Around the world, kitchens and home offices are witnessing an odd new ritual. “Can you review this portfolio and tell me what I should change?” is typed by someone who opens a laptop, uploads a screenshot of their brokerage account (the kind of document you’d normally show only a spouse or a paid financial planner). A few seconds later, a well-organized, frequently excellent critique is provided by a chatbot. Whether ChatGPT can accomplish this is not the question that people are secretly debating. It’s about how much of its output to genuinely trust and whether it should be doing it.
The number of experiments has rapidly increased. GOBankingRates requested in April that ChatGPT create a diversified portfolio for a forty-year-old investor who would have twenty years to retire. An almost textbook-like recommendation was returned by the AI: 50% in the Vanguard Total Stock Market ETF (VTI), 20% in foreign stocks through VXUS, 20% in bonds through AGG or BND, 5% in real estate through VNQ, and 5% in gold through GLD. After reviewing the selections, Thomas Brock, a CFA and CPA at Annuity.org, described them as “sound” but pointed out that they were “a bit light on the growth front” for someone with such a long time horizon. He claimed that a globally diversified stock portfolio already carries embedded real estate exposure, so he increased the equity weight to 80%, increased international exposure to better match global market capitalization, and completely removed the real estate slice.
| Detail | Information |
|---|---|
| Tool | ChatGPT / OpenAI GPT-4 & GPT-5 class models |
| Most Common Use Cases | Portfolio review, asset allocation, stock screening, macro summaries |
| Popular ChatGPT Custom GPT | “Portfolio Analysis” |
| Typical Use-Case Demographic | Self-directed retail investors, DIY long-term savers |
| Common ETFs Recommended | VTI, VXUS, BND, AGG, VNQ, GLD |
| Typical AI-Recommended Allocation | 70% stocks, 20% bonds, 10% alternatives |
| Expert Consensus on AI Portfolios | Reasonable, but often too conservative for long time horizons |
| CFA Reviewer Example | Thomas Brock, Annuity.org |
| Brock’s Key Critique | “A bit light on the growth front” |
| Brock’s Suggested Revision | Increase equities to 80%, eliminate real estate, raise international stock weight |
| Real-World Experiment | Thomas Smith’s $500 ChatGPT portfolio (Palantir, AppLovin, MicroStrategy, Agios Pharma, Hut 8) |
| Result (3 months in) | Volatile — huge rise, then steep drop |
| Peer-Reviewed Study | Romanko et al. (2023), cited 83+ times |
| Academic Finding | ChatGPT is effective at stock selection, weaker at weighting |
| Common Retail Workflow | Upload IBKR statement → ask GPT for allocation review |
| Most Useful GPT Function | Structure, discipline, emotional offsetting |
| Biggest Limitation | Hallucinated figures (e.g., wrong IRA contribution limits) |
| Recommended Approach | Use AI as sanity check, not decision maker |
| Regulatory Status | Not a registered financial advisor per SEC |
It’s a telling pattern. The AI doesn’t give poor advice. It generates general guidance. This is a significant distinction because, for the average investor, consistent application of generic advice likely outperforms custom advice that is never followed. There’s a reason why complex trading strategies aren’t the most common ChatGPT use cases in personal finance. These include things like “compare my current allocation to a global 60/40 benchmark,” “summarize the macro news this month,” and “which of these ETFs overlap in holdings.”” Those are dull inquiries. Additionally, the majority of retail investors genuinely need answers to these questions.
Attention-grabbing experiments typically go in the opposite direction, with high drama and stakes. Journalist Thomas Smith asked ChatGPT to create an aggressive six-month growth portfolio after giving it $500 in a real Robinhood account in September 2025. Palantir, AppLovin, MicroStrategy, Agios Pharmaceuticals, and Hut 8 were selected by the AI. Smith described what came next as a “wild ride.” He regretted not investing more after the portfolio reached highs and then crashed so badly that he almost gave up. After three months, the performance was so erratic that any disciplined investor’s resolve would have been put to the test. The aspect of these experiments that is typically left out of the headlines is Smith’s honest framing: stocks chosen by chatbots are extremely volatile, and real money can be lost.
The more deliberate form of AI-assisted investing, which is genuinely yielding outcomes, has a different appearance. Over the past year, a Reddit user in r/ArtificialIntelligence detailed their workflow, which included subscribing to a few macro newsletters, setting up a ChatGPT project with personal rules and investing books, uploading a monthly snapshot of their IBKR portfolio, and asking the bot to make minor tactical recommendations within stringent parameters. only ETFs. Not a single stock. Not a single trade. They claim to have outperformed the S&P 500 by about 2% with fewer emotional decisions, but the sample size is too small to have any statistical significance. However, the explanation of the AI’s actual actions—enforcing structure, identifying threats, and lowering panic trades—was significant. Alpha isn’t that. That’s behavioral coaching, which is probably what the majority of investors require anyhow.
As this develops, it seems like ChatGPT isn’t really taking the place of financial advisors in personal investing. It fills the gap between paying for an advisor and not having one. A well-prompted AI session provides a structured second opinion that was previously unattainable for the person with $50,000 in a brokerage account who cannot afford to hire a fiduciary planner. Oleksiy Romanko’s peer-reviewed 2023 paper, which has been cited over 80 times, discovered that ChatGPT performed surprisingly well at choosing individual stocks but poorly at allocating ideal portfolio weights. It’s a helpful distinction. Stock recommendations are not necessary for the majority of retail investors. They require assistance with the weights.
The warnings are valid and should be made clear. ChatGPT has numerical hallucinations. This was directly documented by GOBankingRates; during a portfolio review, the AI misstated the IRA contribution catch-up limit; only a reader who was aware of the regulations would have noticed it. Unless granted access, it is unable to view real-time market data. It is unaware of your estate plan, insurance coverage, tax status, or the emotional background that influences your response to a 20 percent market decline. It’s not a fiduciary. It won’t ever be. As this develops, it’s difficult to avoid the conclusion that, in 2026, AI portfolio reviews are a good place to start, a risky place to end, and an oddly democratizing technology that sits awkwardly in the middle.
