Christian Perry is the CEO of Undetectable AI and TruthScan, a deepfake-detection platform.
Deepfakes no longer belong to sci-fi. They are now part of our daily lives. From an estimated 500,000 deepfakes online in 2023, the number surged to around eight million in 2025, an annual growth of nearly 900%. It has spread across industries, costing businesses and enterprises millions.
This year, 63% of cybersecurity leaders express their growing concern that AI is being used to generate deepfakes, posing a threat to digital trust. They are right to be concerned: Around 53% of businesses in the U.S. and U.K. report deepfake scam attempts as of 2024.
But just because there is a growing public awareness doesn’t mean businesses are any closer to solving the challenge. Only 0.1% of people can reliably detect them, according to an iProov study.
In the coming months, you’ll likely encounter more sophisticated deepfakes. What can your business do? As the CEO of a deepfake-detection platform, I have seen a lot of deepfakes and have developed a few strategies that can help you shift how you approach images and videos online. Here are three tips that can help to spot deepfakes:
Conscious Viewing
Many people see an image or video and will jump to a conclusion: “What I’m seeing is AI-generated,” or “What I’m seeing is real.”
There are a few problems with this line of reasoning. First, it’s an example of a bifurcation fallacy, in which one assumes that something is either A or B—real or fake. Even before AI existed, content could be edited, posted without context or altered in ways that misrepresent the truth.
Today, this way of thinking is (arguably) even more dangerous because AI-generated content appears so convincing. The first line of defense against deepfake content is what I call conscious viewing.
First, be aware that anything you see might not tell the whole truth. When looking at an image or video, approach what you see like this:
• What I’m viewing might be real, edited, lacking context or generated by AI.
• The source of the content I am viewing might be mostly credible, sometimes credible or not credible at all. Even credible sources are not necessarily infallible.
• If I jump to an immediate conclusion only based on limited information, I might harm myself or others.
Conscious viewing should apply to every piece of information you consume. The key is to avoid binary judgments with limited information.
As a business decision-maker, you probably want to make good decisions. Good decisions lead to good outcomes and can only be made with good information and sound judgment. When consuming any piece of information, you want to form a mental truth-testing matrix.
If you’d like to see a more complex version of this, check out the hypothesis-testing matrix that inspired the process. Simply put, consider that the truth of what you’re seeing might be A (real), B (not real), C (partially real) or D (real but lacking context), etc.
Focused Analysis
While conscious viewing is the foundation, focused analysis of content should be a part of that process.
The digital world moves fast. Attention spans are shrinking. In 2012, the average American adult shifted focus after 74 seconds; today, that’s down to 47 seconds. Attention to detail is your first line of defense.
Deepfake content usually has visual tells. Some easy-to-identify markers of deepfake content are:
• Gibberish Or Nonsensical Letters Or Text: Often appearing on signage in the background of a video or image, on clothes, or on-screen text.
• Illogical Physics: People or objects moving in impossible ways.
• Unstable Pixels: Many generative video tools demonstrate unstable pixels. It looks like visual buzziness or fuzz (some bad actors may try to hide this by lowering the quality of a video or image to make it intentionally look low-quality).
Those are just a few easily spotted markers I see in deepfake videos. Of course, sometimes there are also watermarks, but bad actors find ways to remove them.
If you want to go deeper, you can slow the video down. Artifacts in facial movement, lip sync and hand geometry become more visible at reduced playback rates, in my experience. Watch the edges. Hair, teeth, ears and fingers are where most generators fail. Look for blending at the jawline, teeth that shift shape between frames or fingers that merge or multiply.
Technical Analysis
As the iProov study cited above shows, almost no one is equipped to identify deepfake content. This is where using tools can help.
There are a few advantages that technical tools have over the human eye. First is metadata analysis. Deepfake detection tools can analyze the data inside an image or video file and spot invisible watermarks. Metadata might be lost if you’re downloading a video from social media, but if someone sends you a raw video or image file, that data should be there (unless it was forcibly removed—but that can also be detected).
These tools work best, though, when paired with other strategies like the ones mentioned above. They should also be trained on large datasets that include both real and AI videos, given the speed at which deepfakes are evolving.
Conclusion
Beyond spotting deepfakes, businesses should also build a system to be ready for their appearance.
No single method (e.g., conscious viewing, focused analysis or technical tools) is enough on its own. Use all three. Slow down, question what you see and make decisions based on verified information rather than assumptions.
Start with people: Train employees to slow down and question anomalies, and verify requests through secondary channels. Build verification protocols into workflows.
In the modern world, treat deepfakes as a primary vector of fraud. Because we are living in a world where seeing is no longer believing. It is not enough to merely react to deepfakes; we need to operationalize our defense against them.
Forbes Technology Council is an invitation-only community for world-class CIOs, CTOs and technology executives. Do I qualify?
