Your Verdict on My ChatGPT Investor

Indeed, my preceding article stirred up quite a response. Not in the same league as the document from Citrini Research, I’m pleased to report, which foretold a future where professional workers, like automatons, consume infants instead of opting for food delivery services. I merely posited that artificial intelligence held an advantage over financial consultants.

For this current week, I had devised some additional prompts in my pursuit of an optimal investment portfolio. ChatGPT had demonstrated an impressive beginning, but I had a multitude of subsequent inquiries. Before delving into those, however, I must address the hundreds of electronic messages I received.

Five principal concerns emerged most frequently. Firstly, I was astonished by the number of readers who confessed to already utilizing large language models for investment guidance. If that is the situation, I pondered, why bother with my weekly ramblings?

Then came the anticipated retort from financial advisers. Artificial intelligence, they claimed, would never supplant the “personal touch.” It lacks the capacity to perform the essential “behind-the-scenes” due diligence. How, they questioned, would these models behave “during a crisis?”

Advisers, naturally, would articulate such points, wouldn’t they? And were I in their shoes, I would underscore the benefit of lengthy, convivial lunches, an experience no AI model can replicate. We shall observe what transpires. More intriguing were the subsequent three apprehensions expressed by numerous individuals.

Many of you underscored that it’s one thing for an ex-investment specialist to provide the prompts. But how would the typical retail investor discern what to inquire about? My response to this is that ChatGPT would have formulated far more astute questions than I could have.

In practical terms, even the least financially savvy individual can input: “Listen, I possess no understanding of finance, but I wish to embark on cruises and similar activities until my demise. Can you assist?” At that juncture, an AI agent assumes control—scrutinizing bank statements, risk appetite through surveys, tax standing, familial circumstances, health information, and so forth.

The appropriate queries would be posed. Far more intelligently than yours. Far more intelligently than mine. But a snag exists—as many of you were quick to discern. How does AI navigate the reality that regulators frequently prohibit the acquisition of numerous investment vehicles without endorsement from human advisers?

Indeed, certain facets of investing operate as an exclusive domain—and in my estimation, this ought to change now that AI decisively outperforms all other methods. Nevertheless, employing ChatGPT, Claude, or Gemini to construct a foundational portfolio of equities and bonds represents a monumental stride for individual investors. Let us not be overly ambitious just yet.

This leads me to the most captivating query hundreds of you sent via email: if I furnish disparate LLMs with an identical prompt, will they yield the same counsel? There are numerous approaches to structuring an investment strategy, but who would entrust their capital to AI if the outcomes diverge significantly?

Before I could even power on my workstation, Dr. Bruce Lloyd, emeritus professor of strategic management at London South Bank University, had processed my initial prompt through ChatGPT, Claude, Gemini, Grok, Copilot, Perplexity, Qwen, and DeepSeek.

The findings? Akin to their human predecessors, each model adopted its unique methodology for constructing portfolios. Copilot and Gemini, for instance, exhibited a greater quantitative and risk-averse orientation. Grok and ChatGPT appeared to favor straightforward, economical execution.

Qwen, meanwhile, displayed considerable awareness of foreign exchange. Perplexity concentrated on hedging strategies and meticulously examined everything through a UK perspective, which I appreciated. DeepSeek is a staunch proponent of diversification and classic portfolio theory. Fair enough to each.

More specifically, Gemini is deeply invested in style biases and the utilization of factors such as small-cap and growth stocks. There is also a substantial divergence across the board concerning the role of alternatives and real assets—Copilot posits I should allocate one-fifth of my portfolio to them. At the opposing end of the spectrum, Perplexity advocates for a 10 percent cash holding for “psychological reassurance.”

Overall, however, each of the models ultimately converged on approximately 60 to 65 percent in equities, with Copilot and Perplexity standing as outliers at 55 percent. DeepSeek and Claude were content to elaborate further on regional allocations, while Gemini and Copilot underscored the importance of holding quality and low-volatility shares.

Regarding fixed-income assets, Grok and Qwen proposed as high as a 35 percent weighting, with a sizable 15 percent allocated to corporate credit. Claude stated I should only possess 20 percent in conventional fixed-income securities. The former pair also concurred that hedging back into pounds is imperative.

These do not represent enormous discrepancies in asset allocation. Nor are there any outlandish proposals (60 percent in emerging market debt!) or absurd recommendations like liquidating all my equity funds last year. Everything was coherent and rationally substantiated.

It is reassuring that all the models comprehended the fundamentals. That diversification is paramount for optimizing risk-adjusted returns. The efficacy of systematic rebalancing. Tax efficiency. Also, what I recently discussed: that determining your entry point is crucial—avoiding dollar-cost erosion.

Bless his heart, Dr. Lloyd then tasked ChatGPT with formulating a composite “best of all possible worlds” portfolio, amalgamating the outputs of all the models—from the intellectually rigorous precision of Copilot and Gemini to the pragmatic guidance of ChatGPT and Grok, given my residence in Britain.

This does not constitute investment advice, but here is the solution: Vanguard FTSE All-World ETF (45 percent); Vanguard FTSE UK All Share ETF (10 percent); iShares MSCI World Quality Factor (10 percent); iShares Core UK Gilts UCITS ETF plus iShares Global Aggregate Bond GBP-Hedged (25 percent); iShares UK Property ETF plus HICL Infrastructure (5 percent); iShares Physical Gold ETC (5 percent); cash (5 percent).

Would I ever implement this particular portfolio? Perhaps. But reflecting upon it revealed to me a defect in AI-driven investing. There is no individual to hold accountable when things go awry.

The author is a former portfolio manager. Email: stuart.kirk@ft.com