What to Ask Claude About Your Portfolio (Once It Can See It)

The comparison people most want and least often run correctly, because dividends and window choice both quietly change the answer.

• “How has my portfolio done against the S&P 500 this year, and over three years?”
• “Which of my holdings are dragging my return, and by how much each?”
• “If I had put everything in a total-market index fund instead, where would I be?”
• “What has actually driven my return: one position, or the whole portfolio?”
• “Show me my best and worst holdings by contribution rather than by percentage move.”

Concentration is the risk most people carry without choosing it, and fund names hide it.

• “What share of my portfolio is in my single largest company, looking through the funds?”
• “Which sector am I most exposed to, and is that deliberate?”
• “Do any of my funds hold substantially the same companies?”
• “How much of my portfolio is in one country?”
• “If technology fell 30%, roughly how much of my portfolio moves with it?”

The one input to future returns you control with certainty, and almost nobody has calculated theirs.

• “What is my weighted average expense ratio across everything I hold?”
• “Which of my holdings is the most expensive relative to a comparable index fund?”
• “How much am I paying in fund fees per year, in dollars?”
• “Is any of this money sitting uninvested?”

The question changes entirely once the assistant knows you own it.

• “I own NVDA. What has it done for me since I bought it?”
• “What does this company actually do, and why might it have moved this week?”
• “How much of my portfolio would I be selling if I exited this position?”
• “Is this position larger than I probably intended?”

The maintenance job people skip, because it means selling whatever has done best.

• “How far has each position drifted from its target weight?”
• “What trades would bring this back to target?”
• “Could I rebalance using new contributions instead of selling?”
• “Which account should I rebalance in to avoid a tax event?”

Sizing a new position against what you already hold, rather than in isolation.

• “If I put $500 into this idea, what would I actually be buying?”
• “Would adding this make me more concentrated or less?”
• “Do I already own this company inside one of my funds?”
• “What would this do to my sector weights?”

The pattern is that each needs either a view about the future or information outside your accounts. A confident answer to any of them is a reason for more scepticism rather than less, because confidence is cheap and the underlying uncertainty has not gone anywhere.

The last two are worth separating from the rest. They are not unanswerable in principle, they are unanswerable by this, because a portfolio connector sees connected brokerage accounts and not your tax return, your property, your pension or your spouse's holdings. See what portfolio analysis can and cannot do.

Ask for the number, then the interpretation. Getting the measurement first means you can judge whether the interpretation follows from it. Reversing the order means you get a narrative and then numbers selected to support it, which is harder to check.

Name the window. Portfolio performance is meaningless without one, and an unspecified window quietly becomes whichever one is most available. Asking for three years and one year and year-to-date, and noticing where they disagree, tells you more than any single figure.

One sequence shows the difference better than a list. Start with “how has my portfolio done against the S&P 500 over three years?” Suppose the answer is that you are four points behind. That is a fact, not a verdict, and the useful move is to ask why rather than what to do.

So: “which holdings account for that gap?” If the answer is one position, the portfolio is fine and one thing went wrong. If it is spread evenly, the cause is more likely structural, which points at the next question: “what is my weighted expense ratio?” A portfolio trailing by four points while paying 0.9% a year has explained most of its own gap.

Then the check that stops a bad conclusion: “what share sits in my largest company and largest sector?” Trailing a US large-cap index while holding bonds and international stocks is the diversification working rather than failing, and knowing your weights is what tells the two apart.

Four questions, each one narrowing the last, and none of them asking what to do. That is the shape that works: the assistant is very good at the narrowing and should not be making the decision at the end.

None of this works until Claude can read the account. Setup is a URL and a sign-in and takes about two minutes: how to connect Walnut to Claude. If you use a different assistant, the ChatGPT setup is the same connector, and what the connector is explains what it exposes and what it does not.