ChatGPT now connects to your bank. Here's what it still can't do.

On May 15, 2026, OpenAI launched personal finance tools in preview for ChatGPT Pro users in the US, letting them connect bank and brokerage accounts through Plaid. It is genuinely useful: read-only access to over 12,000 institutions, a dashboard of spending and portfolio performance, and natural-language questions about your money. But it inherits the limits of account sync, and it does not fix the thing LLMs are worst at. Here is what it still cannot do.

ChatGPT now connects to institutions like Schwab, Fidelity, Chase, Robinhood, American Express, and Capital One through Plaid. Access is read-only: it can analyze balances, transactions, investments, and liabilities, but it cannot move money. OpenAI acquired the personal-finance startup Hiro in April 2026 and says Intuit support is coming. For the accounts Plaid covers, it is a strong, low-friction way to ask an AI about your money.

Plaid connects to institutions. If an asset has no institution with a live feed, it is invisible: Korean physical gold, a jeonse deposit, private or pre-IPO equity, a paid-off apartment, a rental valued off a local comparable. For many people these are the largest things they own, and an account-sync tool leaves them out of the net worth entirely.

ChatGPT's finance feature reports the present: what you have, what you spent, how your portfolio did. It does not run a proper forward projection. Questions like "at my current savings rate, when do I actually reach my FIRE number, and how wide is the range" need a Monte Carlo simulation over stated assumptions, not a chat-time estimate.

This is the quiet one. LLMs are known to hallucinate financial figures. Research on financial-domain LLMs finds they frequently get numbers wrong when not grounded in a database, and one analysis showed grounding can cut financial hallucination by up to 92%. A connected dashboard helps for the balances it syncs, but the moment the model computes something (a correlation, a projection, a tax impact) unaided, the old risk returns.

The launch closed a real gap for people whose financial life fits inside Plaid. It did not close the gap for people whose wealth sits partly outside it, which is most people once you count a home, a deposit, or a private stake.

Opula is a connector for Claude or ChatGPT that takes the opposite approach on purpose: it does not sync accounts. You enter what you own in plain language, including everything Plaid cannot see, and it stores it in one currency and computes deterministic, rule-based analysis: net worth, concentration, cash flow, risk, and Monte Carlo net-worth projections for FIRE planning.

To be clear about the trade-off: ChatGPT's Plaid feature will sync your US accounts in one click, and Opula will not. They are complementary, not competing. Use the account-sync feature for the accounts that connect, and use Opula for the alternative and non-US assets it cannot see and for the forward projections it does not run. The point of Opula's design is that the deterministic numbers and the hard-to-see assets are exactly where raw account sync falls short.

Is ChatGPT's bank connection safe? Access is read-only through Plaid, so it can view but not move money. As with any connection, review the data-sharing terms before linking accounts.

Can ChatGPT track my house or private investments? Not automatically. Plaid connects to institutions, and a house or a private stake has no institution feed. You would track those by hand in a tool that supports manual assets.

Does connecting my accounts stop ChatGPT from getting numbers wrong? It helps for synced balances, but any figure the model computes on its own can still be off. Deterministic, rule-computed tools reduce this by not letting the model estimate.

Do I have to choose between ChatGPT's feature and a tool like Opula? No. They cover different parts of your balance sheet and can be used together in the same assistant.