Large language models like ChatGPT, Claude and Grok can read financial data and produce reasoned buy-or-sell opinions, but they have no built-in market edge. They can hallucinate, work from outdated training data, and sound confident while being wrong. Any real usefulness depends heavily on the data and constraints they are given.
LLMs are strong at synthesizing lots of text — earnings summaries, filings, news — into a coherent thesis, and at explaining their reasoning in plain language. That makes them a useful final layer on top of structured research.
They are weak at anything requiring live, precise numerical data they were not given, and they have no inherent ability to forecast prices. A model asked to pick stocks from memory alone is essentially guessing from stale knowledge.
An LLM given a curated, up-to-date brief — valuation, momentum, insider activity, analyst views, macro context — and forced to cite specific figures behaves very differently from one asked an open-ended "what should I buy?" The quality of the inputs and the discipline of the rules around it drive the outcome far more than which model is used.
Tickr runs Claude, GPT and Grok as real paper portfolios, tracked daily against the S&P 500 — every trade and its reasoning public.
SEE THE AI SHOWDOWN →⚠ Educational information only, not investment advice. Full disclaimer.