You can’t blame your saw when your shelf comes out crooked. You’re supposed to know how the tool cuts.

Chatbots have a unique wrinkle: your saw doesn't insist the cut was straight. A chatbot does. It will make a mistake, claim it double-checked, and sound completely confident while doing it.

Take the iShares Core High Dividend ETF (HDV). Fidelity showed an August ex-dividend date instead of the usual September schedule, and the financial aggregator sites contradicted each other — some still listing the fund as quarterly, others as monthly. Was it a special payout? An error?

How much you learned from Gemini depended on how you asked. Today I asked Gemini the same question two different ways.

Prompt 1: Asking for confirmation

User: I bought shares of HDV. Fidelity shows an ex-dividend date of 08/19/26, but the next published dividend is September. Is this accurate?

Gemini: Yes, that notification is accurate... While HDV historically paid strictly on a quarterly schedule (March, June, September, December), ETF issuers occasionally distribute monthly or special mid-quarter payouts based on the underlying cash flows and portfolio rebalancing.

Result: correct dates, correct amount — but the wrong reason. Cash flows and rebalancing had nothing to do with it. And the answer that mattered never surfaced.

Prompt 2: Asking for the issuer's record

User: I bought shares of HDV. Fidelity shows an ex-dividend date of 08/19/26, but the next published dividend is September. Where is this published by the issuer? Give me the link.

Gemini: The dividend is accurate. iShares changed the distribution frequency for HDV from quarterly to monthly... Issuer source: ishares.com - Shareholder letters

Result: it named the actual event — a formal frequency change — and handed over a page to check. From there, BlackRock's own distribution schedule spells it out: HDV; Changed Frequency from Quarterly to Monthly: June 16, 2026, with ex-dividend on the third Wednesday of every month. August 19 was the third Wednesday.

The takeaway

Both got the dates right. Only one got the reason right.

Asking whether something makes sense invites a chatbot to speculate — and it will, fluently. Asking where it's published sends it looking. One returns a story, the other returns something you can verify.

These tools can be extraordinarily useful. They're just easy enough to use badly that a mistake isn't obvious.

Ears