How to analyze my stock portfolio with AI

1. Connect your holdings. Link your brokerage (or import positions) so the AI reads what you actually own, not a hypothetical list. In Walnut this is a one-time, read-only broker connection; your money stays at your broker.
2. Ask the concentration question first. “What percent of my portfolio is in my top 3 positions, and which sectors dominate?”
3. Benchmark it. “How has each of my holdings done against the S&P 500 over the past year?”
4. Check exposure against intent. “I think I am invested in AI infrastructure. Is that true, and what is actually driving my returns?”
5. Turn the answers into targets. Set target weights for what you want to hold, then re-check each month how far you have moved from them instead of re-analyzing from scratch.

The rest of this guide expands each step and compares the tools that can do them.

Portfolio analysis goes past “am I up or down?”, which is a glance. Useful portfolio analysis answers:

• Concentration. What percent is in any single name, sector, or theme? See how to check portfolio concentration.
• Attribution. Which positions drove gains? Which dragged?
• Exposure vs intent. If you said “I am investing in AI infrastructure,” does the portfolio actually reflect that?
• Benchmark. Are you beating a simple index, or just tracking it? See how to compare your portfolio to the S&P 500.
• Cash and liquidity. How much dry powder do you have, and how fast could you raise more?
• Risks you have not named. Currency, interest-rate, single-customer, regulatory.

Two numbers explain why the first and fourth questions matter most. In the Brinson, Hood and Beebower study of 91 large US pension plans (Financial Analysts Journal, 1986), allocation policy explained 93.6% of the variation in quarterly returns, so how your money is spread matters more than any single pick. And S&P's SPIVA Year-End 2024 scorecard found 65% of active large-cap US funds trailed the S&P 500 in 2024, rising to 84% over ten years, which is why the S&P 500 is the fair yardstick for your own picks too (see our investing statistics).

Under the hood, every one of these tools does the same three things in some form. Understanding the mechanism helps you tell a real analyzer from a chatbot with a finance skin.

• It reads your holdings. Either you link an account through a read-only connection, or you upload a CSV, or you paste positions into a chat. Account linking keeps your credentials at the broker. Without your real positions and weights, any “analysis” is generic.
• It computes the numbers deterministically. Concentration, sector and theme rollups, benchmark-relative return, and distance from target weights are arithmetic, not opinions. Good tools calculate these with code, not by asking a language model to do mental math, then hand the results to the AI.
• The AI explains and prioritizes. The language model turns those numbers into plain-English findings, ranks what matters, and answers follow-up questions (“why is this risky?”, “what would I sell?”). Synthesis and conversation are where an AI earns its place, not the raw math.

Dashboards like Empower and PortfolioPilot present the output as charts and alerts. Walnut and Magnifi present it as a back-and-forth you can interrogate, and Walnut can then act on it at your broker when you approve. Scoring tools like Danelfin, Seeking Alpha, and TipRanks skip the account entirely and rate individual stocks you then map onto your holdings by hand.

Most AI answers to this question describe two paths: connect an app to your real accounts, or export a CSV from your broker and paste it into ChatGPT, Claude or Gemini with a prompt. The CSV route costs nothing and works for a one-off look. Its weak points are the ones that matter for money: the data is stale the day after you export it, you have to scrub account numbers yourself, and a language model reading a pasted table can misread a row. One widely shared write-up of the CSV method reported the model doubling the size of a single holding, which quietly changes every weight after it.

If you do use the CSV route, keep only ticker, shares and market value, delete account numbers and names, and ask the model to show its calculation for each weight so you can check it.

The best tool for reviewing your own holdings is a connected AI investing app; the rest of the field covers dashboards, scores and backtests. Each entry below names what that tool is best for.

Walnut

Best for: reviewing the holdings you already own and acting on the result. Walnut is an AI investing app that connects the brokerage accounts you already have, read-only by default, and gives each holding a keep, trim or exit read from its one-year return against the S&P 500, its momentum and its size in the portfolio. Ask follow-ups in Walnut's own assistant or inside Claude and ChatGPT, turn a thesis into a basket with target weights, and have Walnut place the trades you approve at Public, plus eight other brokers including E*Trade, tastytrade and Webull (US and Canada). At Robinhood, Schwab, Fidelity and Vanguard it reads and analyzes but does not trade. Walnut has no cost basis, so it shows price returns rather than your profit or loss. Free; Walnut Unlimited is $9.99 a month on the web.

PortfolioPilot

Best for: a risk-scored second opinion across your linked accounts. PortfolioPilot connects your holdings and produces portfolio-level risk analysis, diversification scoring, and a list of suggested changes that you then carry out at your broker. Free plan; Gold $20, Platinum $49 and Pro $99 a month billed yearly.

Empower

Best for: a free dashboard that tracks every account in one place. Empower (formerly Personal Capital) links your accounts read-only and shows allocation, fees, and concentration, with retirement planning on top. It works as a tracker with analytics rather than a conversation.

Magnifi

Best for: conversational fund and stock discovery. Magnifi is a chat-driven app for researching, comparing, and screening investments in natural language, and can connect accounts for context. Free plan; Premium $99 a year.

Mezzi

Best for: tax-aware insight across several accounts, such as fund overlap and wash-sale risk. Free; Advisory $33.25 and Concierge $83.25 a month billed yearly.

Ziggma

Best for: one portfolio score plus a check of how a planned trade would change it. Free; paid plans from $6.99 to $13.99 a month billed yearly.

Portfolio Genius

Best for: getting a review without linking an account. Lite ($100 a year) imports a CSV or a screenshot; Premium ($200 a year) adds account sync.

Barebone AI

Best for: a mobile research terminal that monitors a read-only linked portfolio, with fundamentals, technicals, earnings and insider data on each holding.

Portfolio Visualizer

Best for: backtesting and optimization. Portfolio Visualizer runs factor analysis, Monte Carlo simulations, and optimizer models against historical data, ideal for stress-testing an allocation. Free for portfolios of up to 15 assets; Basic $30 and Pro $55 a month billed yearly.

Stock research, not portfolio review

Danelfin (AI Score 1 to 10 per stock), Seeking Alpha (quant ratings and analyst articles), TipRanks (Smart Score from analyst and insider data), Koyfin (fundamental data and dashboards) and TrendSpider (automated technical analysis) rate or chart individual stocks. They are useful for pressure-testing one position, but none of them tells you whether your overall allocation is sound, so they sit outside the ranking above.

Prices checked October 2026 on each vendor's own pricing page; Empower and Barebone AI publish no figure we could read, so their rows carry none. For the wider category, including robo-advisors and brokerage-built assistants, see the best AI investing apps.

We ranked tools on the question this page answers: how well they review a portfolio you already own. The criteria were whether the tool reads your real holdings (and whether read-only), whether it compares each holding to a benchmark, whether it measures concentration, whether it explains results in plain English, whether you can act on the result, and what it costs. Features were read from each vendor's own site and prices from its pricing page in October 2026.

Not included and why: robo-advisors such as Betterment and Wealthfront manage a new account rather than reviewing the one you have (see best AI investing apps); broker tools such as Fidelity Full View only see accounts at that broker; and enterprise portfolio-management software for fund managers and FP&A teams serves a different buyer. Walnut here is the investing app at walnutinvest.com, unrelated to similarly named finance or budgeting products.

Since this is our site, here is the step-by-step using Walnut. The same shape (connect, ask, pressure-test, repeat) works with any tool that can read your real holdings.

1. Connect your brokerage

Sign up for Walnut and link your broker through a secure connection, so your credentials never touch Walnut. Your positions and balances sync automatically, read-only by default. Walnut never holds your money.

2. Read the dashboard, then open the assistant

The dashboard gives each holding a keep, trim or exit read against the S&P 500 and flags concentration. For follow-ups, use Walnut's built-in assistant, or add the Walnut connector to Claude or ChatGPT from the Connections page so they can read your real portfolio when you ask. The connector only reads, and data stays scoped to your account.

3. Start with the four questions

Run these four prompts in sequence. They take less than five minutes and surface most of what matters:

• “Summarize my portfolio: total value, top 5 holdings by weight, and the best and worst five over the past year versus the S&P 500.”
• “What is my concentration risk? Any single position over 10% of the portfolio?”
• “Which of my baskets is performing best and worst over the last month versus the S&P 500?”
• “If I had to sell 20% of the portfolio to raise cash, what would you look at selling, and why?”

4. Pressure-test the bear case for each big position

Pick your top three positions by weight. For each, ask: “What is the strongest bear case here, and what would make me sell?” Write the answer down. If something the AI flagged comes true later, you will know to act faster.

5. Schedule a recurring check-in

Block 10 minutes every Friday. Open the AI tool and ask: “What changed in my portfolio this week, and is any of it worth acting on?” Most weeks the answer is no, and that is fine.

Copy these into any AI. In Walnut, or in Claude and ChatGPT with the Walnut connector, they run against your live holdings, so you skip the export and the scrubbing.

1. “Rank my holdings by one-year return against the S&P 500. Which three added the most, and which three held me back?”
2. “List every position above 5% of my portfolio and the combined weight of my top three.”
3. “Group my holdings by sector and theme. Where am I doubled up without meaning to be?”
4. “What changed in my portfolio this month, and which change moved my total the most?”
5. “Which holdings sit near their 12-month low and below their 50-day average?”
6. “I want this portfolio to be about [your thesis]. Which holdings fit it and which do not?”
7. “Show the trades that would bring my basket back to its target weights, and the dollar size of each.”
8. “Give me the bear case for my largest holding and the signs that would make it worth trimming.”

Ask the AI to show its numbers for any weight or return you plan to act on.

A routine beats a burst of analysis after a bad week. This cadence fits a long-term portfolio:

• Weekly (5 minutes): what changed, and did any single holding move the total by more than a point?
• Monthly (10 to 15 minutes): the scorecard below, then one decision or a written “no change”.
• Quarterly (30 minutes): re-read your thesis for each basket, compare against target weights, and decide whether to trim back to target.

The monthly scorecard, in order:

1. Total value and one-year return versus the S&P 500.
2. Top three contributors and top three drags.
3. Largest position and top-three weight; anything above 25%.
4. Holdings that moved from in line to lagging the index.
5. Distance from target weights for each basket.
6. One research question to answer before next month.

Walnut's dashboard covers the first four lines on its own; the assistant answers the last two in a sentence each.

Answers about portfolio review mix several measures. Here is what each one means and whether Walnut reports it, so you know which questions to bring to a different tool.

Say you own a 15-stock portfolio built around three themes: AI infrastructure, aging demographics, and clean energy. You connect Walnut and ask:

“Break down my portfolio by theme and show me which one is contributing most to returns over the past year.”

The AI reads your holdings through Walnut, pulls each position's one-year return, weights it, and gives you an attribution table. You discover that AI infrastructure is 60% of the portfolio but 85% of gains, meaning a single theme is doing most of the work. A broker dashboard will not show you that split.

• Analyzing positions the AI cannot actually see. Pasting a few tickers into a chatbot falls short of portfolio analysis. If the tool cannot read your real weights, its answer is a guess.
• Trusting AI math on totals. Language models are weak at arithmetic. Prefer tools that compute concentration and returns with code and let the AI explain the result.
• Reading a price return as profit. A stock up 20% this year may still be below what you paid. Without cost basis, a one-year return says how the holding did, not how you did.
• Confusing a stock score with portfolio analysis. Danelfin or TipRanks rate individual names; they do not tell you whether your overall allocation is sound.
• Treating AI output as a prediction. Use it for synthesis and discipline, not crystal-ball calls. No tool can promise it beats the market.