When asked to select the best FTSE 250 stock to buy, ChatGPT recently pointed to Serco (LSE:SRP), while separate queries across mid-cap indexes have yielded suggestions like Balfour Beatty and shipbroker Clarkson.
Why Large Language Models Fall Short as Autonomous Stock Pickers
Let’s be clear: leaning on an LLM to populate a Stocks and Shares ISA is an architectural mismatch. By its own admission, ChatGPT isn’t a stock picker, and its outputs demand a heavy dose of skepticism. Models are notoriously prone to ingesting stale web data—historical tickers, outdated valuations, and expired dividend yields—and presenting them as current metrics.
Yet, treat the technology strictly as a rapid-fire brainstorming tool, and its utility shifts. For investors drowning in the sheer volume of the FTSE 250—an index boasting more than double the components of the FTSE 100—an AI companion can rapidly surface overlooked mid-cap names.
Deconstructing the AI’s Mid-Cap Recommendations
The chatbot’s suggestions across recent mid-cap queries highlight a mixed bag of cyclical plays and infrastructure giants. For instance, when pointed toward infrastructure, the AI singled out Balfour Beatty (LSE:BBY), citing its long order book and history of shareholder returns via dividends and share buybacks.
According to market reports, the Balfour Beatty share price enjoyed a strong run, surging 60% over a 12-month window and 120% over two years. The company tracked toward full-year expectations backed by a 20% jump in its order book, which stood at £18.4bn. With projected earnings growth of 5% and monthly net cash touching £1.2bn, the stock traded on a price-to-earnings ratio of 16.5. While construction remains a bumpy sector prone to project overruns, the underlying data matched the AI’s bullish output.
Shifting focus to maritime services, the chatbot highlighted shipbroker Clarkson (LSE:CKN). Described as a global powerhouse operating across 67 offices in 25 countries, Clarkson navigated a challenging global trading environment that saw revenue increase 3.4% to £661.4m, alongside an underlying pre-tax profit of £115.3m. Furthermore, the firm hiked its annual dividend for the 22nd consecutive year, supported by significant insider buying where two directors purchased over £200k worth of shares.
However, the AI missed crucial macro-level volatility. As noted in Clarkson’s H1 report, the company was forced to navigate a highly complex global environment amid shifting economic conditions and evolving trade dynamics—nuances that static LLM training parameters routinely gloss over.
The Danger Zones: High Yields and Cyclical Traps
Not all AI suggestions hold up under technical scrutiny. When tasked with finding income or turnaround candidates, the model has previously flagged oil and gas producer Harbour Energy, highlighting an eye-popping 10% trailing yield. Crucially, the system omitted the massive capital loss potential inherent in cyclical commodity stocks, failing to contextualize why the Harbour share price had dropped 15% over a single year and 55% over five.
Similarly, the model pointed toward TwentyFour Income Fund for its 9.68% trailing yield, focusing on higher-yielding, less liquid credit assets. While the yield sounds attractive, the underlying share price sat roughly where it did a decade ago, proving that raw yield metrics without capital appreciation context are a classic value trap.
The 30-Second Verdict on AI-Assisted Investing
The robot can talk, but human investors still have to walk the floor.
