ChatGPT Stock Analysis: Build a Research Note You Can Verify
A stock research summary can sound convincing while mixing three different things: what a company reported, what someone expects to happen, and what the AI inferred. Once those statements sit together in a fluent paragraph, it becomes easy to miss the distinctions.
A practical way to use ChatGPT for stock analysis is to build a research note in stages. Gather a small set of dated sources, extract individual claims, check them against the originals, and only then write a conclusion. The useful output is a record another person can follow, including the questions you could not answer.
Start with a question the evidence can answer
“Is this a good stock?” leaves too much undefined. Good at what price, over what period, and based on whose assumptions? Start with a narrower question, such as: “Did revenue growth come with an improvement in operating margin in the latest reported year?”
Write down the company, the reporting period, the comparison period, and the date of your research. If you want to understand a historical investment argument, also set an information cutoff. Do not treat a filing published after that cutoff as evidence that the information was publicly available beforehand.
Keep business analysis separate from the eventual investment decision. Finding evidence that a business improved does not, by itself, establish that its shares are attractively priced. Valuation requires additional inputs and assumptions, and an AI summary cannot decide whether an investment fits your circumstances.
Build a dated source pack
For the margin question, begin with the company’s annual report, the relevant earnings release, and any reconciliation needed to understand adjusted figures. Use the company’s investor relations site or the relevant regulator’s filing database. Save the document title, publication date, reporting period, and page or section containing each figure.
If a financial video prompted the research, include that source too. They Said Buy’s source-linked stock commentary connects quoted statements to the relevant moment in the original video. That gives you a starting point for checking the argument in context; the company’s own reports still source its reported financial results.
Label your materials S1, S2, and S3. Provide accessible documents or clearly identified excerpts, and retain the originals for your own checks. A URL in a prompt does not prove that the model read the page. Ask it to identify which supplied materials it could actually access and which it could not.
The UK’s Financial Conduct Authority explains in its guide “Using AI for investment research” that answers can be inaccurate or out of date, and recommends checking financial information against trusted original sources. A dated source pack makes that checking process manageable.
Ask for a claim ledger before a conclusion
A claim ledger is a simple list in which each entry records one statement and its evidence. It prevents a persuasive summary from hiding missing support. Start with this prompt, replacing the brackets with your own material:
Prompt 1: “Use only the supplied sources to investigate [question] for [company and period]. First, list which sources you could access. Then extract one claim per entry. For each claim, give its type, source ID, exact supporting passage or figures, page or section, publication date, reporting period, and any missing context. Label claims as reported fact, management forecast, commentator opinion, or your inference. Do not fill gaps from memory. If evidence is missing or conflicting, state that explicitly. Do not produce a buy or sell recommendation.”
These prompts are starting templates, not a tested accuracy guarantee. Treat every returned entry as unverified until you check it yourself. A page number or a quotation-shaped sentence can also be wrong.
For each entry, keep a short verification record:
- Claim: one statement, with its units and period.
- Type: reported result, forecast, opinion, or inference.
- Evidence: the original document or video and the precise location.
- Date: when the source was published and when you checked it.
- Status: unverified, verified against the source, contradicted, or unresolved.
“Verified” means that the source supports the statement as written. It does not mean a forecast will come true or that an investment will succeed.
Keep facts, interpretations and forecasts separate
Consider this invented example. Fictional company Harbor Tools reports revenue of $100 million for one year and $120 million for the next, with operating margins of 10% and 12%. Its management expects further growth, while a fictional commentator says the business is becoming more resilient. These are teaching examples, not real company results or quotations.
The reported figures support two calculations: revenue increased by 20%, and operating margin rose by 2 percentage points. A spreadsheet or calculator can verify both. Calling the margin change “2% growth” would confuse percentage points with relative change; the relative increase from 10% to 12% is 20%.
The other statements require different labels. Management’s expectation is a forecast. The commentator’s view about resilience is an interpretation. Neither becomes a reported fact just because it appears next to the financial figures.
Even correct arithmetic can compare the wrong things. Before accepting a calculation, check whether both values use the same currency, scale, period length, and accounting definition. Do not silently substitute an adjusted margin for a reported margin, or compare one quarter with a full year. If the definitions differ, preserve the discrepancy in the ledger.
A reasonable next question in this example is what caused the margin improvement. The numbers alone do not tell you whether it came from pricing, product mix, lower costs, or an unusual item. Keep that explanation unresolved until you find supporting evidence.
Verify what the commentator actually said
Public stock commentary helps you find an argument to investigate, but a shortened summary can lose its conditions. An opinion about the next quarter can look like a long-term prediction; a discussion of a business can look like a recommendation to buy its shares.
When a video introduces a stock idea, start with the original segment. Watch enough surrounding material to understand the subject, time horizon, and qualifications. Keep the video title, publication date, and relevant timestamp in your notes. Record what was said before asking the AI to interpret it.
Separate the speaker’s statement from your own conclusion. If you think a remark implies confidence in a company’s margins, label that as your interpretation unless the speaker explicitly makes that claim. A date and a source link show where a statement came from; they don’t prove the opinion is correct.
See more of our reporting in Google Top Stories, AI Overviews and AI Mode.
Use a second prompt to expose gaps
Once you have checked the individual claims, ask for a review focused on weaknesses rather than a more polished version of the same narrative:
Prompt 2: “Review this claim ledger against the supplied sources. Identify conclusions that go beyond the evidence, inconsistent periods or definitions, forecasts presented as facts, and missing information that could change the interpretation. For each issue, name the claim and explain what original evidence would resolve it. Do not invent a counterargument or mark an issue resolved without supporting evidence.”
This is a way to generate questions, not independent verification. The same model can repeat its earlier mistake. Open the relevant document yourself, recalculate important numbers, and leave a disputed item unresolved if you cannot settle it.
OpenAI’s “Use ChatGPT” guidance likewise recommends checking important numbers, dates, quotations, and claims before using an answer. A polished rewrite should come after those checks, so it doesn’t hide an uncertain foundation.
Turn checked claims into a short research memo
Finish with a compact memo that preserves the boundary between evidence and judgment. You can ask ChatGPT to draft it from your checked ledger:
Prompt 3: “Draft a brief research memo from this ledger. Include the research question and information cutoff, findings supported by checked sources, interpretations and their assumptions, unresolved questions, and the next evidence to obtain. Keep source IDs beside factual claims. Do not turn forecasts or unresolved items into established findings.”
In the Harbor Tools example, the supported finding is that the supplied figures show higher revenue and operating margin. The cause of the improvement remains open. A useful next step is to read the relevant management discussion and notes for an explanation, then test whether that explanation supports any claim about future results.
Date the memo and retain the ledger alongside it. When a new earnings release arrives, you can identify which claims changed instead of rewriting the entire story from memory. That is a practical role for AI stock analysis: helping organise a research process whose evidence, assumptions, and unanswered questions remain visible.
This article is educational and does not recommend any security.
You Might Also Like:
AI Portfolio Analysis Tools
Best Equity Research Software
Best AI Trading App For Beginners
