AI Trading Platforms vs a Boring Index Fund: The Honest Comparison

The number the adverts leave out#

Mercer surveyed 131 global asset managers in February and March 2026 and asked how they use artificial intelligence. Most of them do: 55% have put AI into at least one investment process. Three-quarters use it to automate work. Roughly seven in ten use it to sift research and data.

Six per cent use it to make investment decisions.

That gap is the whole story. The professionals with the biggest budgets, the cleanest data and the strongest incentive to find an edge have mostly decided that AI is a very good research assistant and a poor portfolio manager. Meanwhile the marketing aimed at ordinary savers points the other way, promising that a model can pick your shares better than you can.

Both can be true. But if you are being sold the second, it is worth knowing that the people who run money are, so far, buying the first.

What each side of this comparison actually is#

An AI trading platform is software that ingests market data, news and sometimes social media, then produces trade signals. Some only suggest. Some place the orders. The newer ones are described as agentic, meaning the software is given an objective and the authority to act on it without checking back each time.

An index fund is duller by design. It buys every company in a published list, in proportion to size, and holds them. No forecasting, no view. Vanguard and State Street run S&P 500 trackers that do nothing but own the 500 largest listed American companies and rebalance when the list changes.

Two terms worth pinning down. An expense ratio is the annual slice of your money the fund keeps, quoted as a percentage. Backtesting means running a strategy against historical data to see how it would have done. Backtests are where most spectacular AI trading claims come from, and they are the weakest form of evidence there is, because a model tuned until it fits the past will fit the past beautifully and tell you nothing about tomorrow.

One category error is worth clearing up. Buying an S&P 500 tracker is not a bet against AI. Nvidia, Microsoft, Alphabet and the rest of the AI build-out sit inside the index already, weighted by their market value. The question is not whether you get exposure to artificial intelligence. It is whether an algorithm should be choosing which slices of it you own.

The scoreboard, where someone keeps score#

The longest live experiment in AI stock picking is the Amplify AI Powered Equity ETF, launched in October 2017, which selects US shares using a quantitative model. It has a public track record and an auditor, which makes it more useful than any vendor's backtest.

Here is that record against a plain S&P 500 tracker at the same date.

Both are net of fees and quoted at net asset value. The one-year and five-year figures cover identical periods; the bottom row does not, because the funds are different ages. Put $10,000 into each five years ago and the arithmetic gives you about $12,450 in the AI fund against roughly $18,630 in the tracker. One fund is an anecdote, not a law of nature. It is, though, the anecdote with the longest audited life.

Widen the lens and the picture holds. S&P Dow Jones Indices has run its SPIVA scorecard for a quarter of a century. In the year-end 2025 edition, published on 3 March 2026, 78.78% of actively managed US large-cap funds trailed the S&P 500 over twelve months. Stretch the window and it gets worse: 88.96% over five years, 85.59% over ten, 92.89% over twenty. Across all domestic equity funds measured against the S&P Composite 1500, 95.01% lost over two decades.

Morningstar reaches the same place by a different route. Its mid-2026 Active/Passive Barometer, published on 12 August 2026, tracked about 9,226 funds holding roughly $29 trillion. Just over 40% of active strategies survived and beat their passive equivalent over one year. Over ten years, 25% did. In large-cap growth, the figure was 5%.

Then there is the question people rarely ask, which is whether last year's winner is next year's winner. S&P's Persistence Scorecard for year-end 2025 followed the large-cap funds that finished in the top quarter in 2021. Five years on, none of them were still in the top quarter. Not a small number. Zero.

The costs that never make the landing page#

Fees decide more of this than skill does.

The Investment Company Institute's 2025 fee study puts the asset-weighted average expense ratio of index equity mutual funds at 0.05%, index equity ETFs at 0.14%, and actively managed equity funds at 0.64%. On $10,000 that is $5 a year against $64.

AI trading tools sit in a different bracket again, because most charge a subscription rather than a percentage. Trade Ideas, one of the longer-running platforms, prices its AI-equipped tier at $2,268 a year. On a $10,000 account that is 22.7% of your capital, paid whether the market rises or falls. The model has to be extraordinary before you are merely level.

Trading frequency costs money too, and the tools tend to encourage it. The FCA studied 176,159 UK investors for Occasional Paper 66, published in April 2025. On apps using heavy engagement features such as push notifications and prize draws, 24% of users day traded at least once, against 3.2% on the quietest apps. Returns on the busier apps were worse across every measure the researchers used, and 10% of those users took losses exceeding 2% of their net income, against 4% elsewhere. The FCA is careful here, and so should we be: it could not establish that the design caused the losses, and nearly all the underperformance traced to crypto and CFDs, which those platforms also happen to sell.

The academic literature on frequent trading is blunter. Chague, De-Losso and Giovannetti followed 19,646 people who started day trading Brazilian equity futures between 2013 and 2015. Of those who stuck with it for more than 300 days, 97% lost money. Just 1.1% cleared the Brazilian minimum wage. That study predates the current generation of tools, so it is not a verdict on AI. It is a description of what happens to human money when trading gets easy and frequent, and no software has repealed it yet.

Where AI genuinely earns its keep#

The evidence above is one-sided on stock picking. It is not one-sided on AI.

Go back to the Mercer survey. Of the managers using AI, 69% report efficiency gains and 91% intend to expand their use over the next year. What they do not report is performance: only 8% say returns improved, and 8% say risk fell. The technology is doing real work in research, data processing and operations. It is simply not the work the retail marketing describes.

Academic testing lands in the same neighbourhood. One team ran language-model investing strategies across more than a hundred stocks and two decades of data in a paper finalised in June 2026. Earlier reported advantages largely evaporated once the sample widened and the clock ran longer. The models proved too cautious in rising markets and too aggressive in falling ones, roughly the worst combination available.

There is one finding that cuts the other way, and it deserves airtime. Morningstar's barometer found that active funds in the cheapest fifth by cost beat their passive rivals 33% of the time over ten years, against 20% for the most expensive fifth. Skill is not irrelevant. But the variable that moved the odds most was price, not cleverness, which rather supports the boring conclusion.

If the machine gets it wrong, who pays?#

This is the part investors discover late.

The Securities Investor Protection Corporation covers up to $500,000 if a US brokerage fails. It does not cover losses from a falling investment or from bad advice. A review of five major platforms published on 27 August 2026 found all of them place decision risk on the customer. Robinhood's terms state that customers assume all risk for agent-executed trades.

Regulators have noticed. FINRA's 2026 Annual Regulatory Oversight Report, published in December 2025, warns that AI agents may act "without human validation and approval" and beyond a user's intended authority, and that badly specified reward functions can push an agent towards decisions that harm the investor. IOSCO's supervisory toolkit of May 2026 catalogues hallucination, model opacity and herding, and names AI washing directly: firms "making false statements as to the level of implementation of AI systems in investment research or selection of securities, and reporting false performance attributed to AI systems."

Not hypothetical. In March 2024 the SEC fined two advisers $400,000 between them for exactly this. Delphia claimed an AI that predicted which companies were about to take off. Global Predictions called itself the first regulated AI financial adviser. Neither claim held up.

FINRA's own guidance on auto-trading services offers the most practical filter I have seen: check registration on BrokerCheck, distrust performance claims without audited proof, and insist that any provider explain with specificity how its technology works. A firm that cannot describe its method in plain language, when asked, has told you something.

Key takeaways#

1. Over the five years to 30 June 2026, the longest-running AI-selected equity fund returned 4.48% a year against 13.25% for a plain S&P 500 tracker, while charging about eight times as much.
2. Only 6% of surveyed professional asset managers use AI to make investment decisions, and only 8% report better returns from it. Most use it for research and admin.
3. SPIVA found 88.96% of active US large-cap funds trailed the S&P 500 over five years to end-2025. None of the top-quartile funds from 2021 were still top-quartile in 2025.
4. Cost is the most reliable predictor in the data. Index equity funds averaged 0.05% a year in 2025; one AI platform's subscription equals 22.7% of a $10,000 account.
5. When an AI agent errs, the loss is yours. SIPC does not cover investment losses, and platform terms place decision risk on the customer.

None of this tells you what to do with your money, and none of it is financial advice. It is what the audited record, the regulators and the researchers currently show.

Frequently asked questions#

Do AI trading bots work?

Some execute reliably. Whether they beat a low-cost index fund after costs is a separate question, and the published evidence does not support it. The one AI-selected fund with a long audited record has trailed the S&P 500 substantially over five years, and academic testing of language-model strategies found early advantages disappeared over longer periods and wider samples.

Are AI trading platforms a scam?

Most established ones are legitimate businesses selling legitimate software. The risk is misdescription rather than outright fraud. The SEC has fined firms for overstating AI capability, and IOSCO treats AI washing as an enforcement matter. Check registration before paying anyone.

Why do so many active funds lose to the index?

Mostly arithmetic. Before costs, active investors collectively own the market and so collectively earn the market return. After fees, trading costs and tax, the average must fall short. Skill exists, but it is distributed thinly and, on S&P's persistence data, does not repeat reliably.

Is buying an index fund a bet against AI?

No. An S&P 500 tracker holds the largest AI companies at their market weight automatically. The choice is about who selects your holdings, not which industries you own.

What should I ask an AI trading platform before subscribing?

Whether the track record is live or backtested, whether an independent party has verified it, whether the firm is registered, what happens if the system places a trade you did not intend, and what the total annual cost comes to as a percentage of the amount you plan to invest.

Are robo-advisers the same thing?

No. A robo-adviser typically allocates you across index funds using rules about risk and time horizon, and usually charges a modest percentage fee. That is closer to automated administration than to AI stock selection.

Does any evidence favour active management?

Some. Morningstar found active funds in the cheapest cost quintile beat passive peers 33% of the time over ten years against 20% for the priciest. Costs move the odds more than anything else measured.

What about AI in professional investing generally?

It is being adopted quickly for research, data handling and operations, and 91% of surveyed managers plan to expand its use. The reported benefit is efficiency, not performance.

Glossary#

Index fund A fund that copies a published list of companies rather than choosing between them. It aims to match the market's return, minus a small fee.

Expense ratio The annual percentage a fund deducts from your money. A 0.05% ratio costs £5 a year per £10,000 invested.

Active management Choosing which securities to hold in the hope of beating a benchmark. AI stock selection is active management with a different tool.

Backtesting Running a strategy against historical data. It shows how a rule would have performed, not how it will perform, and is easily flattered by tuning.

Agentic AI Software given an objective and the authority to act without approving each step. In trading, that means placing orders on your behalf.

AI washing Overstating or fabricating the role of artificial intelligence in a product or investment process. A basis for regulatory enforcement.

SPIVA S&P Dow Jones Indices' scorecard comparing active funds against their benchmarks, published since 2002 and adjusted for funds that close.

Persistence Whether a fund that outperformed in one period does so again in the next. Low persistence suggests luck rather than skill.

SIPC The US body that protects customer assets up to $500,000 if a brokerage fails. It does not compensate for investment losses.

References#

1. S&P Dow Jones Indices, SPIVA U.S. Scorecard Year-End 2025, published 3 March 2026.
2. S&P Dow Jones Indices, U.S. Persistence Scorecard Year-End 2025.
3. Morningstar, Active/Passive Barometer mid-2026 edition, published 12 August 2026, as reported by InvestmentNews.
4. Mercer, Moving Beyond the AI Pitch: Asset Managers' Use of AI, survey of 131 managers conducted February to March 2026.
5. Amplify ETFs, Amplify AI Powered Equity ETF (AIEQ) fund page, performance as of 30 June 2026.
6. State Street Global Advisors, SPDR S&P 500 ETF Trust fact sheet, performance as of 30 June 2026.
7. Investment Company Institute, Trends in the Expenses and Fees of Funds, 2025.
8. Financial Conduct Authority, Occasional Paper 66: Digital engagement practices and investment outcomes, April 2025.
9. Chague, De-Losso and Giovannetti, "Day Trading for a Living?", study of 19,646 Brazilian day traders, 2013 to 2015.
10. Li, Kim, Cucuringu and Ma, "Can LLM-based Financial Investing Strategies Outperform the Market in Long Run?", arXiv, final version June 2026.
11. IOSCO, Supervisory Toolkit for AI Use in Capital Markets, FR/02/2026, May 2026.
12. FINRA, 2026 Annual Regulatory Oversight Report, December 2025.
13. FINRA, Know the Risks of Auto-Trading Services Offered by Unregistered Entities, 29 July 2025.
14. US Securities and Exchange Commission, SEC Charges Two Investment Advisers with Making False and Misleading Statements About Their Use of Artificial Intelligence, press release 2024-36, 18 March 2024.
15. US Securities and Exchange Commission, Office of Investor Education and Advocacy, Artificial Intelligence (AI) and Investment Fraud investor alert, 25 January 2024.
16. Trade Ideas, subscription pricing page.
17. Stacker, "What happens to your money if your AI trading agent makes a mistake?", 27 August 2026.