OpenAI: ChatGPT enters personal finance with bank account integration

OpenAI launched a personal finance feature for ChatGPT Pro subscribers in the United States on May 15, 2026. The feature connects to bank accounts through Plaid, which supports integrations with over 12,000 financial institutions including Chase, Fidelity, Schwab, Robinhood, and American Express.

How it works

Once connected, ChatGPT presents a visual dashboard showing portfolio performance, spending patterns, active subscriptions, and upcoming payments. Users interact with this data through conversation, asking questions such as “Has my spending changed this month?” or requesting multi-year financial projections. An Intuit integration is planned to add tax impact analysis and credit approval predictions. Users can disconnect accounts at any time, with data removed within thirty days, and financial memories can be viewed and deleted separately.

Context

OpenAI acquired fintech startup Hiro in April 2026, and this feature is a first product output from that deal. The company reports that 200 million users already ask ChatGPT financial questions monthly, framing this as an expansion of an existing use case rather than an attempt to create a new one.

What it signals for product managers

The launch is a clear move toward vertically specific applications rather than remaining a general-purpose interface. For product managers watching AI product strategy, the approach is worth examining: establish a usage baseline through general conversation, identify domains where users already engage without prompting, then build dedicated infrastructure to serve those domains more precisely. Partnering with Plaid for banking integration — rather than building proprietary connections — also reflects an API-first approach to regulated verticals that others can study when entering similarly sensitive categories.

The combination of financial data sensitivity and AI inference also raises product design questions around trust signals, data transparency, and how to communicate model limitations in a domain where errors carry real consequences.