The Google trick that kept a family on the wrong medication

Today's Tip: Dismantling A Forwarded "Shocking" Article In Under Two Minutes

When a friend forwards you a shocking article, your instinct is to Google the headline. That is the mistake.

Let me explain…

Last Tuesday my neighbor Mateo got a text from his brother-in-law Ruben. It was a link to an article claiming a common blood pressure medication had been "secretly recalled" and that thousands of older adults had already been hospitalized. Ruben had thrown out his own pills that morning.

Mateo is 69, retired 30 years from teaching high school shop class in Albuquerque. He is not a technology person. But his sister was on the phone in tears about her husband, and Mateo knew he had about ten minutes to figure out whether this was true before the family made a mistake nobody could take back.

He did what most of us do first… he Googled the headline. And the top of the page looked terrifying. Three different sites all repeating the same alarming claim. Same wording. Same photo. Same anonymous "insider."

Here is the thing about a Google search… it shows you what is popular. It does not show you what is true. A conspiracy site published at 6 a.m. and copied by two other conspiracy sites by 9 a.m. looks like three independent confirmations. It is one story wearing three coats.

So Mateo opened Grok on his phone instead. Grok is the AI tool made by Elon Musk's company xAI, and its one great strength is that it pulls information straight from the live web and traces where a story actually came from. He pasted the headline in with a specific request. Ninety seconds later he had his answer.

There was no FDA recall. The Associated Press had actually published a debunk piece four days earlier. The original story traced back to a single Facebook post from an account that had been flagged for medical misinformation twice this year. Mateo called his sister back. Ruben was back on his medication by lunch.

Here is the exact prompt Mateo used…

★ Copy-Paste Prompt

Works in Claude, Gemini & Grok

Someone forwarded me the article or claim below and I need to know if it is true before I do anything about it. Please act as a careful fact-checker and answer these five questions in order, using the live web:

1. Is this claim true, false, misleading, or unverified as of today? Give me one word first, then explain.

2. What is the earliest known source of this story — a website, a social media post, a press release, a study? Give me the name of the outlet and the publication date.

3. Has this been reported by any of the following outlets I trust: Reuters, Associated Press, the FDA, the CDC, my state health department, or a major newspaper? If yes, link to that reporting. If no, say so plainly.

4. What do those trusted outlets actually say about the claim — do they confirm it, contradict it, or add context I am missing?

5. In one sentence, what should I tell the person who forwarded this to me?

Do not invent sources. If you cannot find a real link from a real outlet, say "I could not verify this claim." Here is the article or claim:

[PASTE THE HEADLINE, LINK, OR SCREENSHOT TEXT HERE]

Why this works: You are not asking the AI to have an opinion. You are asking it to trace a chain of custody — who said it first, who confirmed it, and what the trusted outlets actually say. The instruction not to invent sources removes the AI's temptation to fill in gaps with made-up citations, which is the single biggest risk in this kind of task.

The magic is in question two. Almost every piece of misinformation that circulates on social media traces back to one original post from one unreliable source. When you ask an AI to find the earliest known version of a story, you strip off all the copies pretending to be independent sources. Suddenly a "widespread report" is one Facebook post from a stranger.

Question three is your safety net. If Reuters, the AP, or the FDA has not touched a health story that supposedly affects thousands of people, that alone tells you almost everything you need to know. Trusted outlets are boring on purpose. They wait until something is confirmed.

★ Today's "Aha!" Moment

Google shows you what is popular. Grok shows you where a story actually came from. Those are two completely different questions — and when someone is scared, you only want the answer to the second one.

I know what you are thinking… "If a chatbot can be wrong sometimes, why would I trust one to fact-check anything?" It is a fair question and honestly the right instinct to have.

Here is the answer. You are not asking the AI to be right. You are asking it to show you the receipts. The links in Grok's answer are the thing you actually trust — not the summary above them. If Grok links you to a Reuters article, click it. Read the first paragraph. That is your fact-check. The AI is just the fastest way to get to the source without wading through twelve pages of noise.

★ One Thing to Watch

Grok can still be wrong on stories that broke in the last few hours — and worse, it can produce a link that looks real but goes nowhere. Always click the link it gives you. If the link is broken, or if the outlet name is one you have never heard of, treat the claim as unverified until a name you recognize — Reuters, AP, FDA, your state health department — has confirmed it.

Next time a friend or family member sends you something that makes your stomach drop, do not answer right away. Do not throw out your pills. Do not call the family in a panic. Open Grok, paste the headline into the prompt above, and give it ninety seconds. Ninety seconds is nothing compared to what a bad decision could cost.

The people who forward you these stories are not trying to hurt you. They are scared, and they want you to be safe. The kindest thing you can do is send them back the truth.

★ ★ ★

Tool Spotlight ★

Grok

Site: grok.com

Made by: xAI

Cost: Free tier available

Grok is the tool to reach for the moment a question touches something happening now. Its one great skill is that it goes out and reads the live web while you wait — news that came out this morning, prices as of this hour, product recalls posted yesterday. Claude and Gemini can do this too, but Grok defaults to it and is faster at surfacing a working link than either of them.

The interface is plain on purpose. No pretty pictures, no long preamble, no "as an AI language model" throat-clearing. You ask a question, it gives you an answer with links. For fact-checking a forwarded article, verifying whether a store still stocks something, checking a company's news for the last 48 hours, or comparing two products by their most recent reviews, this is the one that gets you to the source fastest.

Best for: fact-checking a forwarded article, current events, product comparisons, checking recalls, news from the last 48 hours, verifying a claim with links to the source

★ ★ ★

★ Quick Wins

The One-Line Charity Check — Before you write a check to any organization, open Grok and type "Give me the current Charity Navigator or CharityWatch rating for [charity name], the percentage of donations that go to programs rather than administration, and a link to their most recent tax filing." Grok will pull the live rating and a link within seconds. Works in Grok most reliably; Claude and Gemini can do it but sometimes rely on older figures.

The "Is It Still Made?" Trick — Before you drive across town for a discontinued shampoo, coffee, or replacement part, ask Grok "Is [exact product name and size] still being manufactured and sold as of today? If it has been discontinued, what is the closest current replacement, and which national retailer stocks it new?" Two minutes at the kitchen table beats a wasted afternoon. Works best in Grok because of the live inventory check.

The Two-Minute Recall Check — When you hear about any recall — a car part, a food item, a household product — ask Grok "What is the current status of the [product] recall issued by [brand]? What are consumers being offered: refund, replacement, or repair? What is the deadline, and where do I claim it? Please include the link to the official recall notice." Grok will give you the exact steps and the deadline in the same answer. Works in Grok.

Skill Builder: Catching AI When It Makes Things Up

Here is the hard truth about every AI tool on the market today. They will, on occasion, make things up. Not on purpose. Not because they are broken. It is baked into how they work — they are pattern machines, and when a pattern runs out, they fill in the missing piece with something that sounds right. The technical name is a hallucination. It is the single most important thing you can learn to defend against.

The most common hallucinations are the ones that look most confident. A made-up phone number for a government office. A book title that does not exist by an author who does. A court case citation that reads exactly like a real one. A statistic that sounds plausible but has no source. If you accept these at face value, you can end up on the phone with the wrong number, quoting a nonexistent ruling, or forwarding a fake stat to your book club.

The defense is not to stop using AI. The defense is to change how you ask the question. Most hallucinations happen because the AI feels pressure to give you a complete answer. If you tell it upfront that "I don't know" is an acceptable response, the pressure goes away and the honest answer comes out. Save the prompt below and paste it before any factual question you actually need to be right about.

Try it side by side sometime. Ask any AI for a list of five books on a specific niche topic without this prompt, then ask again with it. The first list will have one or two invented titles buried in there. The second list will be shorter and honest about what the AI actually knows.

★ Anti-Hallucination Guardrails

Works in Claude, Gemini & Grok

I am going to ask you a factual question. Before you answer, please follow these four rules and confirm you understand them:

1. If you are not confident in the answer, say "I am not confident about this" and stop. Do not guess to fill the gap.

2. For every factual claim, name the source — the outlet, author, or organization — and the year it was published or updated.

3. Do not invent URLs, phone numbers, book titles, author names, court cases, or statistics. If you do not have a verified source, say "I do not have a verified source for this."

4. At the end of your answer, add a short section called "What I am less sure about" and list any part of your answer that could be wrong or outdated.

Confirm you will follow these rules, then answer this question:

[YOUR QUESTION HERE]

Why this works: You are giving the AI explicit permission to say "I don't know." Without that permission, it will invent an answer to seem helpful. The "What I am less sure about" section at the end is the most useful part — it flags exactly where you should double-check before acting.

Two other habits are worth building. Ask the same question a second time, phrased differently, and see if you get the same answer. If the two answers disagree, one of them is a guess. And never accept a URL, phone number, or citation without opening it yourself — those are the four things AI invents most often, and they are also the four things that look most like proof.

Once you have this habit, the whole tool gets more useful, not less. You stop treating AI like an oracle. You start treating it like a very fast, very well-read assistant who occasionally gets tired and starts guessing. That is a much more accurate picture of what it actually is.

The Google trick that kept a family on the wrong medication

Read More from the AI Advantage