AI now screens stocks, sums up earnings, reads charts, tests rules and places orders. Here is what each of those jobs looks like with today’s tools, what the research says AI still can’t do, and a newer route: stock perps that trade around the clock.
November 15, 2026
20
min to read
The short answer
AI stock trading means using AI to screen stocks, summarize news and earnings, read charts, backtest rules and automate orders. Tools such as Trade Ideas, TrendSpider, Composer, Danelfin and Robinhood’s Cortex cover parts of the job, and stock perps on trade.xyz now trade around the clock. AI can’t reliably predict prices, so test rules first.
Two kinds of AI do the work. Machine-learning models learn patterns from market data and turn them into scores, signals or detected chart patterns. Large language models (LLMs), the AI behind ChatGPT and Claude, read and write: they summarize, explain, turn your words into a screen or a rule and, connected to the right tools, place orders. Most products mix the two. Here is what each part of a stock trader’s job looks like with AI in November 2026:
| Job | What AI does well | Where it falls short | Tools that do it |
|---|---|---|---|
| Screening | Turns a question in plain English into a stock screen, or ranks stocks with a model score | A score is a probability learned from the past, not a forecast | Danelfin, Robinhood Cortex, Trade Ideas |
| News and earnings | Sums up what may be moving a stock, from news, filings and analyst notes, in seconds | It can misread a number or miss the latest story | Robinhood Cortex Digests; Claude or ChatGPT with web search; Ethy finds why a market moved with a web search (it has no news feed or earnings calendar) |
| Reading charts | Draws trendlines, flags patterns, reads levels off a chart or a screenshot | A pattern is not a prediction, and a screenshot can be misread | TrendSpider, Claude, Ethy |
| Backtesting | Turns a rule into a test on past prices in minutes | Easy to overfit, and a language model may already know the period you test | Composer, TrendSpider, Trade Ideas, Ethy |
| Automating rules | Runs a rule or sends orders to your broker while you’re away | It does exactly what it was told, bugs included | Composer, TradersPost, Trade Ideas, Robinhood Agents, Ethy |
| Journaling | Reads your trade history and finds the habits that cost you | Small samples throw up patterns that aren’t real | Claude, or another assistant that can run code |
The pattern is plain: AI is strongest at reading, sorting and checking, where a mistake costs you time. It is weakest at the thing people most want from it, which is saying where a price goes next. That part comes further down, with the research. Our guide to AI trading covers the same split across every kind of AI trading, not just stocks.
Each tool below was checked on its own site on November 15, 2026. Prices and features change, so treat them as a snapshot. Several of them also appear in our list of the best AI trading bots, next to crypto bots and AI agents.
Danelfin gives US and European stocks and ETFs a daily AI Score built from fundamental, technical and sentiment indicators. The score “estimates the probability of beating the market over the next three months,” and Danelfin itself calls it “one input to a decision, not a recommendation” (Danelfin). Its screener adds an Ask AI box, in beta, that turns a question into filters (screener). The model portfolios Danelfin publishes are backtests, which it labels as such. How a score like this fits money you hold for years is covered in our guide to AI investing.
Robinhood Cortex is an assistant in the Robinhood iOS app, included with Robinhood Gold at $5 a month. Ask it for “large-cap stocks with relatively low P/E ratios” and it builds the screen, then shows the criteria, the list and the source it used (Robinhood). Trade Ideas does the same for live scanners: “Describe a scan in plain English and Holly builds the Alert or Top List window for you” (Trade Ideas).
Cortex Digests are AI summaries of “what may affect an asset’s price or your overall portfolio,” drawn from market data, breaking news, research reports and analyst ratings, for the most popular stocks and ETFs (Robinhood). Cortex can’t browse the open web or read social media, and it won’t make buy or sell calls or predict prices. General assistants reach further: with web search on, Claude answers with citations “so you can easily verify sources yourself” (Anthropic), and ChatGPT shows its citations in the chat too. A good habit is to paste the earnings release itself, ask for the three numbers that matter and what management said about guidance, then check those numbers in the release.
TrendSpider, founded in 2016, draws trendlines and finds candlestick patterns for you. Its help center says the trendlines come from “the mathematical properties of the chart” (TrendSpider), so that part is rules, not AI. Its AI is Sidekick, a chat assistant built on language models that can’t browse the internet or place trades (Sidekick), plus a lab where you train machine-learning models on your own setups. A general assistant can read a chart screenshot too, but Anthropic warns that Claude “might hallucinate or make mistakes” on low-quality, rotated or very small images (vision docs). Check every level it reads off the axis.
Composer backtests any strategy you build or describe, on daily prices going back up to 30 years, and says so plainly: “All backtest results are hypothetical and benefit from hindsight” (Composer). TrendSpider’s Strategy Tester runs on intraday data, with depth set by your plan, and Trade Ideas’ OddsMaker, on its Premium plan, tests a scan over the last 60 days. Whatever you use, our guide to backtesting a trading strategy covers the checks that separate a real result from a lucky one.
Composer is a broker as well as a builder. Composer Securities LLC, an SEC-registered broker-dealer and FINRA and SIPC member, runs and rebalances your strategy in a daily window near the close. Its AI writes the rules but “doesn’t make trading decisions for you,” and you confirm each strategy before it runs (Composer). Our Composer alternatives guide covers it and similar tools in depth. TradersPost takes alerts from TradingView, TrendSpider or your own code as webhooks, automatic messages one app sends another, and places them at brokers such as Robinhood, Interactive Brokers, Webull and E*TRADE (TradersPost). It has no AI of its own and holds no money: it “is not a broker-dealer, clearing firm, or custodian” (terms). Trade Ideas’ Premium plan can also trade your scan strategies automatically, through Interactive Brokers, E*TRADE, TradeStation or Alpaca.
This is the most underused job. Export your trades from your broker as a CSV file and hand it to an assistant that can run code. Claude can process CSV files and write Python to analyze them on every plan, Free included (Anthropic). Ask for patterns, not praise:
Here is a CSV of my trades from the last three months: date, ticker, side, entry, exit, size, and my note on why I took each one. Group the trades by setup and by time of day, and show the number of trades, the win rate, the average win, the average loss and the total for each group. Then list my three most expensive habits, with the trades that show them. Mark any group with fewer than 20 trades as too small to judge. Don’t tell me what to buy or sell.
Our list of trading prompts to copy has more like this one, for charts, research and risk.
| Tool | Main job | Places trades? | Price, November 15, 2026 |
|---|---|---|---|
| Danelfin | AI stock scores and an AI screener | No | Free plan; paid Plus, Pro and Elite plans |
| Robinhood Cortex | Answers, screens and news digests inside Robinhood | It prepares orders that you review and submit | Comes with Robinhood Gold, $5 a month |
| Trade Ideas | Scanners and Holly AI signals | Through your broker; automatic trading on Premium | Free with delayed data; $127 or $254 a month, $89 or $178 billed yearly |
| TrendSpider | Charts, pattern detection and backtests | Through webhooks or broker links | No free plan; trials from $19 |
| Composer | AI-built strategies that run by themselves | Yes, as its own broker | Free; $10 or $40 a month, $32 billed yearly |
| TradersPost | Turns your alerts into broker orders | Yes, at your broker | From $49 a month after a 7-day trial |
| Claude | Research, summaries, code and journaling | Only through a broker connector | Free plan and paid plans |
| Ethy | Chart reads, why it moved, backtests and automations | Yes, stock perps on trade.xyz | Free plan; $15, $59 or $179 a month |
Prices: Trade Ideas, Composer, and each company’s own pages above.
Yes, and in 2026 there are four ways to let it. They differ in who decides, who presses the button and where your money sits.
| Route | How it works | Who places the order | Examples |
|---|---|---|---|
| An agent inside your broker | A chat agent with trading tools works in a separate account that you fund | You, while trade approvals are on; the agent, once you turn them off | Robinhood Agents; Public and Interactive Brokers offer their own versions |
| Your own assistant, plugged into a broker | Claude, ChatGPT or another AI app connects to a broker’s MCP server, a standard plug that lets AI apps use outside tools | The assistant, when you ask, within the broker’s approval settings | Robinhood’s Trading MCP; Alpaca’s MCP server, which starts in paper trading |
| A rules platform that executes | You build or describe a rule once, and the platform runs it | The rule, automatically | Composer, which is its own broker; TradersPost, which sends your alerts to your broker |
| An agent on a perp exchange | It trades stock perps, contracts on the price rather than shares, at any hour | The agent, after a confirmation card unless you turn it off, or a rule you activated | Ethy on trade.xyz |
Whichever route you take, the AI can place the trade but can’t carry the responsibility. Robinhood’s help center puts it in one line: “You are ultimately responsible for the trades your AI agent places in your Agentic account.” It also warns that agents “can make errors, misinterpret instructions, act on incomplete or outdated information, and may behave in unexpected ways” (Robinhood).
Four of our guides go deeper: can Claude trade for you? shows how people wire Claude to a broker and what Anthropic allows, can ChatGPT trade for you? does the same for ChatGPT, Robinhood Agents and the alternatives compares the broker agents, and agentic trading covers the trend and what regulators say about it.
For most of the history of retail trading, an AI tool for stocks ended at a brokerage account. There is now a second place to trade a stock’s price. trade.xyz, an exchange built on Hyperliquid’s blockchain, lists perpetual futures on stocks: xyz:NVDA for Nvidia, xyz:AAPL for Apple, xyz:TSLA for Tesla and many more. A perpetual future, or perp, is a contract that follows a price and never expires. You post USDC, a stablecoin pegged to the dollar, as collateral, and you can go long or short, with or without leverage.
Three differences from owning shares matter more than the rest:
These were the busiest single-company stock perps on trade.xyz in the 24 hours to November 14, 2026, 8:52 PM UTC. The window ran mostly through a weekend, the quietest time for these markets, so read the volumes as a floor.
| Rank | Market | Company | 24h volume | Open interest | Max leverage | 24h change |
|---|---|---|---|---|---|---|
| 1 | xyz:NVDA | Nvidia | $150.8M | $377.2M | 20x | -2.6% |
| 2 | xyz:SNDK | Sandisk | $56.5M | $90.0M | 10x | -4.7% |
| 3 | xyz:SPCX | SpaceX | $46.9M | $118.8M | 20x | -3.4% |
| 4 | xyz:SKHX | SK hynix (Seoul listing) | $40.6M | $264.9M | 10x | -3.8% |
| 5 | xyz:GOOGL | Alphabet, Class A | $34.1M | $154.4M | 20x | -2.3% |
| 6 | xyz:MU | Micron | $25.0M | $146.0M | 10x | -3.6% |
| 7 | xyz:CRCL | Circle | $17.6M | $87.6M | 10x | -6.1% |
| 8 | xyz:AMD | AMD | $13.2M | $39.7M | 10x | -3.5% |
| 9 | xyz:COIN | Coinbase | $12.7M | $16.5M | 10x | -6.3% |
| 10 | xyz:MSTR | Strategy | $11.9M | $32.0M | 10x | -8.9% |
Open interest is the value of positions still open. Tesla and Apple ranked 13th and 14th, with $9.9M and $9.7M of volume. Three of the ten make memory chips: Sandisk, SK hynix and Micron. How prices behave at night and over the weekend, what funding costs, how margin works and how to place a first trade are in our guide to trading stocks on Hyperliquid.
Not reliably. AI does find real patterns in stock returns, but they are small, they fade once many traders use them, and a backtest can make them look far larger than they are. Four lines of research say so.
Shihao Gu, Bryan Kelly and Dacheng Xiu tested a range of machine-learning methods on nearly 30,000 US stocks from 1957 to 2016 (Review of Financial Studies, 2020). Trees and neural networks clearly beat the older regression models, and portfolios built on their forecasts did far better. Still, the best models could predict only 0.33% to 0.40% of the month-to-month variation in a single stock’s return, what the authors call a “tiny signal-to-noise ratio.” A later study by Doron Avramov, Si Cheng and Lior Metzker (Management Science, 2023) found that much of the profit from such signals came from hard-to-trade stocks and turbulent markets, and that it “further deteriorates in the presence of reasonable trading costs.”
Alejandro Lopez-Lira and Yuehua Tang had GPT models rate news headlines for 4,123 US stocks, from October 2021 to May 2024, and the ratings predicted part of the following days’ move (Journal of Financial Economics, 2026). The edge weakened year after year: the strategy’s Sharpe ratio, its return per unit of risk, fell from 6.54 at the end of 2021 to 1.22 in the first five months of 2024, and the authors link the decline to wider use of the same models. It is an old pattern. R. David McLean and Jeffrey Pontiff found that return patterns earned 58% less once they had been published in academic journals (Journal of Finance, 2016).
FINSABER, a study presented at KDD 2026, a leading data-science conference, re-ran published LLM stock-trading strategies over two decades and more than 100 stocks. Advantages reported in short tests “deteriorate significantly” in the wider test, and simple buy and hold was consistently among the best performers (Li and coauthors).
A model trained on text up to a cutoff date has read what happened before that date. In “The Memorization Problem” (2025), Lopez-Lira, Tang and Zhu show that LLMs recall exact market figures from before their cutoff, so they “cannot be trusted for economic forecasts during periods covered by their training data.” Telling the model to ignore what it knows didn’t stop the recall. Not every study finds the effect large, but the safe rule is simple: judge an AI’s stock calls only on dates after its training cutoff, or on a live paper account.
So the useful jobs for AI sit around the prediction, not in it: finding candidates faster, reading more, testing honestly and following rules without flinching. Our guide to whether AI can predict the stock market goes further into the research, with a test on real data.
Most of the ways AI stock trading goes wrong have nothing to do with how clever the model is.
Start where a mistake costs time, not money, and hand over the trading last.
Research is the safe place to start: a screen, a summary of an earnings call, a second read of a chart. You still decide, and you still place the order.
Which stocks you trade, the most you risk on one trade, when you stay out (earnings day, the first 15 minutes, weekends on perps). Paste them into every chat, or keep them where your tool reads them each time.
Something a machine can check, such as “buy when the daily RSI closes under 30, sell above 70, stop 5% below entry.” Backtest it with trading costs, look at the worst losing stretch, compare it with buy and hold, then check it on months it wasn’t tuned on. Our guide to algorithmic trading for beginners takes a rule from idea to live, step by step.
Simulated orders on live prices first, then the smallest real size you can, with approvals on so you see every order before it goes out.
A stop that sits at the broker or exchange, a daily loss limit, a cap on size and a way to stop everything in one tap. Read every trade it makes for the first few weeks.
Ethy is an AI trading agent you ask before you trade. For stocks, it works on trade.xyz’s perps: companies such as AAPL, MSFT, GOOGL, AMZN, NVDA, META, TSLA, AMD and PLTR, ETFs such as SMH and TLT, and indices such as the S&P 500 and XYZ100 (trade.xyz’s own XYZ U.S. 100 Index, not the Nasdaq-100). They are synthetic: you trade the price, long or short, and own no shares, votes or dividends. New trade.xyz listings show up in Ethy without anyone adding them.
What it doesn’t do: hold shares for you at a broker, trade options, or give you an earnings calendar, a news feed or P/E ratios. Pair it with your broker or a screener for fundamentals. To split an earnings week between Claude and Ethy, see Ethy vs Claude. Ethy is software for research and analysis and doesn’t give investment advice.
Yes. Robinhood Agents trades in a separate account you fund, with approvals on by default. You can plug Claude or ChatGPT into Robinhood’s Trading MCP, or Claude into Alpaca’s MCP server, which starts in paper trading. Composer runs a rule set as its own broker, and TradersPost turns your alerts into orders at your broker. Ethy trades stock perps on trade.xyz. In every case you remain responsible for the trades.
It depends on the job. For screening, Danelfin’s AI Score and Robinhood Cortex. For intraday ideas, Trade Ideas and its Holly AI. For charts and backtests, TrendSpider. For rules that run themselves, Composer, or TradersPost with your own alerts. For research, Claude or ChatGPT; for reviewing your own trades, Claude. For stock perps around the clock, with backtests and automations, Ethy.
It can screen, summarize filings and explain a company, but research gives little reason to trust its predictions. In one study, GPT ratings of news predicted part of the next days’ moves, and that edge shrank year after year as the models spread. Another found LLMs remember market data from before their training cutoff, so backtests over that period can look better than the model really is.
Using AI tools to research or place your own trades isn’t illegal in itself; what regulators go after is fraud, often dressed up as AI. What matters is who holds your money: a broker should be registered, and you can look one up on FINRA’s BrokerCheck. Composer’s broker, for example, is a FINRA and SIPC member. Be wary of any AI service that asks you to send money to its own platform.
Several have free tiers, as of November 2026. Composer’s free plan automates one strategy, Trade Ideas offers a free account on delayed data, Danelfin has a free plan, Claude has a free plan that can analyze your trade history, and Ethy’s free plan comes with 500 credits and no card. Robinhood Cortex needs Robinhood Gold, at $5 a month.
Stock perps on trade.xyz trade around the clock. Their reference price follows outside quotes from Sunday 8:05 PM to Friday 8:00 PM ET and the perp’s own order book on weekends, so prices can jump when outside quotes return. A perp gives you exposure to the price only: no shares, votes or dividends, and leverage can liquidate a position.
Some rules make money in some periods, and many don’t. Research finds the predictable part of stock returns is small and tends to shrink once many traders use the same signal. No bot can promise profits. Test a rule on past data, run it on paper, start small, and judge it on live results with the losing stretches included.
Trading involves risk, including the loss of your money. Ethy AI is a research and trading tool and does not give investment advice or recommendations. Backtests are simulations on historical data and do not predict future results.
Deploy your AI trading agent in one click.
