How to Analyze Your Portfolio With ChatGPT (Safely)

You can analyze your portfolio with ChatGPT, and for some questions it is a very good second opinion. It explains what a company does, spots concentration you missed and finds holes in a plan. Where it fails is arithmetic on your real data: prices, returns, currency and dividends. The safe method is simple. Let a portfolio tracker calculate the numbers, give ChatGPT a summary of weights instead of your raw broker export, and ask it to reason, not to count.

You would not be the only one doing this. In eToro's August 2025 survey of 11,000 retail investors in 13 countries, 19% said they use AI tools to pick or change investments, up from 13% a year earlier. Most of them paste their holdings into a chat window and trust what comes back.

This guide shows where that goes wrong for European investors on Degiro, and a six-step workflow that keeps the useful part.

---

What ChatGPT portfolio analysis is good at

ChatGPT portfolio analysis means giving a large language model a description of your holdings and asking it questions about risk, allocation or strategy. The model does not see your broker account. It sees only the text or file you give it, and it fills every gap with what it learned in training.

That makes it strong at language and weak at accounting. A useful rule: if the answer is a sentence, ChatGPT can help. If the answer is a number about your own money, get it from somewhere else.

So is ChatGPT good for investing? As a thinking partner, yes. As a calculator for your own account, no. The rest of this guide is about keeping those two jobs apart.

Six ways ChatGPT gets your portfolio numbers wrong

1. It uses old or wrong prices

Without web search, the model answers from training data that is months old. With search turned on, it can look up a quote, but it has to pick the right one.

ASML trades in euros on Euronext Amsterdam and in dollars on Nasdaq. A model that grabs the wrong listing mixes two currencies in one total. London is worse: prices there are quoted in pence, not pounds, so a missed unit makes a position 100 times too large. Your Degiro export names the exchange in a short code (EAM for Amsterdam, NDQ for Nasdaq, ETR for Xetra). Most people delete that column before they paste. Keep it.

2. It reports total gain as if it were a yearly return

Say you put €1,000 into an ETF on the first day of every month in 2025, and on 31 December the position is worth €12,600. Ask a chatbot for your return and the common answer is 5%: €600 gained on €12,000 deposited.

That is a correct total gain. It is not a yearly return, because most of that money was invested for only a few months. Calculated as a money-weighted return (XIRR), the same result is about 9.4% a year, almost double. Neither number is wrong. They answer different questions, and a chatbot rarely tells you which one it picked. Our guide to time-weighted vs money-weighted return explains which number to use for what.

3. It misreads Degiro's CSV

Degiro's Transactions CSV writes numbers with a comma as the decimal separator, and puts them in quotes. A share price of 412.35 appears as "412,35". A model, or a script it writes for you, can read that comma as a column break or a thousands separator. That is how a €412 share becomes a €41,235 share.

The file is also incomplete for return questions. The Transactions export holds your buys and sells. Dividends, dividend tax, deposits, withdrawals and currency conversions sit in a separate Account statement. Upload only the transactions, ask for your total return, and the answer leaves out every dividend you ever received. It usually does not warn you.

ChatGPT can run Python on an uploaded file, which fixes the adding-up problem. It does not fix the input problem. Code that sums the wrong column gives a wrong answer with more decimals. Any AI portfolio analysis is only as good as the file under it.

4. It forgets that you think in euros

If you give a model dollar prices, it calculates a dollar return. Your account is in euros.

A US stock that rises 10% in dollars while the dollar falls 8% against the euro leaves you up about 1% in euros, not 10%. Degiro also charges a 0.25% AutoFX fee on each conversion, which the model only knows about if that column is in your data. We showed what the euro's rise did to European portfolios this year, with the numbers.

5. It counts dividends before tax

Ask ChatGPT what a 3% yield pays on €20,000 and it will say €600. That is the gross figure.

US dividends paid to most EU investors lose 15% at source with a W-8BEN form on file, and 30% without one. Dutch companies withhold 15% as well. So €600 of US dividends arrives as about €510, before your own country taxes anything. For an income plan, that gap compounds every year. Our guide to dividend withholding tax in Europe lists the rates by country.

6. It cannot see inside your ETFs

To a chatbot, IWDA or VWCE is one line. It knows roughly what the index holds, from training data that can be a year old.

That gap matters more in 2026 than before, because the AI rally made the largest index names even larger. On 31 August 2026, seven US companies were 23.6% of the MSCI World. Hold a world ETF, an S&P 500 ETF and a few Nvidia shares, and your real exposure to AI stocks is bigger than any single line shows. We walk through the full calculation in ETF overlap: how much Mag 7 is in your portfolio.

If you want ChatGPT's help here, paste the top-10 holdings from each fund's latest factsheet. Then the model works with current numbers instead of memories.

A safe way to analyze your portfolio with ChatGPT

Degiro has no public API, so no AI assistant can connect to your account. Copy and paste is the only route. That has one advantage: you decide exactly what the model sees.

1. Export your data from Degiro. On the web platform, open the Inbox on the left of the screen and export the Transactions CSV and the Account statement. The default range is one month, so set it to your full history.
2. Let a tracker do the arithmetic. Import the CSV into a portfolio tracker that reads Degiro's format. It rebuilds your positions, values them at current prices, converts them to euros and adds your dividends. A Degiro portfolio tracker like Zune.Money does this step in about three minutes, and our Degiro portfolio analyzer guide covers what to check once the import is done.
3. Build a short summary table. One row per holding: name, ticker and exchange, weight in percent, sector, country and currency. Add two lines below it: total return and yearly dividend income.
4. Remove anything that identifies you. No name, account number or IBAN. If you upload a raw Degiro file, delete the Order ID column first. In ChatGPT, use a Temporary Chat, or turn off "Improve the model for everyone" under Settings, Data Controls. OpenAI's data controls page explains both. Temporary chats are not used for training, but OpenAI can keep them for up to 30 days.
5. Ask one question per prompt. "Analyze my portfolio" gets a generic essay. A narrow question gets an answer you can use.
6. Check every number that comes back. If ChatGPT states a weight, a return or a yield, compare it with your tracker. When they disagree, the tracker wins. Ask the model to show its working so you can see where it slipped.

Use percentages, not euro amounts. Weights carry everything the model needs for allocation questions, and they say nothing about how much money you have. A table where ASML is 8.2% tells ChatGPT as much as one where ASML is €4,100, and it tells the chat history much less.

ChatGPT investing prompts that work

Each prompt below starts from the summary table in step 3. Paste the table where it says so.

Find concentration.

Here is my portfolio as weights: [table]. Which single companies, sectors,
countries and currencies are more than 15% of it? Count a company once even if
it appears in two rows. Show your working.

Stress-test the mix.

Using only these weights: [table]. Describe what happens to this portfolio if
US technology stocks fall 30% and the dollar falls 10% against the euro at the
same time. Which holdings are hit twice? Do not look up or invent prices.

This is where ChatGPT is at its best. It is reasoning about a scenario you defined, with numbers you gave it.

Test your own reasons.

I hold ASML at 8% of my portfolio because I expect chipmakers to keep buying
EUV machines through 2028. What would have to be true for that reason to be
wrong? List four things I can check in the company's annual report.

Plan new contributions without selling.

My target is 70% world equity, 20% European equity and 10% cash. Current
weights: [table]. I add €500 a month and do not want to sell anything. How
should I split the next six contributions to move toward the target? Show the
arithmetic step by step.

Check the arithmetic yourself. Asking for steps makes errors visible, not impossible.

Understand your own numbers.

My tracker shows a total return of 14% since 2023, but a money-weighted return
of 6% a year. Explain in plain words how both numbers can be correct.

Some prompts are better skipped. "Which stock should I buy next?", "What will ASML trade at in December?" and "Should I sell everything?" ask the model to predict or to advise. It can do neither for you, and it has no duty to get it right.

What ESMA says about AI investment advice

In 2025, the European Securities and Markets Authority published an investor warning titled Using Artificial Intelligence for Investing: What you should consider. It is short, and every Degiro user who pastes a portfolio into a chatbot should read it. The main points:

• Public online AI tools "are not authorised nor supervised by financial regulators."
• They "do not hold any obligation to act in your best interest."
• Their output "can be based on outdated, incorrect or incomplete information."
• Do not share personal data such as your name, age, marital status or financial situation with them.
• If something goes wrong, you may have no access to a financial ombudsman.

The full text is in ESMA's AI warning for investors. In the EU, personal investment advice is a regulated activity under MiFID II, and a general chatbot is not authorised to provide it. That does not make ChatGPT useless. It makes it one source among several, which is how the workflow above uses it.

The split that works

ChatGPT is a good thinking partner and a poor bookkeeper. Give each job to the tool that does it well and most of the risk goes away. The tracker keeps the ledger. The chatbot asks the questions you forgot to ask.

Zune.Money does the ledger part for Degiro investors. Import your CSV and you see every holding with its weight, total return with dividends included, and allocation by sector and country, which is everything step 3 needs on one screen. It has no chatbot built in and it does not calculate TWR or XIRR, so for those two numbers use the methods in our performance guide. New accounts get 7 days of full access with no card, and the free plan keeps up to 10 holdings after that. Start with your Degiro CSV and give ChatGPT numbers worth reasoning about.

---

This article is for informational purposes only and is not financial advice. Tax treatment of dividends and capital gains varies by country, so consult a tax advisor for your situation. Past performance does not guarantee future results.