What ChatGPT Is Genuinely Good At
ChatGPT is a language model, not a calculator or a market-data feed. Once you accept that, it becomes a genuinely useful thinking partner for your portfolio. It is at its best when the task is about words and judgement rather than precise numbers:
- Narrative analysis. Paste your holdings and it will describe, in plain English, what kind of investor your portfolio implies you are — growth-tilted, income-focused, over-exposed to one theme.
- Allocation critique. It is quick to spot that 60% of your money sits in one sector, or that you own five funds that all track roughly the same index.
- Explaining concepts. Ask it what "duration" means for your bond, or why an accumulating ETF is taxed differently, and you get a clear, patient answer tuned to your holdings.
- Generating questions. It is excellent at surfacing the questions you should be asking your own portfolio, which you can then answer with real data.
The one-line rule
Use ChatGPT to interpret numbers, never to compute them. If a figure has to be exact — a return, a gain, a tax bill — bring it in ready-made and ask ChatGPT to explain it, not to calculate it.
Step 1: Export Your Broker Data
ChatGPT can only analyze what you give it, so the first job is getting a clean snapshot out of your broker. Almost every broker offers a CSV or spreadsheet export — usually under a "Positions", "Holdings", "Statements" or "Reports" menu.
- Open your broker's Positions or Portfolio page and look for an Export, Download or CSV button.
- Prefer a holdings/positions export (what you own now) for allocation analysis, or a transactions export (every buy and sell) if you want to discuss history.
- Open the file in a spreadsheet so you can see the columns before you hand anything to an AI.
If you're not sure where your broker hides its export, our broker capabilities directory shows which brokers export, import or sync, with per-broker walkthroughs.
Step 2: Clean and Structure It
A raw broker CSV is noisy — 30 columns, internal codes, settlement dates, sub-accounts. Pasting all of it wastes the model's context and invites mistakes. Trim it to the few columns that matter and give ChatGPT a tidy table instead.
For an allocation review, you usually only need:
Ticker | Name | Asset class | Weight % AAPL | Apple Inc. | Equity | 14 MSFT | Microsoft Corp. | Equity | 11 VWCE | Vanguard All-World| ETF | 30 IEAC | iShares Corp Bond | Bond ETF | 20 BTC | Bitcoin | Crypto | 5 ...
Weights, not just values
If you want an allocation critique, add a weight % column yourself. ChatGPT will happily "compute" weights from your euro values, but on a long list those percentages routinely don't add to 100. Do the division in your spreadsheet and hand it the answer.
Step 3: Prompts That Actually Work
The difference between "analyze my portfolio" and a genuinely useful answer is structure. Give the model a role, the data, and one specific question. Here are six templates worth keeping — paste your cleaned table where indicated.
You are a portfolio analyst. Here are my holdings with weights: [PASTE YOUR CLEAN TABLE] Critique this allocation for a long-term investor with a moderate risk tolerance. Cover: overall balance across asset classes, geographic and sector concentration, and any overlap between funds. Give me the three most important observations, most important first. Do not invent prices or returns — work only from the weights above.
Using the same holdings, act as a risk manager. Identify where I am most concentrated — by single position, by sector, by country and by currency. For each concentration, explain in one sentence what would have to happen for it to hurt me, and roughly how much of the portfolio is exposed. Flag anything above 20% of the total.
Look at the funds and ETFs in my holdings. Which of them are likely to hold many of the same underlying companies (for example two world-equity or S&P 500 trackers)? Explain where I think I'm diversified but am really doubling up. Suggest what a genuinely diversified version of this set might look like, without recommending specific products to buy.
For each holding below, give me one plain-English sentence on what it is and what role it plays in a portfolio (growth, income, ballast, speculative). Then tell me, in two sentences, what story this overall mix tells about the kind of investor I am. [PASTE YOUR CLEAN TABLE]
Here is my current allocation by asset class: Equity 70% | Bonds 15% | ETF (world) 10% | Crypto 5% I want to move toward Equity 55% | Bonds 30% | ETF 10% | Crypto 5%. Describe the direction and rough size of the trades that shift me there, and the trade-offs of getting there gradually versus in one step. Talk in percentages only — do not calculate share quantities or prices.
Given these holdings, which are the likely income generators (dividend payers, coupon-paying bonds) and which are pure growth or speculative? Group them, and explain what a heavier tilt toward the income group would change about my portfolio's behaviour. I'll supply actual yields myself — don't guess specific dividend numbers.
Why every prompt says "don't invent numbers"
Left to its own devices, ChatGPT will fill gaps with plausible-looking figures. Explicitly forbidding it from inventing prices, returns and yields is the single most effective way to keep an analysis honest. Supply any hard number yourself.
Where ChatGPT Breaks (Read This Before You Trust It)
This is the section most tutorials skip. ChatGPT fails in specific, predictable ways on portfolio data — and because its wrong answers are fluent and confident, they're easy to believe. Here is what actually goes wrong.
| What breaks | Why | What it looks like |
|---|---|---|
| No live prices | The model answers from training data with a cutoff; it doesn't know today's quote | Confident valuations built on months-old or invented prices |
| Invented cost basis & returns | It approximates arithmetic over long lists and fills missing data | A "total return" that's internally inconsistent or simply wrong |
| No FIFO / tax-lot logic | It has no model of how lots are matched on a sale | Gains and tax figures that ignore which shares were sold |
| Context limits | Long transaction histories exceed what it holds accurately | It silently summarizes or drops rows, then answers as if complete |
| Currency & unit slips | No enforced schema for GBP vs GBX, or mixed-currency accounts | Values off by 100x, or two currencies added together |
1. It doesn't know today's prices
The base model has a training cutoff. Any current value or "you're up X%" it volunteers is built on stale or imagined prices unless you paste the price yourself. Even with a browsing tool enabled, quotes can be delayed, pulled from the wrong exchange, or misread. Never size a trade or a valuation on a price ChatGPT offers.
2. It invents and miscalculates cost basis and returns
Ask it to work out your gain across several buy lots and it will produce a number — often the wrong one. Language models approximate arithmetic and are prone to "filling in" missing values with plausible figures. On a portfolio of any size, treat every number it computes as a draft to be checked, not a result.
3. It has no FIFO or tax-lot awareness
Real capital-gains tax depends on which shares you sold — FIFO, LIFO or specific identification — and on rules like wash sales and the Irish 8-year deemed disposal. ChatGPT has no model of any of this. Cost-basis and tax figures it produces will quietly ignore the lot-matching that determines your actual bill.
4. It runs out of context on large histories
A decade of transactions across three brokers is thousands of rows. Past a certain size the model can no longer hold it all accurately; it summarizes or drops data without telling you, then answers as though it read everything. The bigger your history, the less you should trust a raw paste-and-analyze.
Confident and wrong is the dangerous combination
ChatGPT's mistakes don't come with error bars. A fabricated return reads exactly like a correct one. That's fine for a qualitative critique — and genuinely risky the moment you act on a number it produced.
Data Privacy: What You're Actually Sharing
Pasting your holdings into a chat window sends them to a third party. On consumer ChatGPT plans, conversations may be used to improve the models unless you turn training off under Settings → Data Controls. Business and enterprise plans don't train on your data by default, but most personal users aren't on those.
- Never paste account numbers or anything that identifies the account itself.
- If exact balances feel sensitive, share share counts and weights rather than euro or dollar totals — the analysis barely suffers.
- Turn off training in Data Controls if you're going to do this regularly.
Want the analysis without the guesswork?
AllInvestView's built-in AI assistant answers these same questions over your real, synced portfolio and live prices — no CSV exports, no hallucinated returns, and cost basis computed with proper tax-lot logic. Ask "what's my best performer this year?" or "how much dividend income did I earn?" and get a number that's actually correct.
Portfolio Trackers With Built-In AI
The honest conclusion is that ChatGPT and a portfolio tracker do different jobs. ChatGPT is a brilliant explainer working from whatever you paste; a tracker holds your actual positions, prices them live, and does the exact arithmetic — FIFO cost basis, multi-currency returns, tax by jurisdiction — correctly and repeatably.
The newest generation of trackers closes the gap by putting an AI assistant on top of your real data. Instead of copying a CSV into a chat, you ask the tracker itself. AllInvestView's assistant has full context of your holdings, performance and history, so you can ask in plain language — and the numbers behind the answer are the same audited figures that drive your tax report.
If you want to see how the two approaches compare head-to-head, read portfolio tracker vs ChatGPT, or browse the best AI portfolio trackers.
Frequently Asked Questions
Analysis you can act on
Track your whole portfolio for free, then ask the built-in AI assistant anything — over real holdings, live prices and correct cost basis.
Start free