AI Investing Assistant: What It Can and Can't Do for You

Short answer: an AI investing assistant is useful when it can see your actual holdings and current prices instead of guessing them. Then it can tell you what's overweight, why your portfolio dropped last week, what risks you're carrying, and explain jargon in plain words. What it shouldn't do is make decisions for you. It isn't a licensed adviser, it can be wrong, and trades stay in your hands.

This guide covers where general chatbots go wrong with money, how a data-connected assistant works, which questions get useful answers, and how to check what it tells you.

What general chatbots get wrong about your money

A general-purpose chatbot is great at explaining what an ETF is or how the Sharpe ratio works. The trouble starts when the question is about your money.

• It can't see your portfolio. You describe your positions from memory, forget a few, misremember your average cost. The answer is built on a partial picture.
• Its prices are stale. The model knows the market as of its training data. Ask "what's Bitcoin at right now?" and, without live data, it either admits it doesn't know or quotes a number from months ago.
• It can make numbers up. Language models are fluent. Last year's fund return, a dividend date, a company's P/E: it can "remember" any of these wrongly and sound completely sure.
• The advice is generic. "Diversify and think long term" fits everyone, so it helps no one. Without your data it can't get more specific.

What a grounded assistant does differently

A grounded assistant doesn't recall facts. It fetches them through read-only tools. Ask about your Apple position and it first pulls your quantity, average cost and trade history, then does the math. Ask about market mood and it checks a fear and greed index rather than paraphrasing year-old headlines.

That last row matters. A good assistant can't change anything. Read-only access protects you from a model that "decides" to sell a position.

For the assistant to see everything, your holdings need to live in one place first. Here's how to do that across brokers and exchanges: track your stock and crypto portfolio.

Questions worth asking (with examples)

The best questions are specific and about your own data. Some prompts that work well:

About your portfolio

• "What share of my portfolio is crypto, and is that too much for me?"
• "Which three positions add the most risk?"
• "My portfolio is down 6% this week. What dragged it down?"

About a single asset

• "When did I buy NVDA, at what average price, and what's my gain now?"
• "How has ETH moved over the last 30 and 90 days?"
• "Are there recent technical signals on SOL, and what's their historical success rate?"

About the market and news

• "Is the market in fear or greed right now?"
• "What news came out about my stocks in the last few days?"

About taxes

• "How much gain have I realized this year, and roughly what tax will I owe?"

If you live in Spain, it helps to understand how that tax works in the first place: see crypto tax in Spain. The assistant gives an estimate; check your return with an adviser.

Questions to skip

• "What should I buy to double my money?" The answer will be either hedged or dangerous.
• "Where will Bitcoin be tomorrow?" Scenarios are possible. A precise forecast isn't.

The limits

It's not a licensed adviser

An AI assistant isn't regulated, isn't accountable for outcomes, and doesn't know your whole life: debts, plans, family. It can share a view and lay out scenarios, but the decision and the responsibility are yours.

It can be wrong

Even with real data, a model can misread a figure, mix up two tickers, or reach a conclusion that's too bold. Technical signals with good statistics also fail regularly; see what that looks like on historical data in do chart patterns work?.

It doesn't execute trades

The assistant only reads. It can't buy, sell or move money, and that's how it should be.

Privacy

To answer, the assistant sends the relevant data (positions, amounts, your question) to a language model. Before using any service like this, find out what data leaves the app, which model provider receives it, whether chat history is stored and whether you can delete it. Only give exchanges read-only API keys.

How to check AI answers: a checklist

1. Do the numbers match the app? Compare the weight, gain and position value with your portfolio screen.
2. Is there a source? Price, headline, signal: where did it come from? If a number isn't from your data, treat it as opinion.
3. Are the dates current? A two-month-old headline doesn't explain yesterday's drop.
4. Are facts separated from opinion? A good answer says "the data shows X, my view is Y, confidence is moderate."
5. Any promises? "Will definitely go up" or "guaranteed" is a reason to distrust the whole answer.
6. Does it fit your risk profile? Telling a cautious investor to add 30% altcoins is a mistake.
7. Ask again, differently. If the answer flips with the wording, the conclusion is shaky.
8. Big decision? Check with a human. Taxes and large amounts deserve a professional's eye.

Using it with a risk profile

The same portfolio can be fine for one person and reckless for another. So it's worth answering a short questionnaire once: your goal, time horizon, how you'd react to a drawdown, experience, income stability and attitude to crypto. That gives you a profile from cautious to bold, with target shares for stocks, crypto and cash.

From there the assistant compares your real portfolio against that profile. "Is this too risky?" stops being abstract. You get something like "crypto is 45% of your portfolio against a 15% target; a 30% crypto drop would cost you this many euros." Update the profile when life changes: a new job, a child, buying a home.

Analyzing YouTube bloggers' videos

Finance YouTube is a use case of its own. A creator says "I'm buying ETH from 2,800, target 3,500," and you can't tell whether that's worth listening to.

Here the assistant needs three things: a summary of the video, a list of the concrete calls (asset, direction, entry, target, stop, time frame) and the creator's track record, meaning how many of their past calls actually worked. With that you can ask:

• "How does this ETH call compare with my own position?"
• "What's this creator's hit rate on crypto over the last year?"
• "If I followed this call, what would my ETH weight become?"

It's a good filter for loud but empty predictions. The same idea applies to watching what big funds and insiders do rather than what they say: see 13F filings and insider trading.

How Signae helps

Signae has an AI assistant built on Claude that answers from your data, not from memory. Here's what it does today:

• Reads your data through read-only tools: your portfolio, your position and trade history in an asset, price change over 7, 30, 90 and 365 days, the fear and greed index for crypto and US stocks, technical signals with their historical success rate, recent news, and a Spanish income tax (IRPF) summary for the year. It can't change or buy anything.
• Risk questionnaire with six questions: a profile from cautious to bold and target shares of stocks, crypto and cash.
• Portfolio review: fit with your profile, strengths, risks and concrete suggestions. You can turn on a weekly or monthly review and get a notification when it's ready.
• Portfolio builder for an amount and a goal, backtested over the past year and compared with SPY and BTC.
• "Ask AI" on a blogger video: the assistant gets the video summary, every call and the creator's hit rate.
• News digest for your assets, in your language.
• Lite and Pro modes: plain words for beginners, terms and numbers for experienced investors.

Every answer is labeled as AI-generated. There are limits on question length and how often you can ask.

Signae is in closed beta. Request an invite and we'll send you one.