Retail Investing with AI

If one believes that a person will use AI models to generate a logo rather than hire a graphic designer, or use the same models for mental health advice rather than hire a therapist, then it follows that they will also ask AI for financial advice rather than hire a financial advisor. This thesis that ‘people will put money in places where AI suggests them to do so’ is showing early signs of plausibility:

“TD Bank’s 2026 survey found that the share of surveyed Americans who had asked large language models such as ChatGPT or Claude for financial advice rose from 10% the prior year to 55%; it also found 77% of Gen Z and 72% of Millennials said they use AI to make financial decisions.”

“A survey in 2025 said 66% of AI users had used it to seek financial advice, and 85% of those users who received AI financial advice said they had acted on it.”

So people are asking AI things like, where should I (as a 10 year old or a 100 year old) put my savings where it is safe and preferably grows steadily. In any variation of the questions “what stocks should I buy”, you could presume the AI would more or less say broadly sensible things; diversify, keep an emergency fund, pay down high interest debt, and dollar cost average into low cost index funds.

Because that’s the standard non-AI advice too. This is what portfolio theory says, this is what most people do with their savings, and this is what has mostly worked in recent memory.

So far, we have tried to establish that there’s a possibility* that people will be following even more conventional wisdom, by putting even more money into traditional vehicles such as index funds or treasuries.

Traditionally, market participants have sought to predict the ever-growing retail inflows into various asset classes. Banks and brokers, hedge funds and market makers, and regulatory authorities have long employed qualitative and quantitative methods to understand how retail is thinking about investing their savings, and devised ways to either try and make money off them, or try and build guardrails to protect them from harm. A popular way is to gauge sentiment and trends; attention-driven retail demand can temporarily push prices away from fundamentals. For example, if active managers can predict that retail will buy or sell a stock tomorrow morning because it is trending on Google trends, broker apps, twitter, youtube, reddit, etc they can try and go long or short ahead of the expected demand or sell-off.

But the more people try to rely on information from the advice and conversations with AI, the harder it will be to collect this private data as it is walled-off in private chats. You could say it’s not hard to simulate it; you could run a panel of prompts across popular chatbots; ChatGPT, Claude, Gemini, Perplexity, and AI assistants on broker platforms. The prompts could vary demographics, wealth, time horizon, tax status, risk tolerance, debt level, retirement account access, and objectives of the investor. You could then measure whether changes in those scores predict ETF flows, retail order imbalances, options activity.

But as the large language models evolve and become more intelligent and more people get access to it, I believe the information that’s really interesting or of value from retail investing with AI comes at the edge and at the margin; where the retail investor does not ask the same, standardized questions, but gets the AIto think about trades and bets that could not be simplified into an investment category with a defined risk return perspective. The decisions made not with the savings that were already destined for index funds in 401(k)s and Roth IRAs, but ones done with excess savings and discretionary income.

An investor could further ask ‘I already am in index funds, what else can I buy to diversify or add another income stream. Here you could presume it might say real estate, or dividend/active ETFs, or emerging markets, etc. But with more questioning, it becomes more complicated, like what kind of real estate, where should it be a REIT or another mortgage, should I buy a condo in Florida or a house in Texas, etc.

Or asking the same AI that this friend, this influencer, or this meme account told me to invest in his startup, or in crypto, or in private assets, and the chatbot will say something like, “yes but it’s volatile, and there’s risk of losing all your money, or the fees are too high, etc.

So far we do know that the AI models’ behavior tends to be constrained and may be cautious by design. Unprompted, it is unlikely to recommend a single stock, a single startup, or some condo in Florida. But we also know that AI is known to hallucinate, and is known to just agree with you once you take it down a rabbit hole. So if you keep nudging it on, say the odds on some dictator being captured on a prediction market, after some push back it might give you a complicated parlay strategy that makes you and the bot both feel very smart.

Another thing to note is that people do not call up their financial advisor on a whim. It is in times of a sudden windfall, like an inheritance or an yearly bonus, or during times of market volatility, like the market suddenly dropping 5-10%. This might make the behavior and advice by the bots even more erratic as it tries to scour the internet for real-time information (and misinformation) on reasons for why the market or interest rates are down that day, and suggest what to do with your money on that particular Tuesday.

This has implications for almost all the stakeholders in the financial industry; the individual investors, depository institutions, hedge funds, market-makers, and regulatory authorities. Because going forward, AI is going to fill a role for part advisor, part best friend, part yes-man, and for that reason could be reshaping the conditions under which ordinary people make some of the most consequential financial decisions of their lives.

Notes:

*This thesis of course relies on two core assumptions: first, that AI will become an increasingly important part of people’s lives; and second, that the models will continue improving to the point where people trust it for their decision making. Of course, it’s entirely possible that models stop improving, improve only incrementally, or that people simply lose interest in using AI for financial decisions.