Can AI Know What Stocks You Own?

The honest answer is a conditional one. Out of the box, no AI assistant knows what stocks you own. ChatGPT, Claude, Gemini, and every other general model start with zero visibility into your brokerage. They have read a lot of the public internet, but they have never seen your account.

That changes only when you take a deliberate step: connecting your brokerage to the AI through a separate tool. A purpose-built app like Walnut or PortfolioPilot links your account, or an MCP connector gives Claude or ChatGPT direct read access to your positions. The connection runs through a regulated aggregator such as SnapTrade or Plaid, and it is read-only by default. After that, and only after that, the AI knows what you own and can answer questions about your real portfolio rather than a generic example. Nothing happens automatically, and nothing happens without your authorization.

ChatGPT does not know your holdings because there is no link between the chat window and your brokerage. A large language model is trained on text; it is not wired into your Fidelity, Schwab, or Robinhood account. When you ask “how is my portfolio doing” with nothing connected, the model has no data to read, so it either answers with hypothetical numbers or asks you to describe what you hold.

This is a deliberate boundary, not a bug. Your brokerage data is private and sits behind your broker's authentication. An AI company cannot reach into millions of accounts, and you would not want it to. The assistant gains visibility only through an explicit, permissioned connection that you set up. That is why asking a bare ChatGPT or Claude about your money gets you generic answers: it genuinely cannot see anything you own. The same is true of Gemini, Perplexity, and any other general assistant. Knowing your holdings is a capability you grant, not a default the model ships with.

You make an AI portfolio-aware by connecting your brokerage to it through a secure aggregator. The mechanics are the same across reputable tools, and the security model is the part worth understanding.

When you connect, you do not hand your broker password to the AI tool. Instead, the tool opens your broker's own login flow through a regulated aggregator like SnapTrade or Plaid. You authenticate at the broker, the broker issues a scoped, read-only token back to the aggregator, and the AI tool reads your positions through that token. Your credentials never touch the AI tool, and the access is read-only by default, which is enough for analysis. The steps look like this:

1. Choose a tool that connects brokerages to AI, such as Walnut, or an MCP connector for Claude Desktop or ChatGPT.
2. Pick your broker from its list (most major US brokers connect through SnapTrade for read access).
3. Log in at your broker. The aggregator opens the broker's authentication flow, so your password is entered at the broker and never stored by the AI tool.
4. Authorize read-only access. The connection defaults to read-only, which lets the AI see your holdings without moving money.
5. Ask away. The AI now reads your real positions and prices and answers questions about your actual portfolio.

For a fuller walk-through of the routes and tools, see how to connect your brokerage to an AI assistant. The short version: a few clicks, no password sharing, read-only unless you decide otherwise.

Once an AI can read your real positions, the conversation stops being generic and starts being about you. A portfolio-aware assistant does four things particularly well.

• Concentration. It can tell you how much of your portfolio sits in a single stock or sector. If 40% of your money is in one name, or if three of your top holdings are all chipmakers, the AI sees it because it can read your actual weights, not a description you typed from memory.
• Overlap. It can find where your funds duplicate each other. If you hold an S&P 500 ETF, a total-market ETF, and a tech ETF, large-cap technology shows up three times, and the AI can quantify that hidden double-exposure across the holdings it reads.
• Performance versus a benchmark. It can frame each holding's return against a yardstick like the S&P 500, so you can see which positions are pulling their weight and which are lagging the market over the window you ask about.
• What-if scenarios. It can model a change before you make it: what trimming a concentrated position would do to your weights, or what adding a theme would do to your sector mix. Because it reads your real starting point, the answer reflects your portfolio, not a textbook one.

What ties these together is that the AI is reasoning over your genuine holdings. It explains and analyzes; it does not decide for you. For more on this capability, see what portfolio-aware AI is.

Letting an AI see your stocks is safe when the connection is built the right way, and the right way comes down to three things: read-only access, no password sharing, and the ability to revoke at any time.

• Read-only by default. A read-only connection lets the AI see your holdings and analyze them but gives it no ability to move money. This is the safer default and the one reputable tools use. The AI gets context without getting the keys.
• OAuth and scoped tokens, not your password. A safe connection routes your login through your broker's own authentication and a regulated aggregator (SnapTrade, Plaid). You enter your password at the broker, the broker hands back a scoped token, and the AI tool never sees or stores your credentials. Be wary of any tool that asks for your broker password directly.
• Revoke anytime. The access you grant is a permission you can withdraw. You can disconnect from the AI tool, from the aggregator, or from your broker's connected-apps settings, and the AI immediately loses visibility into your account. Disconnecting does not touch your holdings.

The risks worth avoiding are the inverse of those three: a tool that asks for your broker password directly, a connection that can trade without your approval, or a provider with no clear way to disconnect. For a deeper look at the security model, see is it safe to give an AI access to your brokerage.

Knowing what you own is not the same as being right about what to do with it. Even a fully connected, portfolio-aware AI has hard limits, and being honest about them matters.

• It is not a fiduciary. A tool like Walnut is informational and is not a registered investment adviser. It does not owe you the legal duty a human adviser does, and it does not know your full financial picture, taxes, or goals unless you tell it.
• It can be wrong. AI assistants can misread data, miss context, or state something confidently that is incorrect. Reading your positions accurately does not guarantee the analysis or any suggestion built on it is right.
• You should verify. Treat AI output as a starting point, not a final answer. Check the numbers against your broker, sanity-check the reasoning, and make your own decisions. The AI is a tool that helps you think, not a replacement for your judgment.

None of this makes a portfolio-aware AI useless. It makes it a powerful assistant rather than an authority. The right mental model is a knowledgeable second pair of eyes that can read your real holdings, not a manager you hand the wheel to.

Walnut is a portfolio-aware AI financial assistant that knows your portfolio because you connect it. You link your existing brokerage through SnapTrade, a regulated aggregator, so your broker login is entered at the broker and never stored by Walnut. From there, Walnut reads your real positions and lets you analyze and manage them by talking through Claude, ChatGPT, or a built-in assistant, with each holding framed against the S&P 500.

The defaults match the safe pattern described above. Walnut is read-only by default, so it sees your holdings without the ability to move money. If you choose to enable trade access at a broker that supports it, every order is approval-gated, so nothing executes without your explicit go-ahead. You keep your existing broker, and you can disconnect whenever you want. Walnut is also available as an MCP connector, so Claude Desktop, ChatGPT, or Cursor can read your connected portfolio directly. Walnut is not an investment adviser; it is informational.

From a connected account you can ask about a specific stock, an ETF you hold, or a theme you want exposure to, against your real portfolio. If you want to know which brokers support an AI connection, see which brokers have an AI assistant.

Can AI know what stocks you own? Not by default, but yes if you connect it. On their own, ChatGPT and Claude are blind to your brokerage and can only give generic answers. The moment you connect your account through a secure aggregator, read-only by default, the AI can read your real positions and answer questions about your actual portfolio: concentration, overlap, performance versus a benchmark, and what-if scenarios. That is what a portfolio-aware AI financial assistant like Walnut does. The connection keeps your login at your broker, stays read-only unless you decide otherwise, and can be revoked anytime. Just remember the limits: the AI is not a fiduciary, it can be wrong, and you should verify. Walnut is not an investment adviser.