Misinformation, AI and your responsibility

0:00 Misinformation and verification

It's eleven at night. A video lands in your DMs. A famous business founder, looking straight at the camera, says she's backing a new crypto app and your followers can double their money this week. It looks real. It sounds real. Your thumb is already on the share button. Stop. In this lecture you'll learn why misinformation is now your professional problem, a ten-minute verification workflow, the special care zones, and exactly what to do when someone fakes you.

0:34 Three risks

Why does this matter so much now? Generative AI makes convincing text, images, audio and video cheap to produce at scale. That creates three risks for you. You might publish something false because a tool hallucinated it. You might amplify a fake someone else made. Or your own face and voice might be used in a scam. Creators have real influence. Sharing a fake, even with good intentions, can hurt people, damage your trust, and in some cases break consumer protection, defamation or false information laws, as well as platform policies.

1:14 The five checks

Think of verification like airport security. Nobody is accused of anything. Every bag goes through the same scanner, and most pass in seconds. The point is consistency. Your scanner has five checks. Source: who posted it first? Corroboration: are reputable outlets or official bodies saying the same? Reverse search: did this image appear years ago in another context? Provenance: are there Content Credentials or platform AI labels? And emotion: is this designed to make you angry or afraid? That last one is a signal to slow down.

1:52 Clues vs evidence

A quick word on visual clues. Odd hands, warped text, mismatched shadows, unnatural blinking. These used to be reliable. They're becoming less reliable every few months as models improve. So treat them as hints, never proof. The same goes for AI detection tools. They produce false positives and false negatives. Use them as one signal among several. Provenance is different. When a Content Credentials manifest is present and valid, it tells you who signed the file and, often, whether AI was used. But if there's no manifest, that proves nothing, because most platforms strip metadata on upload.

2:34 Special care zones

Some topics need extra care. Health and medical claims, especially around outbreaks or products. Financial content, like investment tips and guaranteed returns. Elections and politics, where many countries and platforms have strict rules on synthetic content. Religion and communal issues, where in Pakistan, the Gulf and elsewhere false content can cause real-world harm and carries serious legal risk. And breaking news, where early information is often wrong. In these zones, raise your bar: two independent confirmations, or you don't post.

3:09 Example 1: UK finance creator

First example. A UK personal finance creator gets that crypto video. She runs the five checks. The account that first posted it is three days old. A reverse search finds the same background in a real interview from two years ago. There's no coverage from any reputable outlet, and the founder's official channel has a pinned warning about deepfake scams. She doesn't share it, not even with a question mark. Instead, she posts a short video warning her audience about deepfake investment scams, without linking to the fake.

3:47 Example 2 (illustrative): Dubai brand deepfaked

Second example, a realistic business scenario with illustrative details. A Dubai skincare brand discovers an AI-cloned video of its founder, speaking fluent Arabic, offering a fake giveaway that asks for card details. Here's the playbook they run the same day. Document everything: links, screenshots, times. Report to each platform using impersonation and synthetic media tools. Post a warning on all official channels saying the founder never asks for payment details in giveaways. Escalate to the relevant cybercrime reporting channel because fraud is involved. And pin a how to verify it's really us post. Illustratively, the fake accounts came down within days, and customer service handled far fewer confused messages because the warning went out early.

4:37 Watch me do it: flood photo

Watch me do it. I've got a photo claiming to show flooding at a well-known Riyadh mall today. First, source. I search the caption in quotes; the earliest post is from an anonymous account with no history. Second, reverse search. I run it through a reverse image search and find the same photo attached to a flood in a different country, three years ago. Third, provenance. I open it in a Content Credentials checker; no manifest, which proves nothing on its own. Fourth, corroboration. No reputable outlet, no official statement. Decision: do not share. I log it in our verification sheet with the date, the claim, the reverse search result and my initials. Total time, about seven minutes.

5:28 Common mistakes

Common mistakes. Believing it because it looks real. Resharing with the words is this true, which still spreads it to thousands of people. Trusting a detector score as if it were a verdict. Quietly deleting your own error instead of correcting it. And asking a chatbot whether something is true and accepting the answer. A model can help you plan your searches, but it can invent sources. Make it list what to check, then open every source yourself.

6:02 Corrections and prevention

When you get it wrong, and everyone eventually does, correct it promptly and visibly. Edit or remove the original, and post the correction in the same place, to the same audience. Aim to do this within a day of finding the error. And prepare for being faked before it happens. Keep a pinned post that says how to verify it's really you, and lists what you will never do, like ask for money in DMs. Turn on two-factor authentication everywhere, and limit who can post as you.

6:40 Make it a team habit

If you work in a team, make verification a shared habit, not a heroic individual effort. Keep one verification log: date, link, claim, first source, reverse search result, provenance, corroboration, decision, and who checked it. It takes a minute per item. Over a few months, that log becomes gold. You'll see which kinds of accounts keep pushing fakes, which topics trip you up, and how long checks really take. It also protects you. If a client or regulator asks why you shared something, you can show exactly what you checked and when. And when you use a chatbot to plan your searches, paste its suggestions into the log too, so everyone can see what the AI said and what the sources actually showed.

7:33 Recap + try this now

Recap. AI makes fakes cheap, and your reach makes them spread. Run the five checks every time: source, corroboration, reverse search, provenance, emotion. Treat visual clues and detectors as hints. Take extra care with health, finance, politics, religion and breaking news. If you're faked, document, report, warn, escalate and pin. Try this now: write and pin your how to verify it's really me post today, and set up the verification log from the lesson so your team logs every check. Next up, copyright and training data.