Deepfakes have been in the news for years, and the conversation tends to follow a predictable arc. First, there is a shocking example. Then there is alarm. Then there is a discussion about what tech companies, governments, or platforms should do. Then the story fades until the next shocking example.

What gets left out of this cycle is the practical question: what can you actually do, right now, as a person who encounters video and audio content online?

Start with the honest acknowledgment that deepfakes have gotten very good. Early deepfakes had obvious tells: blurry edges around the face, unnatural blinking patterns, weird artifacts around hair. Most of that is gone now. A well-produced deepfake of a public figure can be genuinely difficult to spot with your eyes, which means visual inspection alone is no longer a reliable method.

What is more reliable is context. When you see a video clip that seems surprising, ask a few questions before you share or believe it. Where did this clip come from? Is the original source credible? Is this the kind of thing this person would actually say? Has any mainstream outlet reported on it independently? The clip exists in a context, and that context usually tells you more than the clip itself.

Sound is often easier to detect than video. AI voice clones still have subtle tells: slight timing mismatches between the audio and the visual, an uncanny smoothness to speech patterns that sounds just slightly off, and a tendency to flatten out the natural variation in tone that real human speech contains. Someone who pauses, stumbles slightly, or changes volume mid-sentence sounds human. Someone who sounds like they recorded a perfect take in one shot sounds suspicious. Trust your gut if something sounds a little too clean.

Cognitura’s research on media literacy for general readers, not just professionals or researchers, suggests that the most effective protection is not technical detection tools but rather habits of skepticism about surprising claims regardless of the format they arrive in. If the video is making you feel something strongly, whether outrage, fear, or excitement, that is exactly when to slow down and check.

Governments and platforms are slowly moving on regulation. The European Union’s AI Act has provisions around synthetic media labeling. Several US states have passed laws specifically targeting deepfakes in political advertising and non-consensual intimate images. These are not nothing, but enforcement is hard and the technology moves faster than legislation has historically been able to keep up with.

The realistic near-term situation is that you will encounter more convincing synthetic media over the next few years, not less. The tools to create it are getting cheaper and more accessible every month. The platforms are playing catch-up. The practical response is not to become paranoid about everything you see, but to treat surprising video and audio claims the same way a good journalist would: interesting, maybe important, but not confirmed until there is additional independent evidence backing it up.