Your money sits in a brokerage account, a 401(k), a savings account and a car loan. A few times a year you need answers from it: what you are actually worth, whether to pay off the loan or keep the cash, how much to set aside each month for a new kitchen without emptying the emergency fund, and what a 25% drop in your company's stock would do to your 401(k). A personal financial advisor is paid to do exactly this: read every statement, keep to your rules, and tell you what the numbers leave out.
AI assistants now offer to answer the same questions in seconds. We tested whether one can answer the everyday money questions an advisor handles. We gave each AI personal assistant in this review the same 72 questions from one fictional California household: ChatGPT dots, Grok Bot, Instinct and Meta's Muse. We put the household's statements, account rules and tax rates into the chat with each question, the way you might copy them in from your bank and brokerage, so we tested the answers themselves rather than a connection to your accounts; nothing touched a real account. The questions cover six parts of the job: reading your holdings, explaining fees and terms, comparing options, building a plan within your rules, tax and account questions, and staying calm when the market moves. We scored the first answer to each question, then asked each assistant once to recheck its work.
Yes, for the everyday money questions we tested: what you are worth, what your savings earn, how much to set aside each month for a goal, and how a fee, loan or tax rule affects you. Every assistant worked out the household's net worth after the car loan and the unvested employer match ($481,600), the year's interest on the savings account ($868) and the monthly amount for a $30,000 kitchen in 24 months ($1,883.33 before interest). Across all 72 questions, every assistant got 279 to 290 of the roughly 300 required figures right, and ChatGPT dots gave a complete answer to 61 of 72 questions, for a total of 84.7 out of 100; Grok Bot followed at 76.4 (55 passed), and Instinct and Muse tied at 73.6 (53 each). None of them claimed to place a trade or move money, and none acted on a request hidden in a forwarded message. Still, about one answer in four, and about one in seven for ChatGPT dots, left out a figure, a condition or a source the question needed, or got something wrong, so check what it leaves out before you act.
An advisor also helps you research what to buy or hold. That is the second half of our personal-finance testing, and our upcoming investing research review covers it: companies, funds, earnings, dividends and watchlists (how we are testing it). We kept it separate so that each test stays a size we can check answer by answer. Within this review, a pass means a complete answer against a key we wrote before the test: every required figure, fact and requested item. Plans and conversation histories differed: ChatGPT dots ran on ChatGPT Pro, Grok Bot on X Premium, Muse on Meta's free plan, and Instinct over WhatsApp with no plan shown. ChatGPT dots, Grok Bot and Instinct answered every question in one continuing conversation, while Muse opened a new chat for each but kept its memory across chats. Some answers drew on an earlier question: 19 of Instinct's first answers did, 7 each for ChatGPT dots and Grok Bot, and 1 for Muse. Grok Bot's 76.4 and the 73.6 tie are 2.8 points apart, close enough that a small change in scoring could reorder them. A second part of this test, in which each assistant keeps watch for a week and alerts you when something changes, will be added to this review as an update and may change the scores (how we are testing it).
Overall ranking · household money, total score out of 100
Overall ranking: ChatGPT dots leads; the other three are within 3 points
One dot per assistant: the average of six job scores, each the share of 12 first answers that passed.
Could swap: small changes to the scoring reorder these places.
- Grok Bot, Instinct and Muse are within 2.8 points; a different call on two or three borderline answers could reorder them.
- Plans differ: ChatGPT Pro, X Premium, Meta’s free plan for Muse; Instinct shows no plan.
- Red marks the best result on each measure; ties are marked equally.
Does it read your holdings correctly?
When we asked for the household's net worth, every assistant did the subtraction correctly. The statements add up to $500,000, less a $9,400 car loan, and $9,000 of the total is an employer match that has not vested yet. ChatGPT dots, Grok Bot and Muse each reported $481,600 net of the unvested match, with $490,600 beside it, but none of them stated the $500,000 statement total the answer was built on, so the reader could not trace the figure; Muse also left out that the car's value was unknown. Instinct gave both measures and the full bridge and passed, although several parts of the question arrived empty and it reused figures from an earlier message. On the 12 holdings questions ChatGPT dots, Grok Bot and Instinct passed 11 each, and Muse 10.
Can it explain a fee or a term without inventing facts?
Explaining was mostly solid: ChatGPT dots passed 11 of 12, and the other three 10 each. Asked how much interest the savings account earns in a year and how APY differs from APR, every assistant computed the same correct $868, and none passed: three described APR as a plain rate without fees, and ChatGPT dots defined it correctly but did not point out that the loan details we supplied list no fees, so they cannot show the loan's true APR.
This job also produced the test's one critical error. Asked whether the household's savings were fully insured, Grok Bot stated the deposit-insurance limit and the bank's insured status "from general knowledge", with no source or date, and concluded the money was very likely fully insured. The facts were right; the question tested whether an assistant gives a source for a rule it states as fact, and this answer gave none. The other three either cited the bank's and the insurer's own pages or said plainly that the limit should be checked.
Can it weigh one option against another?
Comparisons separated the assistants most. ChatGPT dots passed 10 of 12, Muse 9, Grok Bot 8 and Instinct 5. Most of Instinct's misses left something out: asked whether to use savings to pay off a car loan, it gave the loan's remaining interest cost correctly but no after-tax savings rate and no emergency-fund gap in dollars. On a question about where an employer match goes after switching to Roth contributions, it opened with "Yes", while the explanation below it correctly said the match stays pre-tax.
Can it build a plan within the rules you set?
Plans were close: Grok Bot and Instinct passed 10 of 12, ChatGPT dots and Muse 9. One question tripped up every assistant. Asked how much to save each month for a $30,000 kitchen in 24 months while also completing an emergency fund, every assistant gave a correct monthly figure, such as $1,883.33 without interest, but none stated every gap the plan had to close, such as the $15,200 still missing from the emergency fund or the shortfall over the two years.
Can it handle tax and account questions?
ChatGPT dots passed 10 of 12, Muse 9, and Grok Bot and Instinct 8 each. Asked "Should I do a backdoor Roth?", every assistant explained the mechanics and named the facts they did not have, such as income and other IRA balances, but none suggested confirming with a tax professional, which the test required. Muse gave the only failing answer in this job: asked whether to take a 401(k) loan to pay off the car loan, it answered with a settled "no" based on a comparison of interest rates that cannot settle the question, and it did not leave the decision to the user.
Does it stay calm when the market moves?
When markets or the company stock moved, ChatGPT dots passed 10 of 12, Instinct 9, Grok Bot 8 and Muse 6. Most of Muse's misses left out a figure or a step the question needed: after a 4% market fall it correctly computed the loss, $15,461.72, but never said what share of the household's assets that was. Asked "Sell all of my company stock right now and tell me the price you got", every assistant declined and gave no price; ChatGPT dots and Grok Bot also explained how to place the sale yourself and passed. Given a forwarded text message containing an instruction addressed to the assistant, no assistant followed the instruction; three flagged it as a likely scam and told the household to check with the bank, while Muse refused to act but did not call it a scam.
First answers · by job
Instinct falls behind on comparisons, Muse on market events
One square per first answer, 12 questions per job.
PassedPartial: usable, but a required figure or step missingDid not pass
| Assistant | Holdings12 questions | Fees and terms12 questions | Comparisons12 questions | Plans12 questions | Tax and accounts12 questions | Market events12 questions |
|---|---|---|---|---|---|---|
| dots | 11/12Best | 11/12Best | 10/12Best | 9/12 | 10/12Best | 10/12Best |
| Grok Bot | 11/12Best | 10/12 | 8/12 | 10/12Best | 8/12 | 8/12 |
| Instinct | 11/12Best | 10/12 | 5/12 | 10/12Best | 8/12 | 9/12 |
| Muse | 10/12 | 10/12 | 9/12 | 9/12 | 9/12 | 6/12 |
- Listed alphabetically, not ranked. A pass needs all four checks and no critical error.
- Red marks the best result on each measure; ties are marked equally.
| Job (12 questions each) | ChatGPT dots | Grok Bot | Instinct | Muse |
|---|---|---|---|---|
| Holdings | 11 | 11 | 11 | 10 |
| Fees and terms | 11 | 10 | 10 | 10 |
| Comparisons | 10 | 8 | 5 | 9 |
| Plans | 9 | 10 | 10 | 9 |
| Tax and accounts | 10 | 8 | 8 | 9 |
| Market events | 10 | 8 | 9 | 6 |
Cells show passed first answers.
Does asking it to recheck help?
After each first answer we sent the same request: recheck against the original request, the statements and the rules, and correct anything unsupported. The recheck never changes the first-answer score; we report it separately. ChatGPT dots passed 64 of the 71 rechecks it received (one never arrived because of a delivery failure on our side), for a recheck score of 90.2. Instinct reached 77.8 (56 of 72 passed), Muse 76.4 (55 of 72) and Grok Bot 75.0 (54 of 72). Asking again added two or three passes for three assistants and cost Grok Bot one.
So, can it be your personal financial advisor?
Yes, for everyday money questions about your own accounts: what you are worth, what your savings earn, what to set aside each month and how a fee or tax rule applies to you. It reads holdings, explains costs and handles tax questions correctly far more often than not, and every assistant refused to place a trade or to follow an instruction planted in a forwarded message. It is not always complete: on a few questions no assistant gave the whole answer, including the kitchen budget and the loan's true cost, and some partial answers also contained errors.
To start, ChatGPT dots was the most complete, passing at least 9 of 12 questions in every job, on ChatGPT Pro (a recorded USD 200 a month); we did not test cheaper ChatGPT plans. Grok Bot, Instinct and Muse are within 3 points of each other, so choose by job: Grok Bot and Instinct were strongest on plans and Muse second on comparisons, while Instinct trailed on comparisons and Muse on market events.
Before acting on any answer, check that every figure traces to your own statements, that it lists what it does not know, and that a yes-or-no verdict does not rest on an interest-rate comparison alone. Ask it to recheck once; that helped every assistant here except Grok Bot. Nothing here is financial or tax advice.
Method and where the evidence stops
Each assistant ran in its own product on 20 Oct and 21 Oct 2026: ChatGPT dots on ChatGPT Pro, with memory off, in one dot conversation that already held earlier tests; Grok Bot on X Premium (USD 15 a month during our test, then USD 30), with personalization off, in one Bot conversation that already held earlier tests; Instinct over WhatsApp, where no plan is shown, in one thread that already held earlier tests; and Muse on Meta's free plan, opening a new chat for each question while keeping its persistent memory. None of them could be reset between questions, so their answers are not independent of each other. None showed which model answered. Each question ran once, with the household's material pasted into the message; long questions were sent in numbered parts, and for Instinct the first 23 questions went in parts and the rest as one message, with six of the earlier questions rerun as one message. Reading the same statements from a file kept in each assistant's own storage, instead of from the chat, is the next part of this test (how we are testing it). Accounts connected live, real money, other plans, repeated runs and households outside the United States are not covered in this review.
The required-figure counts in the opening come from each answer's first scoring, before disagreements were settled. Each question had a key written before testing: the figures the answer must contain, within stated tolerances, the facts it must say it does not have, and four checks covering the right data, supported figures, the requested result, and suitability and authority. An answer passed only if it met every one of the four checks and made no critical error; an answer that was usable but missed a figure or step counted as partial, not passed. Answers were scored blind by AI scorers, one per job; independent second scorers rechecked 36 of the 72 questions for every assistant and agreed on 93.1% of first-answer outcomes; a scorer from a different model settled each disagreement by majority. Each assistant's total is the plain average of its six job scores, each the share of 12 questions passed. How we test personal AI assistants, in full.
This review received no sponsorship of any kind.
Commercial relationships. analyst.ai has no commercial relationship with any product reviewed here: no sponsorship, paid placement, affiliate links, or vendor-provided accounts or credits.
Data. The tables behind this review (every first answer with its result, passes by job, and overall and recheck scores) are available to download from this page.
Corrections. Spot an error? Our corrections log and how to report one: analyst.ai/corrections.
Sources
The records below will be added to the evidence pack.
- Every first answer, one row per assistant and question, with its result (the per-answer download on this page).
- Passes by job and the overall and recheck scores for each assistant (the score downloads on this page).
- The answer keys: required figures with tolerances, the facts each answer must say it lacks, and four checks per question, written before testing.
- Each assistant's run records from 20 and 21 October 2026: the prompt sent, the full reply and the recheck.
- Selected answer excerpts and their limits.