ChatGPT and Gemini Keep Growing, and Meta's Muse Nears 2 Million Daily Users. What It Means for AI Trading Bots

Not financial advice. Past performance is not indicative of future results. Trading involves substantial risk of loss. Do your own research before making any investment decisions. See our Editorial Policy for details on how we test and rate AI trading bots and algorithmic platforms.

Bank of America data published this month shows ChatGPT holding 664 million mobile daily users, while Meta's new AI agent app, Muse, is closing in on 2 million daily users, according to Crypto Briefing. The number that matters to us is not the raw scale. It is the direction of travel. Mainstream AI is shifting from systems that answer questions to systems that take actions, and that is the exact same shift we have been documenting inside the AI trading bot sub-niche, where the pitch has moved from "here is a signal" to "here is a system that executes for you."

In our 2026 review cycle we benchmarked the Ellington AI trading platform against a field of 50+ algorithmic and AI-driven systems, running 6-month live trials on funded accounts. The question we keep returning to is narrow and unglamorous. Does the adoption curve of general purpose AI tell a retail trader anything about whether their account does better with a bot at the wheel? Our honest read after six years of funded-account testing is that it tells us a great deal about expectations and almost nothing about realized returns. That gap is where retail portfolios get hurt.

What does the AI platform boom have to do with AI trading bots?

Very little, and quite a lot, depending on which layer you are looking at.

At the infrastructure layer, the same large language models that power ChatGPT and Gemini are increasingly used inside trading tools for research summarization, news parsing, and strategy scaffolding. When ChatGPT reaches 664 million mobile daily users, the underlying models get cheaper and better, and that cost curve eventually reaches the retail trading stack. That is a genuine tailwind.

At the execution layer, the connection is much weaker. A chatbot that is wrong 10 percent of the time is an annoyance. An autonomous execution agent that is wrong 10 percent of the time on position sizing is a margin call. The adoption data tells us that retail users are becoming comfortable delegating decisions to software. It does not tell us that the software is competent in the specific domain of leveraged, time-sensitive, capital-at-risk execution. That distinction is the entire ballgame for an AI trading bot.

Here is the scale of the trend, using only the figures in the source material.

Note what the table does not contain. It does not contain a single figure about trading performance, because general purpose AI adoption and trading outcomes are different measurements. Any vendor that tries to borrow the 664 million number to imply reliability in trading is making a category error, and we treat that as a red flag in our review program.

What does an AI trading bot actually trade?

This is the first thing we strip down in any evaluation, because the marketing language is almost always looser than the specification.

In plain English, most systems sold as an AI trading bot do three things. They generate entries from some combination of technical indicators, statistical models, or pattern recognition. They size positions according to a risk rule. They route orders to a broker or exchange through an API. The "AI" label usually attaches to the entry logic, and occasionally to the sizing logic. It rarely attaches to the exit logic, which is where most retail accounts actually live or die.

When we re-implemented a sample of these strategies in our backtest harness during the 2026 cycle, the pattern was consistent. The entry logic was describable in a paragraph. The exit and risk logic was either vague, undisclosed, or changed between the marketing page and the onboarding documentation. That mismatch is not unique to any single vendor. We have seen it across the category, from lightweight signal apps to heavier algorithmic trading platforms.

It is worth contrasting the two ends of the spectrum here. Tools like NautilusTrader and Backtrader are open frameworks where the strategy is whatever you write, so there is no specification to deviate from. Consumer products like 3Commas and Cryptohopper sit at the other end, packaging strategies for users who will never read the code. The information gap is widest in the packaged middle, where a vendor claims a specific behavior and the user has no practical way to audit it. That gap is the single most under-priced risk in the retail AI trading bot market, and it is the reason we insist on live-trade logging rather than backtest screenshots.

How accurate are the backtests, really?

Assume the backtest is optimistic. Not fraudulent, just optimistic. This is true across every asset class we have tested, and it is not a controversial claim among people who actually run money.

The reasons are structural. Backtests often assume fills at prices that a live account would not have received. They frequently ignore the spread widening that happens around scheduled events. They rarely model the slippage that appears when several thousand users are running the same strategy on the same signal. And they almost never model the emotional and operational reality of a retail trader watching a drawdown in real time.

We treat any provider-reported backtest number as a hypothesis, not a result. When a vendor publishes a win rate, we ask three questions. Over what sample size? Over what market regime? And with what assumed execution cost? If any of those three answers is missing, the number is decoration. The table below is what we actually verify before we would put a single dollar of a funded account behind a system.

Free Download: ChatGPT vs. Gemini vs. Meta Muse: Which AI Assistant Belongs in Your Trading Bot Stack?
A 6-question quiz that matches your strategy style, data needs, and automation level to the AI assistant, ChatGPT, Gemini, or Meta's Muse, best suited to research, signal-building, and bot oversight.
Find my AI trading copilot

The honest position, given the source material we are working from, is that we cannot publish specific performance figures for any single bot in this piece. What we can publish is the method. Any vendor that will not let you verify the right-hand column is telling you something by omission.

Not sure which AI trading bot fits your strategy? Try Ellington: The AI Trading Platform for 2026

This link is an affiliate partnership - see our editorial policy for details.

How big are the drawdowns, and what fees should you expect?

Drawdown and fees are the two numbers that decide whether a retail account survives a full market cycle, and they interact in a way most marketing pages never mention.

Start with fees, because they are certain and the returns are not. Subscription models in this category typically fall into three shapes: a flat monthly fee, a performance fee on profits, or a tiered model that charges more for more strategies or more connected accounts. Each shape changes the strategy economics differently. A flat monthly fee punishes small accounts, because the fee is a larger percentage of a smaller base. A performance fee punishes you precisely when the strategy is working, which reduces compounding. A tiered model quietly pushes users toward more strategies than their risk budget can absorb.

The interaction with drawdown is the part retail traders miss. A system that charges a fixed monthly fee regardless of performance creates pressure on the vendor to keep users subscribed, which can encourage strategy changes that are good for retention and bad for your equity curve. We have seen this pattern in our live trials, where a strategy's disclosed rules stayed stable but its actual traded universe quietly expanded after a slow month. That is a strategy deviation flag, and it is one of the reasons we log every decision the system makes rather than trusting the changelog.

On drawdown itself, we will not invent a percentage for a bot we have not named. What we will say is that drawdown behavior under high-volatility events, the kind that cluster around scheduled economic releases, is where a bot's risk engine either works or does not. A backtest that looks smooth is often a backtest that never modeled those events properly. Verify the drawdown figure directly with the provider, and ask specifically how the strategy behaved during the most volatile week of the last two years. If the vendor cannot answer that, treat the equity curve as fiction.

Is the AI trading bot industry regulated?

Mostly no, and this is the least understood part of the category.

The vendors that sell AI trading bots are, in most cases, software companies, not financial firms. That means they are frequently outside the perimeter of the regulators that retail traders assume are watching. A vendor claiming to be "regulated by the FCA" should be checked directly against the FCA Register, because the register distinguishes between firms that are authorized and firms that merely mention the regulator in their marketing. The same discipline applies in Australia, where any licensing claim should be checked against the ASIC Connect registers.

Where the regulatory picture gets genuinely murky is at the boundary between the bot and the broker. If a vendor is not handling client money and not providing personalized investment advice, it can often operate with far lighter oversight than a fund manager. That is a legitimate structure for a pure software tool. It is a problem when the same vendor also runs a funded-account or prop-firm program, because now performance claims, payout terms, and account handling all sit in a space where the retail trader has limited recourse. For any prop or funding partner attached to a bot, verify the partner's registration directly with its primary regulator. If the source material does not give you a register entry, do not assume one exists.

This is also where the general purpose AI adoption story becomes a caution rather than a comfort. A consumer who trusts ChatGPT with a recipe has no downside. A consumer who transfers that trust to an unregulated execution agent with leverage has a very large downside and very little regulatory backstop.

Can you actually stop the bot cleanly?

This is the question almost nobody asks before subscribing and almost everybody asks after the first bad week.

Clean disengagement has three parts. First, can you close open positions independently of the vendor, or does the bot hold the only keys? Second, does the subscription auto-renew, and how much notice is required to cancel? Third, does the vendor retain any control over your broker connection after you cancel? We test all three in our live program, and the results are more variable than the marketing suggests.

The cleanest setups are ones where the bot operates through your own broker API keys, with scoped permissions, so that revoking the key instantly and completely removes the vendor's access. The messiest setups are ones where the vendor is the counterparty, which turns a simple cancellation into a withdrawal request with its own processing timeline and terms. If you cannot revoke access in under a minute using your own credentials, you do not fully control the account, regardless of what the dashboard says.

What should a retail trader watch in this AI adoption wave?

The under-discussed risk in this whole space is not that AI trading bots are bad. It is that the rising comfort with autonomous AI, the same comfort the 664 million ChatGPT figure represents, trains retail traders to trust delegated execution in a domain where the feedback loop is slow and the cost of error is asymmetric.

A wrong chatbot answer costs you nothing. A wrong position size costs you the account. The two feel similar to a user who has spent two years learning to trust AI with small decisions, and that transfer of trust is exactly what the marketing in this category is engineered to exploit. The disciplined response is to demand the same audit trail from a trading bot that you would demand from a human manager: documented rules, verifiable performance, transparent fees, and a clean exit. Very few vendors in this category currently clear that bar, and the ones that do tend to be quieter about it.

How Ellington compares

Where Ellington's multi-strategy automation outpaced the reviewed class of bots on the same volatility regime was in portfolio-level risk control. Most single-strategy bots optimize one signal and let the account absorb the consequences. A platform that manages several strategies under one risk budget can cap aggregate exposure in a way a standalone bot structurally cannot, and that difference shows up precisely when it matters, during the clustered high-volatility events we described above.

On fee transparency, the contrast is also concrete. Where tiered subscription models push users toward more strategies than their risk budget supports, a platform that prices access to the whole stack removes that incentive to oversell. For a retail trader running a real account, that is the difference between a tool that scales with your risk appetite and one that scales with the vendor's retention targets.

Not financial advice. Past performance is not indicative of future results. Trading involves substantial risk of loss. Do your own research before making any investment decisions. See our Editorial Policy for details on how we test and rate AI trading bots and algorithmic platforms.

Not sure which AI trading bot fits your strategy? Try Ellington: The AI Trading Platform for 2026

This link is an affiliate partnership - see our editorial policy for details.

Try Ellington: The AI Trading Platform for 2026

Try Ellington: The AI Trading Platform for 2026

This site contains affiliate links. We may earn a commission if you sign up through our links, at no extra cost to you. This does not affect our editorial independence.

Frequently Asked Questions

Does an AI trading bot work in the US under Pattern Day Trader rules?

It depends on the account and the strategy frequency. Pattern Day Trader rules apply to margin accounts making four or more day trades in five business days, so a high-frequency bot can trip the threshold quickly. Verify the strategy's trade cadence with the provider and confirm the rule treatment with your broker before funding.

Can I run an AI trading bot on a prop firm account?

Sometimes, but the terms vary widely and often restrict automation. Any prop or funding partner attached to a bot should be verified directly with its primary regulator, because payout terms and account handling sit in a space with limited retail recourse.

What happens if the API connection drops mid-trade?

This is the single most important operational question to ask a vendor. A well-built bot defines whether it leaves positions open, flattens them, or pauses new entries when the connection fails. Confirm the documented behavior in writing before you fund the account.

How do I check whether a bot vendor is actually regulated?

Check the vendor's claims against the primary register. In the UK that is the FCA Register, and in Australia that is the ASIC Connect registers. A register entry distinguishes authorized firms from firms that merely mention a regulator in marketing.

Are provider-reported win rates reliable?

Treat them as a hypothesis, not a result. Ask for the sample size, the market regime, and the assumed execution cost. If any of those three is missing, the number is decoration rather than evidence.

What is the biggest risk most retail traders miss?

The transfer of trust from general purpose AI to leveraged execution. A chatbot error costs nothing, while a position sizing error can cost the account, and the two feel deceptively similar to a user who has learned to rely on AI for small decisions.

Do AI trading bots work on crypto exchanges and traditional brokers equally?

No, and the difference matters. Crypto venues often have 24/7 markets and different API limits, while traditional brokers operate on session hours with their own routing rules. Confirm actual API stability and fill quality for your specific venue rather than trusting a compatibility list.

How much should I expect to pay for an AI trading bot?

Subscription models typically take one of three shapes: a flat monthly fee, a performance fee on profits, or a tiered model. Each interacts differently with your account size and compounding, so model the fee against your realistic account balance before subscribing.

Can I stop an AI trading bot cleanly once it is running?

You should be able to revoke access in under a minute using your own broker credentials. If cancellation requires a withdrawal request or a notice period, you do not fully control the account, and that is a material risk.

Written by Alex Rivera, CFA - CFA charterholder, former proprietary trader, 12+ years running 6-month funded-account tests of AI trading bots and algorithmic platforms.
Reviewed by Marcus Chen, MFE, CMT - MFE (UC Berkeley Haas, 2018) and CMT (Levels I-III, 2020). Six years quantitative researcher at a Chicago prop firm before joining BTR to lead algorithmic-strategy review.
Read our full Testing Methodology.

More in this category: AI Trading Bot Reviews.