Key Takeaways

  • Source it, date it, file it, challenge it, list it
  • SEDAR+ filings beat AI summaries on any disagreement
  • Unverified claims are opinions — weight them that way

The Most Dangerous AI Answer Isn't the Wrong One — It's the Plausible One

Every guide in this cluster has ended with the same warning: verify the output. This final guide turns that warning into a method. AI stock research fails in predictable ways — figures from the wrong quarter presented as current, a confident claim with no source behind it, a ratio calculated from a share price that stopped existing months ago. None of these errors announce themselves. They read exactly like the correct answers.

So before any AI-assisted research touches real money, run the research itself through these five prompts. Think of it as the cluster's closing audit — the companion to where this series started, with Can ChatGPT Pick Winning Stocks? What AI Gets Right and Wrong.

Prompt 1: Flag Every Claim That Has No Source

"Go through your previous answer about [company] and list every factual claim you made — figures, dates, events, comparisons. Beside each one, state what source would prove it: a specific filing, report, or official document. If you cannot name a checkable source for a claim, mark it UNSOURCED."

This prompt changes the power dynamic of the whole conversation. An AI that has to show its receipts separates, in one step, into two piles: claims you can verify this afternoon, and claims that were fluent filler. Invest only on the first pile.

Prompt 2: Interrogate the Dates

"For each figure in your analysis — share price, earnings, ratios, dividend yield — state what date or reporting period it appears to come from. Flag anything you are unsure is current, anything from a different period than the others, and anything you estimated or derived rather than took from a source."

Mixed-period analysis is the most common silent error in AI stock research: this year's price divided by last year's earnings, producing a ratio for a company that never existed. The dates prompt drags the seams into the open where you can see them.

Prompt 3: Take It to the Filing

"Here is the relevant section of the company's actual report: [paste from the earnings release or financial statements]. Compare it line by line with your earlier summary of the same figures. List every difference, no matter how small, and correct your summary to match the document."

In Canada, the documents themselves are public: company filings, financial statements and material news releases live on SEDAR+, free for anyone to read. The workflow this prompt teaches — AI summary in one hand, primary document in the other, differences listed — is the closest thing retail investing has to a superpower. It takes ten minutes and it catches the errors that matter.

Prompt 4: Force the Other Side of the Argument

"You previously laid out the case for this stock. Now argue the strongest possible case against it, using the same figures. What would a smart skeptic say is wrong with my reasoning? Which of my assumptions, if it failed, would hurt the investment case most?"

AI assistants are agreeable by design — show them a thesis and they'll help you build it. That agreeableness is a genuine risk when the thesis is yours. This prompt rents you a skeptic: not to talk you out of every investment, but to make sure the counter-argument exists somewhere other than the market's future behaviour.

Prompt 5: End With the Two Lists

"Summarize this whole analysis as two lists. List one: facts I have verified against a primary document, with the document named. List two: claims and assumptions I have NOT verified. Do not move anything into list one unless I confirmed it in this conversation."

Pin those two lists to the top of your research notes — in the general workflow from How to Use ChatGPT to Analyze Stocks Before Buying: 6 Powerful AI Prompts, and in every specialist guide that followed it: earnings, valuation, dividends, ETFs, portfolio risk, and Canadian account taxes. List two is not a failure; it's a to-do list. The only failure is confusing the two lists.

What Fact-Checking Can't Guarantee

It can't make a good company cheap, a risky portfolio safe, or next quarter predictable. A fully verified analysis can still lose money — that's investing, not a malfunction. What verification changes is the kind of risk you're taking: from "the AI sounded right" to "I checked the figures, I know my assumptions, and I sized the position for being wrong." One of those is a process. The other is a hope.

The Complete ChatGPT-for-Investors Series

The Twikup Take

This cluster gave you a research department: prompts that read earnings, judge prices, test dividends, compare funds, map risk and untangle Canadian accounts. This final guide gives that department an editor. Source it, date it, file it, challenge it, list it — five prompts, run every time, and the AI works for you instead of the other way around.

Image: Illustrative image.

Sources