Spotting Fakes and Misinformation

Visual and audio "tells" can sometimes reveal a fake, but they're getting harder to spot every year, so the most reliable question is not "does this look real?" but "where did this come from, and who else confirms it?"

Images: where to look

Video: where to look

• Lips and words slightly out of sync, especially on sounds like "p," "b," and "m."
• Blinking too rarely, too regularly, or oddly.
• Edges of the face shimmering or blurring when the head turns, or when a hand passes in front of it.
• Objects that change between frames: a necklace that disappears, a mug that becomes a glass.
• Physics that feel off: water, hair, or fabric moving strangely; people walking through things.
• Very short clips with no before or after. Many AI video tools make clips only seconds long.

Audio: what to listen for

• Flat or oddly placed emotion: panic that sounds a little too even, or laughter in the wrong spot.
• Unnatural breathing: no breaths at all, or breaths at strange moments.
• Too-clean sound: no background noise when there should be, or background noise that loops.
• Mispronounced names or family words the real person would say correctly.
• Avoiding back-and-forth: short, scripted-sounding answers; changing the subject when you ask something personal.

The honest truth about tells

Every list of tells has an expiration date. Each new generation of AI tools fixes the flaws people learned to spot. Research has repeatedly found that ordinary viewers, and even experts, often can't reliably tell high-quality AI images or voices from real ones. That's not a reason to give up; it's a reason to lean on the habits in Sections 4 and 6, which don't depend on spotting flaws at all.

Voice-clone scams combine a copied voice with an invented emergency and a demand for fast, secret payment, and they succeed not because people are foolish but because the scam is designed to overwhelm careful thinking with fear and love.

The scale of the problem

Source: FBI Internet Crime Complaint Center (IC3), 2025 annual report, as summarized by AARP. These are only reported losses; many victims never report.

How the scam unfolds

1. Research. The scammer finds family names, relationships, and a voice sample, often from public social media videos.
2. The hook. A call comes in, sometimes late at night. A familiar voice: "Grandma? It's me. I'm in trouble." Crying makes voices harder to judge.
3. The emergency. A car accident, an arrest, a hospital, a kidnapping, stranded abroad.
4. The handoff. A second person, a "lawyer," "police officer," or "bail bondsman," takes over to give instructions. This person is usually a live human, not AI.
5. The secret. "Please don't tell Mom and Dad." Or a claimed "gag order." Secrecy stops you from calling someone who would spot the scam.
6. The payment. Gift cards, wire transfers, cryptocurrency or bitcoin ATMs, payment apps, or cash handed to a "courier" who comes to your door.
7. The second round. If you pay, they often call back: "The judge raised the bail." Victims who pay once are frequently targeted again.

Variations to know

• Fake kidnapping: a screaming "child" in the background and a demand for ransom.
• The boss call: a cloned executive's voice telling an employee to wire money or buy gift cards.
• The bank call: a "fraud department" voice asking you to "move your money to a safe account."
• Voicemails and voice notes as well as live calls.
• Fake video calls using a deepfaked face as well as a voice (Lesson 4, Section 3).

What to say

A real grandchild in real trouble will understand a two-minute check. A scammer will push back hard, and that pushback is itself the answer.

Why smart people fall for it

Victims are often embarrassed, and that shame keeps many from reporting. But these scams are built by professionals who rehearse scripts and exploit the best parts of us: love, loyalty, and the instinct to protect family. Talking about them openly, without blame, is one of the best protections a family has.

AI makes it cheap to produce convincing fake articles, fake product reviews, and entire networks of fake people online, so a claim appearing in many places, or a product with thousands of glowing reviews, is no longer good evidence on its own.

Fake news sites

AI can generate hundreds of articles a day. Researchers who track misinformation have identified well over a thousand websites publishing largely AI-generated "news" with little or no human oversight, often with generic, trustworthy-sounding names like "Daily Times Report" or "Metro Herald News."

Warning signs:

• No named reporters, or reporter names with no history anywhere else.
• No "About Us," no physical address, no way to contact an editor.
• Huge numbers of articles on every topic, many oddly worded or repetitive.
• Leftover AI phrases, like an article that accidentally includes "As an AI language model…" or "I cannot fulfill this request."
• A name that imitates a real outlet with a small difference (an extra word, a different ending like ".co" instead of ".com").

Fake reviews

AI can write a thousand varied, natural-sounding five-star reviews in minutes. In the U.S., the Federal Trade Commission adopted a rule in 2024 that bans fake reviews and testimonials, explicitly including AI-generated ones, and allows penalties for companies that knowingly buy or sell them. Enforcement is ongoing, but fake reviews remain common.

Fake accounts and bot networks

AI-powered accounts can post, reply, and argue like real people, around the clock. They're used to make fringe ideas look popular, to stir up anger between groups, to promote scams, and to build trust before a romance or investment scam.

Warning signs:

• A recently created account with a generic or AI-looking profile photo.
• Posting at inhuman volume or at all hours.
• Replies that are fluent but oddly generic, or that don't quite respond to what was said.
• Many accounts posting nearly identical messages at the same time.
• A stranger who quickly becomes warm, flattering, and then mentions an "investment opportunity" or asks for help with money.

Why it matters beyond your wallet

When fake crowds flood online spaces, it becomes harder for everyone to tell what real people actually think. That affects elections, public health, and trust in each other. Every time you pause before sharing, you're protecting more than yourself.

Instead of studying a suspicious page closely to decide whether to trust it, professional fact-checkers quickly leave it and open new tabs to see what other sources say about it; this "lateral reading" is faster and far more accurate, and anyone can do it.

What researchers discovered

In a well-known study published in 2017, Stanford researchers compared how three groups judged unfamiliar websites: professional fact-checkers, history professors, and top university students. The professors and students read carefully down the page (vertically), studying the design, the "About" page, and the logos, and were often fooled by slick sites. The fact-checkers barely read the page at all. They opened new tabs and searched what others said about the source. They reached accurate conclusions faster.

The SIFT method: four moves

Researcher Mike Caulfield developed a simple, widely taught method called SIFT:

1. S – Stop. Before reading, reacting, or sharing, pause. Ask: do I know this source? What's my goal here?
2. I – Investigate the source. Open a new tab and search the website's or person's name. Who are they? What's their track record? Who funds them?
3. F – Find better coverage. Search the claim itself. Are reliable outlets reporting it? If a major event is only on one obscure site, be very skeptical.
4. T – Trace to the original. Follow quotes, statistics, photos, and videos back to where they first appeared. Claims often get distorted as they're passed along.

Quick tools anyone can use

Can AI help check?

Yes, carefully. A chatbot with web search can quickly gather what several sources say about a claim. But as Lesson 3 explained, AI can also hallucinate sources. Use it as a starting point, then open the sources yourself.

Technology is being built to label real and AI-made content, through hidden watermarks, tamper-resistant "content credentials," and AI detectors, and these tools are genuinely useful clues, but none of them is complete or foolproof, so they add to your judgment rather than replace it.

Three kinds of tools

Where things stand in 2026

• Growing adoption. Major AI companies, camera makers, and platforms have joined the content credentials effort. Some phones, including Google's recent Pixel models, can attach credentials to photos and videos taken with the built-in camera.
• Checking is getting easier. In 2026, Google added the ability to ask whether an image carries its SynthID watermark through Google Lens in Search and Chrome, and in the Gemini app. Several other AI companies have adopted the same watermarking.
• Labels on social media. Some platforms now show "AI info" or "Made with AI" labels when they detect watermarks or credentials, and some show authenticity labels for verified camera photos.
• Big gaps remain. Many AI tools, especially open-source ones anyone can run at home, add no watermark at all. And a simple screenshot often strips credentials away.

How to read the results

The strongest protections against AI fakes are simple human habits (a family safe word, a call-back rule, a pause before paying or sharing, and open conversations without shame) because they work no matter how realistic the fakes become.

Setting up a family safe word

1. Choose something memorable but not guessable. Not a pet's name, street, or birthday that could be found online. An odd combination works well: "purple canoe," "Aunt Ruth's biscuits."
2. Share it in person or by phone, never by text, email, or social media.
3. Include everyone who might be impersonated or targeted: grandparents, parents, grown children, teenagers, and caregivers.
4. Agree on the rule: any call, message, or video asking for money or urgent help must include the safe word, or you hang up and call back.
5. Practice it once. It feels silly, and that's fine. Practice makes it automatic under stress.
6. Change it if it's ever used in a situation where someone else might have overheard.

Alternative: a "challenge question" only the real person could answer, such as "What did we eat at Dad's 70th birthday?" Avoid anything that's been posted in photos online.

Five more simple defenses

Protecting someone you love

If you're helping an older parent, a neighbor, or a relative who lives alone:

• Walk through the grandchild-call scam together, calmly, as "something going around."
• Put key phone numbers on the fridge: family members, their bank, and a trusted neighbor.
• Ask their bank about adding a trusted contact to their accounts, who can be called if unusual activity appears.
• Consider call-blocking features on their phone, and silence calls from unknown numbers if that works for them.
• Check in regularly. Isolation makes people more vulnerable; connection makes them safer.