Learn · 8 min read
AI investing: what AI is genuinely good at, and where it misleads
AI has changed how quickly an investor can understand a company. It has not changed what makes a decision good. The useful framing is narrow: AI is excellent at compressing and organising evidence, and unreliable as the source of the evidence itself. Everything below follows from that split.
Where AI genuinely helps
Compressing long documents
Filings, transcripts and press releases are long, repetitive and structured. Summarising them into the handful of facts that matter is exactly what language models do well, and it removes hours of reading before a decision.
Normalising numbers into a comparable shape
Different companies label the same line items differently. Restating results into a consistent format — revenue growth, margin direction, cash conversion, share count — makes two businesses genuinely comparable.
Framing scenarios instead of one prediction
A bear, base and bull case built from stated assumptions is far more honest than a single target. The value is in seeing which assumption the outcome depends on, not in the number itself.
Explaining at the right level
The same quarter can be explained to a first-year investor or to someone who reads cash flow statements for a living. Adjusting the explanation, not the facts, is a real strength.
Surfacing candidates and questions
Screening a wide universe and generating the questions a thesis has to answer is cheap for AI and expensive for a person. Confirming the answers is the part you still own.
The four ways it misleads
Stale or approximate data
- What goes wrong:
- A general chat assistant may answer from training data rather than from today's market. Prices, multiples and estimates go out of date within days.
- The guard:
- Insist on the as-of date for every number. If the tool cannot state when the figure was measured, treat it as unverified.
Fabricated figures
- What goes wrong:
- A confident-sounding EPS, margin or growth rate can be produced without a source behind it. The fluency of the answer is not evidence of the number.
- The guard:
- Ask for the source for each figure and check one or two against the filing or the provider. Numbers that cannot be traced should be discarded, not discounted.
False confidence
- What goes wrong:
- Uncertainty tends to disappear in the phrasing. A weak inference and a well-supported one read the same way.
- The guard:
- Require the assumptions to be stated explicitly. If changing one assumption flips the conclusion, that is the finding.
No accountability for the decision
- What goes wrong:
- AI output is research, not advice, and it carries no responsibility for the outcome. Treating it as a recommendation transfers the judgement without transferring the risk.
- The guard:
- Keep the thesis in your own words. If you cannot restate the case without the output in front of you, you have not done the work yet.
A verification checklist
- Every figure carries a source and an as-of date.
- Reported and adjusted numbers are labelled separately, never blended.
- Per-share figures are consistent with the current share count after any split or buyback.
- Forecasts state their assumptions, and the scenarios differ because those assumptions differ.
- The bear case is specific enough to be tested, not a token paragraph.
- The conclusion still holds when you check the two numbers it most depends on.
Where behaviour and tooling meet
AI makes it easier to produce a confident-looking conclusion, which makes overconfidence cheaper to manufacture. A tool that can generate a thesis in seconds can also generate the feeling of having done research in seconds — a different thing entirely.
The defence is the same as it is without AI: write the thesis in your own words, state what would prove you wrong, and size the position so you can survive being wrong. AI should shorten the reading, not shorten the thinking.
Ticker-Terminal is a research tool, and every AI output it produces is generated from market data for informational purposes only — not investment advice.
The mechanics are covered in the seven-step AI research workflow, which AI is best for stock research and AI research vs answer engines.
Continue with trading psychology and market cycles.
Try AI research with sourced data
Ticker-Terminal builds research reports, valuation and 12-month scenarios from provider-verified fundamentals, with the data source and freshness shown. Free plan includes three AI research analyses a week.
Free guide series
Get the “how to research a stock” checklist
Two emails. The 7-step checklist now, then the next guide in the series — starting with Which AI is best for stock research.
- The 7-step “how to research a stock” checklist, sent immediately
- The next recommended guide in the series, two days later
- Occasional new guides only — no daily market noise, unsubscribe in one click