are a genuine addition to the toolkit for underserved investors. So do I. The problem isn't the tool. It's the gap between what the tool appears to offer and what it actually provides. That gap is where people's actual financial lives get damaged.

Regulators will eventually wade into this. Until they do, the only protection users have is knowing exactly where that gap is.

---

Dorothy "Dot" Williams covers small business and Main Street economics for Buzzrag.