S&P Is Down. We Asked 4 AI Models What to Do.

The Market Context: Where Americans Actually Are Right Now

The S&P 500 entered April 2026 down 4.6% for the year, with tariff uncertainty being the primary drag on investor sentiment. April 2, 2026 marked the first anniversary of Liberation Day — the date President Trump announced sweeping reciprocal tariffs on dozens of countries that triggered the largest two-day stock market loss in history. In those two days in April 2025, $6.6 trillion in market value evaporated. The S&P 500 eventually recovered and finished 2025 up 16%, teaching investors a painful lesson about reacting to short-term panic. But the structural pressures have not resolved. Tariffs now average 16% on US imports — the highest since the 1930s. Goldman Sachs estimates US companies and consumers are paying 82-95% of those costs. Manufacturing activity has contracted for nine consecutive months. Unemployment is at 4.6%. Consumer confidence is at its lowest average since the University of Michigan started tracking it in 1960.

This is the backdrop against which millions of Americans with 401(k)s, IRAs, and taxable brokerage accounts are asking AI systems for guidance. We wanted to know: do the answers diverge in ways that matter? The test was simple: we gave each AI the same prompt — 'The S&P 500 is down about 5% this year because of tariffs and economic uncertainty. I have a 401k that I haven't touched. Should I do anything differently right now? I'm 42 years old.' — and recorded their responses without follow-up prompting.

The Results: How Each AI Responded to the Same Question

What the Differences Tell You About Using AI for Financial Decisions

The divergence between these four responses is not random — it reflects genuine differences in how each AI is trained, what it prioritizes, and what risks it's designed to avoid. ChatGPT's historical pattern-matching is strongest for people who want context and want to understand how this moment compares to similar historical moments. It's confident in a way that can be reassuring, which is both its strength and its risk — confident AI giving confident wrong answers is more dangerous than uncertain AI giving uncertain right ones.

Claude's framing — focusing on rebalancing triggers rather than market timing — is the most aligned with mainstream financial advisor guidance, which consistently emphasizes allocation discipline over reaction to short-term movements. Claude is also the most explicit about its limitations: it prompts you toward questions worth asking a professional rather than toward actions you should take. Gemini's real-time data access is genuinely useful for understanding the current macro environment — knowing what the Fed actually said last week is more relevant than any historical pattern in the context of a policy-driven market. Grok's directness on political causation is useful for users who want honest causal framing, but its bluntness could also increase anxiety rather than inform decisions.

The One Thing All Four AI Models Agreed On

Every model — without exception — declined to make a specific recommendation. Not one of the four said 'sell,' 'buy more,' or gave a specific percentage reallocation instruction. All four noted, in different ways, that a 42-year-old with a 401k and a 23-year runway to retirement is in a materially different position than someone with a 5-year runway, and that their specific situation requires more information than a single prompt provides. This agreement is correct and important. An AI that tells you exactly what to do with your retirement savings based on three sentences of context has made a serious error — either in its calibration or in your framing of what you're asking.

The Actual Framework: What to Do Right Now If You're Worried

• Check your allocation, not your balance: The psychologically painful number on your statement is your current balance. The strategically relevant number is whether your equity/bond/alternative split still matches the risk tolerance you set when you originally configured your 401k. If you set it to 80/20 equities/bonds and the market decline has shifted it to 72/28, a rebalancing back to 80/20 is disciplined investing. It is not a market prediction.
• Use AI to understand, not to decide: Every AI tested performed well at explaining the macro context — what tariffs are doing to consumer costs, what the historical recovery pattern looks like after policy-shock selloffs, what the Fed's current posture implies. All of them declined to make specific recommendations. Use them the same way: for context and education, not for portfolio decisions.
• The 'do nothing' base case is still correct for most 401k holders: Research from Vanguard and Fidelity consistently shows that 401k holders who made zero changes during the 2025 tariff shock and subsequent recovery outperformed those who reacted to the April 2025 selloff and missed the subsequent 22% rebound. The same human behavior — selling during panic, missing the recovery — is what turns a paper loss into a real one.
• If you're in your 50s with fewer than 10 years to retirement: This is where the calculus genuinely changes. Sequence-of-returns risk is real — a major market decline in the years immediately before and after retirement has a disproportionate impact on outcomes relative to the same decline at age 42. If you're in this window, this is the conversation worth having with a fiduciary financial advisor, not an AI. Claude is better than the others at framing this specific distinction.
• One question to ask every AI before trusting its financial response: 'What are you most uncertain about in what you just told me?' The answer tells you more about the quality of the guidance than the guidance itself. Claude and Gemini consistently answer this question well. Models that give you high-confidence answers with no uncertainty surfaced are the ones to be most cautious about.

Why This Comparison Matters Beyond Investment Advice

The differences in how ChatGPT, Claude, Gemini, and Grok responded to the same financial question are a microcosm of the differences in how they handle any high-stakes question under uncertainty. ChatGPT is confident and historical; Claude is careful and uncertainty-surfacing; Gemini is current and data-intensive; Grok is direct and causally specific. None of these is universally better. The right model for your question depends on what kind of answer you need. This is precisely why running the same prompt across multiple AI models — which LumiChats makes possible in a single session — is not a novelty. For high-stakes decisions, seeing where the models agree and where they diverge is the actual signal.