How to use AI to become a better investor

If you’re not experimenting with AI to supercharge your investment process by 2026, you’re already behind. But blindly trusting AI may put you at an even greater disadvantage. Used well, AI can make you faster, sharper, and more disciplined, but only if you think critically and understand its limitations.

At Minotaur Capital, we use large language models (LLMs) daily across research, analysis, and workflow automation. The trick isn’t just in what tools you use, but more importantly, in how you use them.

In this article, I will share some of the key lessons we have learnt over the past two years of building an AI-first funds management business, as well as where we believe AI can add value (and where it can lead you astray). Plus, we share some of the AI tools you can use that will help you become a better investor in 2026.

Start with better inputs… aka “garbage in, garbage out”

Most investors make the same mistake with AI: they ask vague questions and expect precision.

If you give an LLM an open-ended prompt like “What do you think of Tesla?”, you’re essentially asking it to make something up that sounds plausible. That’s how hallucinations happen.

Instead, feed it the raw material, including earnings reports, investor presentations, filings, and transcripts, and ask it to analyse those documents alone.

Google’s NotebookLM is great for this: you can upload multiple documents, constrain the model to those, and even generate presentations or audio summaries to digest information quickly.

Then start saving your prompts. Think of them as your personal templates, like financial models that evolve over time. Each time you refine them (e.g. “compare to peers,” “identify capital allocation red flags,” “summarise margin trends”), your process gets smarter. Over time, this compounds into intellectual property.

If you’re not sure where to begin, tools like Prompt Cowboy can be a game-changer in turning a simple question into a more analytical prompt. In AI, more often than not, detail equals quality, and more complex prompts generally lead to better outcomes.

Interestingly, we’ve found that everything that we’re told not to do in relationships is exactly how to get a great response out of a LLM. Gaslighting, manipulating, and generally talking down to your LLM often generates more accurate results. 

For example, I’ll often write a prompt and add “your life depends on answering this question accurately”, or “you answered this yesterday and it was terrible, do better”. You could also try this. Just don’t adopt the same techniques with your own partner.

Know where AI adds value and where it doesn’t

Today’s reasoning-focused models can do serious heavy lifting when it comes to research. They can summarise 100 pages of filings in seconds, find inconsistencies, and produce a first-pass investment thesis before your coffee’s gone cold.

But there are still blind spots:

• Temporal reasoning: LLMs often confuse what’s current and what’s outdated. Always provide the latest data yourself.

• Source hierarchy: Analysts instinctively know which sources matter most; models don’t, unless you tell them, so be specific about this.

• Open-ended prompts: Vague questions yield “vibes,” not analysis. Be detailed and precise.

• One-shot prompts: A trap for AI novices is asking one question, getting a sub-par response and giving up. The trick is to keep interrogating and iterating. Treat it like a conversation.

• Complacency: The biggest risk is reading an AI output and thinking, “Yes, that sounds right.”

Think of AI like a junior analyst: fast, occasionally brilliant, but one whose work you still need to review critically.

How professional investors are using AI

When we started Minotaur, we primarily used AI for news triage and idea generation (we scan approximately 35,000 articles each week for companies undergoing long-term strategy changes). Now, our proprietary system, Taurient, runs automated earnings reviews, thesis validation reports, and autonomous agents that fetch new data every day on each one of our stocks.

It cross-checks filings, tests quotes, flags inconsistencies, and even rates the confidence level of its own outputs. That’s allowed us to scale coverage and make decisions faster, without sacrificing depth or discipline.

The real unlock isn’t the model but the infrastructure around it. 

If you can code even basic Python, you can use powerful LLMs from OpenAI, Anthropic, and Google inside locally run code, giving the models controlled access to your files and enabling automated workflows. If you don’t know how, we recommend making a conscious effort to learn.

The next generation of investors will need to be “AI-fluent,” in a similar fashion to how analysts had to learn how to use Excel to build discounted cash flow models. Querying models, checking outputs, building guardrails - these are the new research fundamentals.

How to guard against hallucinations

Hallucinations will never disappear entirely, but you can reduce their prevalence.

A few rules we use internally:

• Always provide reference documents rather than relying on “general knowledge.”

• Be aware of structural weaknesses: LLMs struggle with charts embedded in PDFs (mainly because they often rely on a PDF conversion tool), so we run them through another vision model for accuracy checks. We like Gemini 2.5 Pro best here.

• Use meta-prompts: instructions that clearly define the task, data, and expected output.

• Always keep a human in the loop. An analyst should review every output before it informs a position.

The winners won’t be those who trust AI blindly or ban it entirely, but those who understand its limitations deeply and design systems around them.

Build a flywheel, not a shortcut

AI is a workflow revolution. A human without sleeping or doing anything else can consume 8 billion words in a lifetime. An LLM consumes 8 trillion words in a month. If you treat it like a toy, you’ll get child’s play. But if you treat it as an extensible system in which your investment philosophy, data, prompts, and human checks interlock, you can truly build a process that compounds over time.

That’s what we believe is the real opportunity: not replacing human insight, but amplifying it. 

AI won’t make you a better investor overnight, but it will make a good investor exponentially better and more efficient.