And what their answers tell you.
In mid-December 2025, I asked five AI systems – ChatGPT, Claude, DeepSeek, Grok, Perplexity – the same deliberately unfair question: “given current market condition and historic patterns, what is likely to be the highest returning asset class available to US investors in 2026?” The question is unfair because nobody can know that until 2027, and because it ignores all of the important stuff, information about the investor’s horizon, needs, and temperament, which is vastly more important than raw returns information.
Every system rushed to answer rather than challenge the (dangerous, stupid) premise. They marshaled citations, built confidence through length, and buried disclaimers after pages of analysis. DeepSeek: 1,200 words of scenario planning. Grok: comparative tables. ChatGPT: implementation steps.
The consensus answer? All confidently predicted U.S. equities would lead, particularly AI-driven tech. The runner-up consensus favored international and emerging market equities (citing extreme valuation discounts to U.S. markets), with private equity and long-duration Treasury bonds appearing as high-conviction contrarian plays depending on whether the economy achieved a soft landing or tipped into recession.
Only Claude in incognito mode – stripped of context about who was asking – pushed back meaningfully: “Here’s what concerns me about your question: asking ‘what is likely to be the highest returning asset class’ suggests a search for a single answer when the evidence points to profound uncertainty.” Then it asked the questions that actually matter: “What’s your situation? Time horizon? Current allocation? Risk tolerance?”
But you would note the hedge words that followed every forecast: “if earnings materialize,” “assuming soft landing,” “barring geopolitical shock.” These weren’t uncertainty disclaimers; they were reminders that the confident frameworks were built on unknowable conditionals.
This is cognitive offloading weaponized. An advisor doing this face-to-face reads body language, probes anxiety, and discovers what’s really being asked. The algorithm just answers – fluently manufacturing certainty from insufficient information. It doesn’t know it’s performing a lobotomy. It’s just following the affordances of the technology: you ask, it responds, everyone’s satisfied with having done something.
These concerns are not hypothetical. They will become manifest, both as you sit across from your financial planner and when your portfolio manager sits down to select equities. Surveys of wealth‑management firms show that roughly three‑quarters now report using some form of AI, up from about half just two years ago, with independent RIAs leading the way and using these tools for meeting prep, note‑taking, document review, marketing, and workflow automation (Grant Thornton, Global survey: AI is transforming asset management, 12/2025). At the institutional level, a global Mercer survey (2024 Global Manager Survey) finds that about nine in ten investment managers are either already integrating AI into their investment strategies and asset‑class research or actively planning to do so, primarily to expand data sets, speed analysis, and generate trade ideas, with AI outputs increasingly informing, but in some cases proposing, specific equity and asset‑allocation decisions.
A slightly more troubling development is the willingness to accept AI recommendations with, let’s call it, minimal reflection. The CFP Board’s 2023 Consumer Sentiment Survey (shows that 31% of investors feel comfortable implementing financial‑planning advice from a generative‑AI tool without verifying it, and that rises to 52% if a financial planner verifies the AI’s recommendations and a 2025 J.D. Power survey finds that 51% of U.S. consumers already turn to AI for financial advice or information, with another 27% considering it.
In other words, AI is now deeply embedded in both the advisory conversation and the portfolio construction engine behind it, often without being named explicitly, which means your financial life can be quietly shaped by systems you never see and questions you never knew to ask.
Three ways to avoid portfolio lobotomy in 2026:
- Treat AI as a research assistant, not an oracle. Use it to gather data (“which ETFs with ‘quality’ in their name or investment strategy have the highest Sharpe ratio over the past 18 months?”), check your thinking (“if US interest rates are likelier to decline in 2026 than to rise, does that imply that non-US bonds of a similar issue quality should outperform US ones?”), find counterarguments (“what are the three best reasons to avoid funds that invest in Nvidia?”). Never for decisions requiring judgment about your specific circumstances or temperament.
- If an AI gives you a confident answer to an unknowable question, run away! The fluency is the danger. AIs are designed to be helpful, which generally means they are designed to be cheerfully confident and affirming while making shit up. Wisdom often sounds uncertain because it accounts for what can’t be known.
- Judge your advisor by the questions they ask, not the answers they provide. Anyone – human or AI – who prescribes without diagnosing is practicing malpractice. AI is hidden just an inch below the surface of many planning practices; whether “model portfolios” or “optimized allocations,” many advisers – especially young folks at big firms – are simply expected to sell the packages and to sell them quickly, which means glossing over complicated questions about one’s life goals and fears and hopes and family history. The hard work isn’t generating recommendations; it’s understanding what problem you’re actually trying to solve.
