Asked to turn EUR 500 into as much money as possible in one month, using three stocks only, no ETFs, currencies, gold or other protective assets, all three chatbots independently put ASML on the final shortlist - and ASML fell more than 12% over the period. Every portfolio ended in the red: ChatGPT minus 2.49% (a EUR 12.45 loss on Nvidia, Rheinmetall and Coinbase), Claude minus 8.79% (a EUR 43.96 loss, its Super Micro Computer pick down more than a quarter from USD 30.71 to USD 22.05) and DeepSeek minus 8.55% (a EUR 42.74 loss, every one of its three positions a loser - ASML, Celestica and Nu Holdings). Only Coinbase gained, about 9%. When asked what went wrong, Claude blamed political risk and specifically the US tariffs announced on 2 April - an event that happened a year earlier, a clear date and timeline hallucination. DeepSeek largely blamed the person who set the task, and cited ASML export restrictions announced in March that had not happened that year. ChatGPT's explanation was that it had supplied long-term ideas for a one-month task and should have made at least one bet against the market.
1 Answer
Expert: George Timoshin, Junior Strategist at Freedom Broker All three portfolios lost money, and the losses were not bad luck: every model independently converged on the same handful of themes. ASML appeared in all three shortlists; Nvidia and Coinbase appeared in two each; the names clustered in AI infrastructure, defence and crypto. George Timoshin of Freedom says the experiment should be read with caution, and his structural critique is the part worth keeping: the models mixed two different levels of recommendation, opening with classical allocation by region and sector and then delivering a concentrated thematic basket of a few stocks that no longer matched that structure. Strategic allocation was de facto transformed into a speculative basket in which idiosyncratic single-company risk dominates and diversification effectively disappears. He also notes the picks reflect current market consensus narrative rather than a search for alpha, which makes the exercise a demonstration of what the market currently treats as volatile rather than an investment process. The two clearest failures are qualitative, not just arithmetic. Claude attributed its loss to tariffs announced on 2 April - a year out of date - and the experiment's own scorecard deducted a point for the hallucination: not knowing what year it is living in is alarming. DeepSeek produced the third-worst return and spent its explanation blaming the person who set the task, while also citing ASML export restrictions that had not been announced in the period in question. ChatGPT gave the most honest readout, but it was a readout of its own error: it supplied long-term ideas for a one-month horizon and never hedged, and it conceded that at least one bet against the market was required. The takeaway for anyone using chatbots to pick investments: the models do not have a view of their own, they reproduce today's consensus, and their post-hoc explanations of why a position failed can contain outright false dates and events. Treat the output as a summary of market narratives, verify any factual claim about what happened and when, and do not mistake fluent reasoning for verified investment logic. Source: https://en.oninvest.com/article/chatgpt-claude-and-deepseek-invested-my-eur500-none-of-them-made-any-money
Your answer
Sign in to verify this AI response.
Don't trust us — or the AI. Ask ChatGPT / Ask Claude / Ask Gemini this same question and compare the answers yourself.
More from this topic
Queried on more than 2,000 brokers, ChatGPT warned users that a broker was a scam in 94.2 per cent of cases - so 5.8 per cent of scam cases drew no warning - but only 34.3 per cent of the cases it classified as safe were actually correct. In other words, roughly two in three of its safety confirmations were wrong, even though it was comparatively effective at pointing out outright scams.
Asked 45 common UK pension questions three times each, Copilot, ChatGPT, Gemini and Claude produced answers that read as confident and helpful: 89% scored two or more out of three for accuracy, and on questions about paying into a pension and taking money out accuracy was above 96%. But the same answers included guidance a saver could act on at a cost: early pension withdrawal was described as "usually penalised at 40 to 55% tax" rather than being normally unavailable before age 55 (rising to 57 in 2028); carry forward was explained without the cap that personal contributions only attract tax relief up to 100% of UK earnings; transfers were discussed without noting that moving a defined benefit pension worth more than £30,000 legally requires regulated advice; and the Pension Protection Fund was described as guaranteeing defined benefit pensions without the 90% compensation cap that applies below scheme pension age.
Between 25 and 28 August 2026, after French companies published their half-year results, the communications agency Reputation Age and the brand-visibility platform Bubbling ran 8,671 conversations with ChatGPT and Gemini under an automated 'economic journalist persona' and generated 34,684 answers about the half-year accounts of 39 CAC 40 companies. The study reports that 40% of the figures the two assistants returned were incorrect, with correct retrieval varying by indicator: net income 67%, net debt 66%, revenue 65%, organic growth 64%, free cash flow 62%, EBITDA 57% and published growth last at 42%. By company the ranking runs from Capgemini (81% correct) and Airbus (80%) down to Credit Agricole (32%) and EssilorLuxottica and BNP Paribas (35% each); only sixteen of the 39 companies are named. The authors split the errors into two families: figures that are outdated or simply invented (from 19% of responses on net income to 36% on EBITDA), and answers that cite a genuine official figure that does not correspond to the question asked (24% of responses about published growth), including confusions between organic and published growth and between quarter and half-year figures. Their summary of the pattern: 'the AI finds the right document but takes the wrong figure from it.'