Tax-Loss Harvesting With AI

Three kinds of software do this without an advisor. Automated robo-advisors harvest inside accounts they manage. Direct-indexing accounts hold an index’s stocks one by one, so they can harvest a single loser while the index is up. Flagging tools read your accounts and tell you what to look at, and you place the trade. Fees checked October 2026.

Best for the review around a harvest on the broker you keep: Walnut, free. It does not harvest on its own and it does not see cost basis for shares bought elsewhere, so it sits outside the ranking above. It lists your holdings by account, flags positions bought through Walnut that are below cost with a wash-sale note, and places the sale or replacement buy you approve at a trade-capable broker.

How we evaluated

We ranked on four things: how much of the harvest is automatic, what it costs, the balance at which harvesting starts, and how far its wash-sale checks reach across your accounts. Fees and thresholds were read from each provider’s own site in October 2026. E*TRADE Core Portfolios, the Interactive Brokers tax-loss tool and TaxHarvest also come up in this space; we could not confirm their current terms, so they are named rather than ranked. For the tax-efficiency picture beyond harvesting, see tax-aware AI robo-advisor alternatives.

Tax-loss harvesting means selling an investment that has fallen below its purchase price to realize a capital loss, which offsets your capital gains dollar for dollar and, up to a yearly limit (currently 3,000 dollars for most filers), offsets ordinary income, with any excess carried forward. You typically move the proceeds into a similar but not substantially identical fund so you stay invested. The main trap is the IRS wash-sale rule, which disallows the loss if you buy the same or a substantially identical security within 30 days before or after the sale. It only works in taxable accounts. The IRS sets out the rules in Topic 409, capital gains and losses and Publication 550, which covers wash sales. For the full mechanics, see the tax-loss harvesting and wash-sale rule guides.

The hard part of harvesting by hand is not the concept, it is the bookkeeping: knowing which of your positions are below cost right now, and remembering what you bought recently. An assistant can do that arithmetic quickly, but it needs the inputs. Brokerage connections, including the ones AI investing apps use, generally pass positions and market values, not what you paid. So the starting point is your broker's unrealized gain and loss page or tax-lot view, which shows cost basis per lot. Give the assistant those figures and it can list the positions below cost and size each loss.

It can also flag wash-sale risk if you tell it what you bought in the last 30 days, in every account, including dividend reinvestments, and it can explain, in plain words, how the loss would offset your gains and what the wash-sale rule requires. That combination, finding candidates, warning about collisions, and explaining the rules, is exactly what a good research assistant should do: put the relevant facts in front of you so a decision is easier. It is surfacing information, not issuing instructions.

Just as important is what an AI assistant cannot responsibly do. It is not a tax adviser, so it should not be treated as giving tax advice or a compliance guarantee. The wash-sale rule turns on whether a replacement fund is “substantially identical,” a phrase the IRS has never fully defined, which makes it a judgment call for a professional, not a rule an algorithm can settle cleanly.

An assistant also does not see your cost basis unless you give it, does not see your entire tax picture the way a professional preparing your return can, and does not file anything. So the responsible flow is: let the AI surface candidates and risks, confirm the plan with a licensed tax professional, and then execute the trade yourself at your broker. AI shortens the finding step; it does not replace the deciding or the filing step. Treat everything it surfaces as context to verify, not tax advice.

It is worth separating two very different things that both get called “AI tax-loss harvesting.” An automated robo-advisor manages your money and executes harvesting on your behalf, selling and rebuying under its own rules, often daily, inside the accounts it controls. That is hands-off, but you hand over discretion and pay for the management, and the trades happen whether or not you were watching.

An AI assistant on top of your own broker is the opposite posture. It usually has no cost basis, so it works from purchase prices you give it, and it does not manage your money, auto-sell, or give tax advice. You stay in control, you confirm with a professional, and any trade is one you place or approve at your broker. Neither approach is universally better; they trade control against convenience, and which fits depends on how hands-on you want to be.

The line is consistent: AI is strong at the information tasks (listing positions below the cost you supply, checking the buys you list against the wash-sale window, explaining the mechanics) and out of its depth on the judgment and execution tasks (substantially identical calls, accounting for your full situation, filing, and placing the trade). Keeping that boundary is what makes an AI assistant genuinely useful without pretending to be a tax adviser. Verify anything it surfaces with a professional.

Walnut is an AI investing app, and its part in a harvest is the review and the trade, not the automation. The brokerage connection it uses passes your positions and their market values, not what you paid, so for broker holdings you bring purchase prices from your broker’s tax-lot view and the assistant (its own, or inside Claude and ChatGPT) does the arithmetic and explains the rules. For positions you bought through a Walnut basket, it records the price you paid, and its dashboard shows when those are worth less than cost, with a note on the wash-sale rule. Each broker holding is tracked against the S&P 500 as a one-year price return, not your gain or loss. It never sells on its own: any trade is one you approve and your broker executes. For more on how these tools compare, see best AI portfolio analyzer and AI portfolio analysis tools.