How to Spot AI-Generated Fake News Before You Share It

The Scale of the Problem Is Bigger Than You Think

Here is a number that should make you pause: in a controlled study by iProov in 2025, researchers showed participants a mix of real and AI-generated images, videos, and audio clips. Only 0.1% of people correctly identified every piece of synthetic media. Not 10%. Not 1%. One-tenth of one percent.

That means if you are reading this and feeling confident about your ability to spot fakes, the statistics say you are almost certainly wrong. Do not feel bad. The technology has outpaced human perception in ways that were hard to predict even two years ago.

The numbers tell a stark story. Deepfake files surged from 500,000 in 2023 to a projected 8 million in 2025 — a growth rate of roughly 900% per year. In Q2 2025 alone, the verification platform Resemble tracked 487 discrete deepfake incidents, a 312% year-over-year increase. Voice cloning has crossed what researchers call the “indistinguishable threshold” — a few seconds of source audio now produces synthetic speech complete with natural breathing, intonation, and emotional inflection that human listeners cannot reliably distinguish from the real thing.

And it is not just sophisticated state actors doing this. Free and low-cost tools have made deepfake creation accessible to anyone with a laptop. The barrier between “could a nation-state do this” and “could a teenager in a basement do this” has effectively collapsed.

This guide is not about understanding the technology behind deepfakes. It is about giving you a practical, usable system for evaluating content before you believe it, share it, or act on it. Think of it as a pre-flight checklist for information.

The Five-Point Verification Checklist

When you encounter a claim, image, video, or audio clip that triggers a strong emotional reaction — outrage, fear, disbelief, vindication — that emotional spike is your signal to slow down. Misinformation is engineered to bypass your critical thinking by hijacking your emotions. The stronger the reaction, the more carefully you should verify.

Here is the checklist. It takes two to five minutes. That is a small investment against the cost of spreading something false.

1. Check the source. Where did this content originate? Not who shared it with you, but where it first appeared. A screenshot of a headline is not a source. A link to an article on a domain you have never heard of is not a credible source. Look for the original reporting. If you cannot find the original, that is a red flag. Established news organizations with editorial standards — AP, Reuters, BBC, NPR — are not infallible, but they have correction mechanisms that social media accounts do not.

2. Reverse search the media. For images, right-click and select “Search image with Google Lens,” or upload it directly at images.google.com. For video, use the InVID browser extension to extract key frames and search them individually. This tells you whether the image existed before the current claim, whether it has been taken out of context, or whether it was generated from scratch. If Google has no record of the image before yesterday, and it supposedly shows a major world event, be skeptical.

3. Look for corroboration. Is any other credible outlet reporting this? If a story is true and significant, multiple independent organizations will cover it. If only one obscure site is reporting something explosive, it is far more likely to be fabricated or wildly distorted. Check Google News, not just Google search — it filters for recognized news sources.

4. Examine the content for AI artifacts. This is getting harder as the technology improves, but current-generation fakes still leave traces. We will cover specific visual, audio, and text indicators in the next sections.

5. Check what fact-checkers say. Google’s Fact Check Tools (toolbox.google.com/factcheck) aggregates findings from organizations like PolitiFact, Snopes, and Full Fact. AFP’s Fact Check database covers international claims. If someone has already investigated the claim, you do not need to reinvent the wheel.

How to Spot Fake Images, Video, and Audio

Each media type has its own tells. None of these indicators are foolproof on their own — the best fakes will pass any single test. But taken together, they give you a reliable detection framework.

Images. AI-generated images have improved dramatically, but several patterns persist. Look for an overly cinematic sheen — a polished, almost hyper-real quality that real smartphone photos rarely have. Check backgrounds for impossible geometry: signs with unreadable text, buildings with mismatched windows, fences that merge into trees. Examine hands and jewelry closely — AI still struggles with ring placement, watch faces, and finger proportions when hands interact with objects. Skin texture in AI portraits is often unnaturally smooth, lacking the pores, blemishes, and uneven tones of real photographs.

Video. Deepfake video has two persistent weaknesses. First, temporal consistency — watch the edges of the face, especially where it meets the hairline and ears. In deepfakes, these boundaries sometimes flicker or shift between frames in ways that real faces do not. Second, lip sync under stress — when a deepfake subject turns their head to an extreme angle or speaks rapidly, the generated face often shows subtle distortion or misalignment. Slow the video to 0.25x speed and these artifacts become more visible.

Audio. Cloned voices now sound remarkably natural in short clips. The weaknesses emerge in longer segments: inconsistent breathing patterns (real people breathe mid-sentence; some AI voices do not), flat emotional range (the voice sounds expressive but the emotional transitions feel abrupt rather than gradual), and background audio anomalies (the voice might sound like a studio recording while the context suggests a phone call or press conference). If audio quality seems unusually clean for the supposed context, that itself is suspicious.

Text. AI-generated text is the hardest to detect because language models produce genuinely fluent prose. Look for vague sourcing (“experts say” without naming experts, “studies show” without linking studies), hedged precision (specific-sounding numbers that cannot be verified), and emotional manipulation patterns (language designed to provoke outrage rather than inform). AI-generated articles also tend to be suspiciously comprehensive — covering every aspect of a topic smoothly without the gaps, biases, and rough edges that characterize human reporting.

Free Tools That Do the Heavy Lifting

You do not need to rely on your eyes and ears alone. A growing ecosystem of free and accessible verification tools can help.

A few notes on using these tools effectively. No single detector is reliable enough to use alone. AI detectors produce false positives and false negatives at meaningful rates. Use them as one input alongside your own assessment and source verification. If three different signals all suggest the content is synthetic, you can be fairly confident. If one tool flags it but everything else checks out, investigate further rather than drawing a conclusion.

The C2PA Content Credentials standard deserves special attention. Major camera manufacturers, news organizations, and platforms including Adobe, Nikon, Sony, the BBC, and Microsoft are embedding cryptographic provenance data into photos and videos at the point of capture. This creates a verifiable chain of custody — you can confirm not just that an image exists, but that it was taken by a specific camera at a specific time and has not been materially altered. Adoption is still in its early stages, but this will likely become the most reliable verification method within the next two to three years.

The Psychology of Why We Fall for It

Understanding the technical detection methods is necessary but not sufficient. The reason misinformation spreads is not primarily technological — it is psychological. If you understand the cognitive patterns that make you vulnerable, you can build better defenses than any tool provides.

Confirmation bias. You are dramatically more likely to share content that confirms what you already believe. A fake study “proving” your political position correct feels true in a way that a real study challenging it does not. This is the single largest vulnerability that misinformation exploits. The uncomfortable discipline: apply your most rigorous verification to the content you most want to be true.

Authority bias. A deepfake video of a recognized figure saying something provocative gets shared at exponentially higher rates than the same statement from an unknown person. The AI-generated content exploits the trust you have already built with that figure. When a public figure appears to say something surprising, verify through their official channels before assuming it is real.

Urgency bias. “Breaking news,” “just revealed,” “share before they take it down” — this language is engineered to make you act before you think. Real breaking news will still be breaking in ten minutes. If you wait and verify, you lose nothing. If you share and it is false, you become part of the distribution network for misinformation.

The illusory truth effect. Content you have seen before feels more true than content you are encountering for the first time, regardless of its actual accuracy. This is why misinformation campaigns repeat the same claims across multiple channels and formats. By the third time you encounter a false claim, your brain processes it as familiar, and familiarity feels like truth. Be especially skeptical of claims you have “heard before” but never verified at the source.

The World Economic Forum’s March 2026 analysis on cognitive manipulation and AI emphasizes that the next generation of disinformation will not just create convincing fakes — it will be designed to exploit these specific cognitive biases at scale, personalizing the manipulation to each individual’s psychological profile based on their online behavior.

Frequently Asked Questions