Using AI to Make My Portfolio More Tax-Efficient
The other day at the playground, I got into a conversation with my friend Nick, who knows a lot about investing. We started talking about our portfolios, and I mentioned a large capital gains distribution I had just received. Nick explained why ETFs typically don’t have these distributions, unlike mutual funds. The explanation — something about how ETFs work with brokers and market makers to avoid selling underlying holdings — was fascinating but slightly over my head.
When I got home, I decided to dig deeper. I used ChatGPT and asked:
“Yesterday my friend was trying to describe to me how you can defer capital gains and ETFs compared to mutual funds because mutual funds have to distribute their capital gains or ETFs even if they want to sell their holdings they don’t have to actually sell them they can sell them to a market maker or a broker. Can you explain this to me? I don’t really fully understand it.”
The reason the prompt isn’t extremely clear and is ‘casual’ in tone is because I often just use voice mode to talk to the LLMs. They don’t care if your prompt is refined. You can just talk to them and they almost always figure it out. I highly recommend this form factor.
The response helped me understand the mechanics behind ETFs’ tax efficiency and the difference compared to mutual funds. That realization made me think: What other tax inefficiencies might I be overlooking in my own portfolio?
This curiosity set me off on a quick but surprisingly impactful process of using AI tools to better understand my portfolio and take actionable steps. Here’s what I did:
How I Used AI to Analyze My Portfolio
-
Gathered My Data
I downloaded a PDF from my brokerage account with all the details: what I owned, current gains, embedded capital gains, cost basis, and so on. -
Uploaded It to Claude And Built A Prompt
I created a project in Claude and uploaded the PDF. Using Claude’s prompt builder, I narrated what I wanted to do: identify simple ways to make my portfolio more tax-efficient. This is the meta-prompt I used - again, this was just me speaking to it.“I have a PDF extract of my investment portfolio. The investment portfolio is split across a taxable brokerage account and several retirement accounts, including IRAs, Roth IRA’s and 401(k)s. All accounts share the same assets meaning that there are funds and ETFs that are purchased in both. Where certain assets are held might not be that tax efficient. I want to design very specific instructions to change my portfolio to make it more efficient. I do not want to incur large tax bills due to embedded capital gains. I am looking for a few key changes I can make that are essentially no-brainers.”
This generated a detailed prompt, which I then fed into the Claude project.
-
Reviewed Claude’s Suggestions
It gave me a strong first pass. Two or three of its suggestions were obvious changes I could make right away. These were actionable and straightforward, and they immediately made sense to me. -
Refined and Learned
I went back and forth with Claude, refining the outputs to make them more specific. The back-and-forth wasn’t just about improving the recommendations; it also helped me better understand the dynamics of tax-efficient investing strategies. This was as much about learning as it was about deciding what changes to make. -
Created an Implementation Plan
Finally, I worked with Claude to refine a document that summarized the key takeaways and outlined an implementation plan. One of the most specific and actionable recommendations was to:- Swap two investments: I had assets in my taxable account that generated ordinary dividends, taxed at my marginal rate. Claude recommended moving these assets into a retirement account where they could grow and compound tax-free.
- Simplify the process: Since I had similar assets in both my taxable and retirement accounts, I could essentially swap their locations. Selling the assets in the taxable account incurred a small capital gains event, but it was minimal.
- Understand the payoff: Claude helped me calculate the long-term tax savings from this move, factoring in the compounding benefits over the years.
The document wasn’t just for me — it was something I could share with my wife to walk her through what we were doing and why it made sense. If you’re curious, here’s the document without the specific numbers. Just to be clear, I didn’t write any of this – Claude did. Also, the real document has very specific numbers – it tells me exactly what I need to do, how much I will save, and what the payoff period is.
What I Learned
This whole process took about 30 minutes. The insights were actionable, and the tool helped me break down decisions that had felt intimidating before. It even identified a few additional changes I could consider in the future, though those would take more effort.
I’ve always thought of myself as pretty informed about investing, but this reminded me that there’s always more to learn. Tools like ChatGPT and Claude don’t just explain things; they can help you take real steps to improve your situation.
Even if you’re comfortable managing your portfolio, I think it’s worth exploring how tools like this can push you a step further. For me, the economic benefit was real, but more importantly, I came away with a better understanding of my own portfolio.