How to Use ChatGPT to Analyze Your Portfolio

Start here because it is the step most people skip. Ask plain ChatGPT “how is my portfolio doing?” and it will answer with hypothetical examples and general principles, because it has no connection to your brokerage account. It does not know your tickers, your share counts, your weights, or today's prices. The advice is correct in the abstract and useless for your actual money.

That gap is the whole problem to solve. Any real analysis (your concentration, your overlap, how your account has done versus the S&P 500) requires the model to know your real positions. So before you ask anything useful, you have to get your holdings in front of it. There are two ways to do that, and they trade off differently on speed, accuracy, and upkeep.

The right mental model is an AI financial assistant that knows your portfolio. ChatGPT supplies the plain-language interface; the holdings supply the substance. Walnut is the layer that connects your real, read-only holdings so the assistant answers from your account rather than from market averages. Walnut is not an investment adviser, so what you get back is information about your own positions, not a directive to buy or sell. People sometimes call this “talking to your portfolio,” which is a fair description of the experience, but the practical question is just how you feed ChatGPT your holdings.

The fastest way to analyze your portfolio with ChatGPT is to paste it in. Copy a list of your positions, ideally tickers with share counts or dollar values, drop them into the chat, and ask your question. ChatGPT reads the list as text and reasons over it. A clean paste looks like “NVDA 30 shares, AAPL 50 shares, MSFT 25 shares, VOO 40 shares,” and then a question such as “how concentrated am I, and what is my largest sector exposure?”

This genuinely works for a one-time look, and it is free. But it half-works, and the limits matter:

• It is a static snapshot. Prices go stale the moment you hit send. The analysis reflects the numbers you pasted, not where your account stands tomorrow. To get a fresh read, you re-paste.
• It invites arithmetic mistakes. A list of 30 or 40 tickers gives the model a lot of numbers to add up, and it can miscount a weight or misread a value confidently. You cannot easily catch the error because you did not see the math.
• It needs constant upkeep. Every time you buy, sell, or rebalance, your pasted list is wrong until you redo it. For a recurring habit, that friction adds up fast.
• It only knows what you typed. If you leave out a position, omit current prices, or forget cost basis, ChatGPT cannot fill the gap. It analyzes the text in front of it, nothing more.

So pasting is the right tool for a quick, occasional check, and a poor tool for tracking your real account over time. If you only want to feel out the kind of questions ChatGPT can handle before you commit to anything, this path is a fine place to start.

The durable way to analyze your portfolio with ChatGPT is to connect your brokerage so the assistant reads your live positions itself, instead of you pasting them. This removes the snapshot problem entirely: the data is current every time you ask, the assistant pulls prices and weights directly, and you never re-paste a list. The tradeoff is a short one-time setup.

ChatGPT cannot connect to a brokerage on its own, so this needs a bridge. Walnut is that bridge: it links your existing broker account read-only through a regulated aggregator, then lets you ask about your real holdings through ChatGPT, through Claude, or through a built-in assistant on the site. Here is what the setup looks like:

1. Connect your brokerage, read-only. You link the account you already have through a regulated aggregator. The aggregator opens your broker's own login screen, so your password is entered at the broker and never stored by the tool. Walnut connects most major US brokers through SnapTrade and is read-only by default, which is all that analysis needs. For the full walkthrough, see how to connect a brokerage to an AI assistant.
2. Pick your assistant. With Walnut you can ask through ChatGPT, through Claude, or through the built-in assistant. The connection underneath is the same; the medium is your preference.
3. Ask in plain language. Type the question the way you would say it. The assistant reads your live positions to answer, framing each holding against the S&P 500 where that helps. No re-pasting, no stale prices.

Under the hood, the assistant has tools that read your connected portfolio: your positions, their current prices, how each has moved over a window, and how they sit relative to a benchmark. You see plain English; the data plumbing stays out of the way. If you want the concept behind whether an AI can read your account at all, see whether AI can know what stocks you own.

Once ChatGPT can see your holdings, whether pasted or connected, the fastest way to get value is to ask sharp, specific questions. These are grouped by what you are trying to understand. A portfolio-aware assistant answers each from your real positions rather than from generic averages. The answers are informational, not recommendations.

Concentration: how much rides on one bet

• “How concentrated is my portfolio?”
• “What are my five largest positions, and what percent of the account is each?”
• “How much of my portfolio is in my single biggest holding?”
• “If NVDA dropped 20%, roughly how much of my total account would that move?”

Overlap: where your holdings double up

• “Do any of my holdings overlap?”
• “If I own both VOO and QQQ, how much do they hold in common?”
• “Am I effectively buying the same companies through more than one fund?”

Performance versus the S&P 500

• “How is my portfolio doing versus the S&P 500 this year?”
• “Which of my holdings has done best and worst over the last twelve months?”
• “Is AAPL keeping up with the market, or lagging it?”
• “Which positions are dragging on my overall return?”

What-if and hypotheticals

• “What would change if I added MSFT to this account?”
• “If I sold half my NVDA, how much less concentrated would I be?”
• “How would adding a broad index ETF shift my tech exposure?”

Tax-efficiency and structure

• “Which of my holdings tend to be the least tax-efficient to hold in a taxable account?”
• “Do I hold anything that pays a high distribution I am being taxed on?”
• “Which positions are short-term versus long-term in how I have held them?”

Sector and theme exposure

• “Am I too exposed to one sector?”
• “How much of my account is in technology versus everything else?”
• “Group my holdings by theme so I can see where my money actually is.”
• “Do I have any exposure to energy, healthcare, or financials at all?”

Notice the pattern: every question is something you could ask a knowledgeable friend who happened to have your account open in front of them. That is the experience this recreates. For a structured single-pass read of the whole account rather than a back-and-forth, an AI portfolio analysis covers the same ground in one go.

Being honest about the failure modes is what keeps ChatGPT useful for this. It is a fast way to understand what you own. It is not an oracle, and it is not a fiduciary. The errors cluster in a few predictable places.

• Hallucinated numbers. ChatGPT can invent a price, a market cap, or a weight that sounds right and is not, stating it with full confidence. This is most common when it is filling a gap the data did not cover. Never act on a specific figure without checking it.
• Stale data. A pasted list is frozen at the moment you sent it, and the model's own training data has a cutoff. Prices, holdings, and fund compositions move; the answer may reflect last year, not today. A live connection reduces this, but does not eliminate the model's tendency to reach for an old figure.
• Arithmetic slips. Adding up weights across a long holdings list is exactly the kind of multi-step math language models fumble. A concentration figure can be off by several points if one position is miscounted.
• No cost basis. Most broker feeds do not pass cost basis, so returns are usually framed as window returns (how a position moved over a period) rather than realized profit and loss. ChatGPT cannot conjure a number that is not in the data.
• It does not predict prices. No model knows where a stock is going. It reports on what you own and how it has done; anyone claiming a forecast is guessing.

The rule that fixes all of this: treat ChatGPT's output as a fast first pass, then verify anything you are about to act on, especially specific figures, against your brokerage statement. You stay the decision-maker. Walnut is not an investment adviser, so the conversation is information, not a directive.

The choice comes down to how often you want to do this. Pasting is right for a quick, occasional check when you just want a read in the next two minutes and do not mind that it goes stale. Connecting is right when you want to ask about your real account regularly and want the data to stay current without re-pasting. Here is how the two approaches compare across the things that actually matter.

Neither is wrong. Pasting is the fast first taste; connecting is the durable habit. Walnut is built for the second case: read-only by default, with any trade gated behind your explicit approval, so the assistant reads your real positions but never moves money on its own.

To use ChatGPT to analyze your portfolio, you first have to get your holdings in front of it, because ChatGPT cannot see your brokerage account by default. You either paste your positions in, which is fast but a static snapshot that goes stale and invites math errors, or you connect your brokerage so the AI reads your live positions, which stays current with no re-pasting. Once it can see your account, ask sharp questions about concentration, overlap, performance versus the S&P 500, and sector exposure, and verify any specific number against your statement. Walnut is an AI financial assistant that connects your real, read-only holdings so you can ask through ChatGPT or Claude, with any trade gated behind your approval. Walnut is not an investment adviser, so keep the final decision your own.