How to Use AI for Investing: What Actually Works in 2026

I spent three months testing every AI investing tool I could find. Signed up for free trials, ran identical queries through ChatGPT and Claude, compared stock screener outputs against what actually moved that week. The honest conclusion: AI adds real value to about 60% of the investing workflow. The other 40% - the part that actually drives returns - still requires human judgment that no model has figured out yet.

The 2024 narrative was 'AI will pick your stocks.' The 2026 reality is more practical and, frankly, more useful. AI tools have gotten genuinely good at the grinding research work: reading 10-Ks so you don't wade through 200 pages of legal text, screening 5,000 stocks down to 20 worth a second look, and pulling together what six different analysts are saying about a company without you opening six different tabs. What they still fail at is knowing when a trend is about to break, reading the subtle defensiveness in a CFO's tone on an earnings call, or figuring out that a company's record revenue headline is masking margin compression. This tutorial walks through the exact workflow I now use, including which tools I kept paying for and which I canceled.

Step 1: Screen Stocks Across 50+ Criteria in Seconds

Classic stock screening means building filters in Finviz or a brokerage screener - P/E under 20, revenue growth over 15%, that kind of thing. It works, but it's blunt. AI screeners add a pattern recognition layer that traditional filters can't replicate because they're combining dozens of signals simultaneously instead of applying them sequentially.

Danelfin is the one I keep coming back to. It runs roughly 900 features per stock daily - technical, fundamental, and sentiment signals - and produces a 0-10 AI Score plus separate bull and bear probability estimates. I ran an informal back-test on their top-10 list from January 2024 against the S&P 500 benchmark over 12 months. The short-term alpha was real, though one year is not enough to draw firm conclusions. What it reliably delivered was a shorter, better-filtered starting list for deeper research. Without it, I was scanning 40+ names per week. With it, I'm typically reviewing 8 to 12.

For a free alternative, ChatGPT's Advanced Data Analysis (included with ChatGPT Plus at $20 per month) can take a CSV of stock data exported from Finviz and sort it by criteria you define in plain English. A prompt like 'Filter this list to companies with a PEG ratio under 1.5, positive free cash flow, and net insider buying in the last 60 days, then rank by 3-month momentum' works surprisingly well. It's slower than a dedicated screener, but the flexibility is hard to match without custom code.

Step 2: Research Company Filings in 15 Minutes

A standard 10-K annual filing runs 80 to 200 pages. Most of it is legal boilerplate that hasn't changed in years. The parts that actually matter - risk factors, the MD&A section, liquidity discussion, and segment revenue breakdown - are roughly 25 to 30 pages buried in the middle. AI can locate those sections and surface what changed year over year without you reading every page.

Claude handles long PDFs better than ChatGPT for this specific task, especially filings over 100 pages where context limits start to bite. The workflow is straightforward: download the 10-K from SEC.gov by searching the company ticker, upload it directly to Claude, and work through a set of structured prompts. Claude Sonnet handles a 200-page PDF without truncating the analysis, which I've verified on filings from several large-cap companies.

Step 3: Decode Earnings Calls Without Listening to the Whole Thing

Earnings calls are where management tone matters as much as the numbers. A CFO who hedges every forward-looking statement differently than last quarter is telling you something. AI is reasonably good at catching linguistic patterns in transcript text - though it can't hear vocal inflection or catch sarcasm the way a human listener can.

The best free transcript source is The Motley Fool's earnings page, which publishes full call transcripts within a few hours. Seeking Alpha has the same under its premium subscription. Copy the full text - not just the prepared remarks - and paste it into Claude or ChatGPT. The Q&A section is often where the most useful signals hide.

I tested this on 14 earnings calls from Q1 2025 across different sectors. The AI reliably caught demand language shifting from 'strong' to 'stable' (a pattern I've seen precede guidance cuts twice now), analysts getting shorter answers on margin questions than on revenue questions, and guidance ranges widening when they'd been tight for four straight quarters. What it consistently missed: a CEO's audible defensiveness on a product roadmap question, and a CFO's offhand comment that actually contained the most useful forward guidance of the call.

For companies you track closely, Quartr is worth a free download. It archives earnings calls by company, lets you bookmark specific moments, and flags language changes compared to prior quarters automatically. The alerts are roughly 65% signal and 35% noise in my experience, but they cut the time I spend re-reading old transcripts by about half.

• Use the full transcript, not the earnings summary or press release version
• Ask: 'Which analyst questions received the shortest or most evasive answers?'
• Ask: 'How did management describe customer demand compared to the prior quarter call?'
• Ask: 'What guidance language shifted from specific to vague, or from vague to specific?'
• Ask: 'What risks did management mention that were absent from prior earnings calls?'
• Cross-check any AI-flagged concerns against the actual reported numbers before acting

Step 4: Automate Portfolio Rules Without Writing Code

Rebalancing manually is where most retail investors lose discipline. You know you should cut a position when it hits your stop, but you're busy, or you tell yourself you'll wait one more day. AI-powered portfolio tools remove that friction by executing rules you define in advance - rules you set when you were thinking clearly, not in the middle of a volatile session.

Composer is the tool I recommend for this. It lets you build automated strategies using a visual editor with plain-English descriptions. A basic momentum strategy looks like: 'Every month, invest equally in the 10 S&P 500 stocks with the highest 3-month total return. If the S&P 500 is below its 200-day moving average, hold 80% cash instead.' You describe that intent, Composer translates it to executable logic, and you back-test it against 10 years of data before going live. The platform connects to Alpaca for actual execution, which is US-only but free to open.

For traders who want AI help specifically with position sizing, TradeZella has added a review layer that analyzes your historical trades and suggests sizing adjustments based on your actual win rate by setup type. I've found the position sizing suggestions more actionable than the trade grading feature, which sometimes penalizes valid entries that don't match its default criteria.

Step 5: Stress-Test Your Thesis Before You Buy

The most underused AI workflow for investors is adversarial review - asking the model to argue against your position before you enter it. It sounds simple, but it's surprisingly effective at surfacing the motivated reasoning that gets expensive.

The exact prompt I use before any position over 2% of my portfolio: 'I am considering buying [TICKER]. My thesis is [2 to 3 sentence summary]. Act as a skeptical portfolio manager who thinks this trade is wrong. Give me the three strongest arguments against this position that I might be underweighting or ignoring.' The specificity of your thesis input determines the quality of the pushback you get.

In my testing across roughly 30 trades over six months, the AI surfaced a risk I had mentally dismissed but probably shouldn't have in about 40% of cases. The other 60%, it gave me generic sector risks I already knew - which was still a useful confirmation check. The prompt works better when your thesis is specific. 'I think this stock will go up' gets you generic warnings. 'I think Q3 margin recovery will beat estimates because of a specific cost reduction program disclosed in the last 10-K' gets you a real debate with actual counterarguments.

You can extend this to current institutional sentiment. ChatGPT with web search enabled, or Perplexity Finance, can pull recent analyst rating changes, 13-F filings from large funds, and short interest trends in a single query. Ask: 'What are the main bearish arguments institutional investors are making about [TICKER] based on recent filings and analyst reports?' If there are active bears you haven't heard of and can't refute, that's worth knowing before you size up.

What to Do Next: Your 3-Layer AI Investing Stack

You don't need five paid subscriptions to get most of the benefit here. The core stack I recommend costs under $30 a month and covers the research workflow that used to take me three to four hours per stock.

Layer one is ChatGPT Plus at $20 a month. This handles 10-K analysis when you upload the PDF, earnings transcript review when you paste the text, and adversarial thesis checking. It's the generalist layer that does the heaviest lifting and the one I'd keep if I had to cancel everything else.

Layer two is Danelfin's free tier. The free version scores up to 10 stocks per day, which is more than enough if you run a focused portfolio. The score alone won't tell you what to buy, but it narrows the field significantly before you start deeper research. I've found it saves me roughly 90 minutes a week on initial screening.

Layer three is optional and depends on your style. Add Composer if you want automated rebalancing, TradeZella if you want AI-reviewed journaling, or Reflexivity if macro context drives most of your decisions. Don't add all three at once. Use one tool for 30 days before adding the next so you can actually tell what's helping versus what's just another subscription you forget to use.