How to Fact-Check ChatGPT Before a Customer Sees It

The comment found me in the Publix parking lot, engine off, waiting on a curbside order. Somebody had left it under my reel about editing every AI draft before it goes out. One word I won't print, then this: "don't let a silicone brain think for you." I laughed out loud. Then I sat there a minute longer than I meant to, because he's half right.

If you want to fact-check ChatGPT, start by accepting this: it gets things wrong, and it does it with a straight face. When 22 public broadcasters tested more than 3,000 answers from ChatGPT, Copilot, Gemini and Perplexity, 45% had at least one significant problem. So before anything it writes reaches a customer, I check four things: the numbers, the names and links, anything that sounds like a promise, and how the whole thing sounds read out loud. That's my entire system. It takes a few minutes, and it's why I can use AI every single day and still sleep fine.

Why it sounds so sure when it's wrong

I've spent 25 years around confident people, in direct sales rooms and corporate boardrooms. The ones who cost a company money were the people who always had an answer, ready or not.

ChatGPT is that guy. It's built to write the most likely next words, so it hands you a smooth, finished paragraph whether it actually knows the answer or not. Sometimes the facts are old. Sometimes it fills a gap with something that sounds right. In that same broadcaster study, 20% of answers had major accuracy problems, including details that were made up or out of date, and 31% had problems with where the information came from.

Even the company that makes it tells you, right there on the screen, that it can make mistakes. I take them at their word.

And plenty of us are using it now. In a February 2026 survey, Pew found that 37% of adults 50 to 64 and 19% of adults 65 and older have used ChatGPT. That number only goes one direction, which leaves one real question: how do you use it without putting your name on something wrong?

The Air Canada story every business owner should know

In November 2022, a man in British Columbia lost his grandmother. He went to Air Canada's website to book a flight, and the chatbot told him he could buy a regular ticket and apply for the bereavement discount afterward, within 90 days. That wasn't the airline's policy. When he asked for the money back, Air Canada said no.

It went to a tribunal. Air Canada's position, in the tribunal's own words, amounted to suggesting "the chatbot is a separate legal entity that is responsible for its own actions." The tribunal called that "a remarkable submission" and ruled against the airline in February 2024, ordering it to pay about $650 in damages plus interest and fees.

Six hundred and fifty dollars won't sink an airline. What stuck with me is the reasoning. The chatbot was part of their website, so what it said, they said. That goes for you and me too. Your website chat, your automatic text replies, the email AI drafted for you at 11 at night. All of it goes out under your name.

Small business owners already feel this in their gut. In a May 2026 Bluehost survey of 350 owners, accuracy and quality topped the list of AI worries at 32%, and only 6% said they highly trust AI with their brand voice. Good. Stay a little suspicious. You just need a routine so suspicious doesn't turn into stuck.

How to fact-check ChatGPT: my four checks

I run a big part of my business with AI helpers now. Every one of them works under the same rule I do: no number goes out without a link to where it came from. Before anything AI drafted reaches a customer, a client or my email list, it goes through this.

1. Every number gets traced to its original source. Prices, percentages, dates, limits, anything with a digit in it. I ask ChatGPT where it got the number, then I go to that source myself (the government page, the company's own site, the actual study) and I look at the date. A real number from 2023 is still the wrong number today.
2. Every name, quote and link gets clicked. If it mentions a study, a rule or a person, I look it up. If it gives me a link, I open it. AI is famous for links that go nowhere and quotes nobody ever said.
3. Anything that sounds like a promise, I rewrite myself. Prices, refunds, guarantees, results, health, what someone could earn. AI loves to round up because it sounds friendly. It will tell a customer "you'll save hundreds" without blinking. Those lines come out, or I write them word by word. Call it the Air Canada check.
4. Then I read it out loud. My ear catches what my eyes skip: a claim that's a little too shiny, a sentence I would never say. I wrote about that habit in how to make ChatGPT sound like you, and it does double duty here.

If something fails a check and I can't confirm it in a couple of minutes, it gets cut. I'd rather send a shorter email than a wrong one.

If you're newer to AI and want the starter prompts and safety tips I give people first, they're in my $7 AI Hacks guide.

What I let it do without a second look

It's a short list. AI can brainstorm subject lines for me, turn my rambling voice notes into a first draft, summarize a long document I already have, and clean up formatting. If it gets one of those wrong, I lose a minute and nobody else ever sees it.

Everything else, it drafts and I decide. Back in August I called AI my research assistant for finding hidden savings, and that's still the job description. And I said it in May when people asked whether AI would replace network marketers: AI cannot be trusted on its own. Nothing I've seen since has changed my mind.

Anything with a customer's name on it goes past my eyes before it goes anywhere. So does anything touching money or health. Those are the places where a confident wrong answer costs you trust, and trust is the one thing in my business I can't automate.

Back in the parking lot

About then my groceries came out to the car, and I put the phone face down on the seat. I'd already decided what I thought of that comment.

He's right that a silicone brain shouldn't do your thinking. Mine doesn't. It does my first drafts and a lot of my digging, and then I check its work the way I'd check any new hire's first week. That's how a woman who restarted her business at 57 gets to move fast without handing her reputation to a machine.

So use it. Just don't give it the last word.

If you want to start using AI this week without the overwhelm, grab AI Hacks for $7: an 18-page guide with 9 copy-and-paste prompts and a 7-day plan, so your first win happens this week instead of someday.

Questions I get about checking AI

Can you trust ChatGPT for business?

For drafts and ideas, yes. For facts, only after you check them. I treat every number and every claim it gives me as unconfirmed until I've seen the original source with my own eyes.

Why does ChatGPT make things up?

It's built to predict the most likely next words, so it writes a confident answer even when it doesn't have a real one. Some of what it learned is also out of date. That's why I trace numbers back to the source and always check the date.

Is there a prompt that stops ChatGPT from making mistakes?

No prompt fixes it completely. Asking it to show where each fact came from and to tell you when it isn't sure does help, and I keep that in my own instructions. You still have to click the links and check.

What should I never let AI send without checking?

Anything about a price, a refund, a guarantee, a result, your health or someone's income. Anything with a customer's name on it, too. Those are the lines that cost you trust, and sometimes money.