We Pointed an AI at a Real 26-Position Portfolio. It Was One Position.
Last updated August 2026
Short answer
We connected a real brokerage account read-only and asked for an analysis. The account holds 26 positions. One of them, an S&P 500 index fund, was 76.3% of the money. The bottom thirteen holdings came to 6.2% between them. Measured from average cost, the whole portfolio was up 9.0% and the index position on its own was up 8.2%, so twenty-five individual stock selections moved the total by roughly 0.8 percentage points. We had set out to hunt for fees and found about four dollars a year. Walnut is informational and is not an investment adviser.
The numbers below are real. They come from a live account connected read-only on 7 August 2026, with the account holder's permission, computed from actual quantities and that evening's closing prices. We publish no dollar amounts, because the structure is the finding and the balance is not.
Where the money actually was
| Holding | Share of portfolio | Note |
|---|---|---|
| SPY | 76.3% | The whole portfolio, effectively |
| Next 2 holdings | 6.5% | INTC and PLTR combined |
| Holdings 4 to 13 | 11.0% | Ten names, each around one percent |
| Bottom 13 holdings | 6.2% | Thirteen names, none above 0.6% |
The account statement lists twenty-six lines, each rendered the same size, sorted in a way that has nothing to do with how much money is in them. Sorting by value and adding a running total takes about a minute and changes the picture completely: the cumulative percentage passes eighty at the second line.
Four findings
1. The portfolio is one position wearing twenty-six names
SPY was 76.3% of the account. The other twenty-five holdings together came to 23.7%, and the bottom thirteen of those came to 6.2% between them. Eight positions were small enough that a single percent move in any of them changes the total by roughly nothing. The account statement lists twenty-six lines, all the same size on screen, and nothing in that list communicates that one of them is three quarters of the money.
2. The stock picking moved almost nothing
Measured from average cost to the close that evening, the whole portfolio was up 9.0%. The SPY position on its own was up 8.2%. So twenty-five individual selections, each requiring a decision and a reason, moved the total by roughly 0.8 percentage points. That is not an argument that the picks were bad. Several were excellent. It is an observation that at these weights, being right barely registers.
3. The best and worst picks were both invisible
The best performer was up 127% and the worst was down 42%. Both were around 1% of the portfolio or less, so neither materially changed the outcome. A holder checking individual positions would have felt both of those strongly, and the account would not have noticed either.
4. The fee hunt found four dollars a year
This was the test we actually set out to run, and the answer was almost nothing. Only three of the twenty-six holdings are funds, and they carry a blended expense ratio of about 0.099%, which works out to roughly 0.078% of the whole portfolio annually. In a portfolio of individual stocks there are no ongoing fees to find, and the fee-audit framing that suits an advised or fund-heavy account had nothing to work on here.
Get a recommendation for your situation
Walnut is the AI that knows your portfolio: ask anything in plain English, research any fund, and get an honest second opinion. On the broker you already use, read-only, and you approve every trade. Walnut is not a registered investment adviser.
Method
| Element | Detail |
|---|---|
| Account | A real individual brokerage account, connected read-only, used with permission |
| Date | 7 August 2026, using that evening's closing prices |
| Holdings | 26 equity positions, mostly fractional. No options, no crypto |
| Return basis | Average cost per position against the closing price. Not time-weighted |
| Fees | Published expense ratios for the three funds held, weighted by position value |
| Not published | Dollar amounts and account identifiers. The structure is the finding |
Everything here is arithmetic on data read from the account rather than an estimate. That is the specific difference between an assistant connected to a brokerage and one answering from memory: the second produces a fluent description of a portfolio it has never seen. See where AI answers are and are not reliable.
Three things this does not show
1. The returns are not time-weighted
Each position was bought at a different time, so comparing them to each other measures holding period as much as selection. The portfolio-versus-index-leg comparison inherits the same limitation, and it is the reason we describe the gap as small rather than attributing it precisely.
2. One account is one account
This is a case study rather than a survey. What makes it worth reading is not that it is representative, it is that every number in it came from a real account rather than from an illustration, and the shape it revealed is a shape most people never see in their own.
3. Concentration in an index fund is not the same as concentration in one company
Being 76% in a broad index is a very different risk from being 76% in a single stock, and nothing here says otherwise. The point is that the portfolio behaves as an index fund with a rounding error attached, which is worth knowing if you believed you were running something else.
The third is the one worth holding on to. Nothing here argues that this account is badly built. Being largely in a broad index fund is a defensible place to be and many people would call it correct. The finding is about the gap between what the holder is doing and what the holdings list appears to say they are doing.
Run this on your own account in five steps
| Step | How |
|---|---|
| Get every position with its quantity and average cost | From the broker, or by connecting the account to an assistant that can read it |
| Multiply by current price and sort by value | The sort is the whole exercise. Statements list alphabetically or by ticker |
| Add a cumulative percentage column | Where the running total passes 80% is the honest answer to how many holdings you have |
| Compare the total against your largest holding alone | If the gap is small, the rest of the portfolio is doing less than it appears |
| Check what the smallest positions are contributing | Anything under half a percent is a decision you are maintaining for no effect |
The third step is the one that produces the surprise. Almost nobody knows where their cumulative weight crosses eighty percent, because no brokerage interface presents it, and the answer is frequently one or two holdings rather than the number of lines on the statement. The fifth step is the cheapest cleanup available: positions under half a percent are decisions you are still maintaining, still watching and still feeling, for no measurable effect on the outcome.
Related: AI portfolio analysis, what an advisor would check in your portfolio, and getting a second opinion.
FAQ
What did the AI find in the real portfolio?
That a 26-holding account was 76.3% a single S&P 500 index fund. The other 25 positions came to 23.7% between them, and the bottom 13 came to 6.2%. Measured from average cost, the whole portfolio was up 9.0% while the index position alone was up 8.2%, so 25 stock selections moved the total by about 0.8 percentage points.
Is holding 26 stocks diversified?
It depends entirely on the weights, which is what a list of holdings hides. Twenty-six positions where one is three quarters of the money is not twenty-six positions in any meaningful sense. The number that matters is where the running total of your sorted holdings passes eighty percent, and for this account it passed there at the first line.
How do I find out if my portfolio is actually concentrated?
Sort your holdings by current value rather than alphabetically, then add a cumulative percentage column. Where that running total crosses eighty percent tells you how many holdings you effectively have. Most brokerage statements sort by ticker, which is precisely the ordering that conceals this.
Does this mean stock picking is pointless?
It means picking at these weights is close to costless in both directions. The best pick in this account was up 127% and the worst down 42%, and neither changed the outcome, because both were around one percent of the money. If you want your selections to matter, they have to be large enough to matter, which is also the argument for keeping them deliberately small.
Can an AI assistant really see my portfolio?
Only if you connect an account to it. A general chatbot knows nothing about your holdings and will answer from whatever you paste in. An assistant with a read-only connection is reading rather than recalling, which is what made every number on this page checkable rather than plausible.
Why not publish the dollar amounts?
Because the structure is the finding and the balance is not. Weights, returns and tickers are what make the analysis reproducible; the size of the account changes none of the conclusions and is nobody's business. Every percentage on this page comes from real quantities and real prices.
What should someone in this position actually do?
Nothing automatically. Being mostly in a broad index fund is a defensible place to be, and plenty of people would call it the right answer. What is worth avoiding is believing you are running an actively selected portfolio when you are running an index fund with twenty-five decorations, because the two call for very different amounts of your attention.
Related articles
Walnut is informational and is not an investment adviser, and nothing here is investment advice. Figures are from one real account on one date, computed from average cost rather than time-weighted, and are a case study rather than a representative sample. Past performance does not indicate future results.