The Visual Fact Check: How to Spot Deepfakes, Hidden Ads and Digital Fakes Without Forensic Software

Imagine this: a video shows up in your family group chat. A famous politician, looking worn out, makes a surprising announcement. Within an hour, it’s been shared hundreds of times. Your uncle is angry. Your cousin claims it’s fake. Who’s right, and how can you tell?

A few years ago, the best advice was to “look closely.” Now, that doesn’t really help. In one study, people trying to tell real faces from AI-generated ones got it right only 48.2% of the time, slightly below the 50% you would expect from guessing, and they even rated the fake faces as more trustworthy than the real ones (Synthetic face perception study). A meta-analysis of 56 papers with 86,155 participants found that people’s overall accuracy at spotting deepfakes was not reliably better than chance (Human deepfake detection meta-analysis). We also tend to think we’re better at it than we actually are (Deepfake overconfidence study).

This article suggests a new approach. Rather than just relying on what you see, focus on asking the right questions. The tips below are based on advice from professional fact-checkers, peer-reviewed research, and the AOFIRS course lesson on searching news, articles and blogs (AOFIRS course source note). All you need is a phone or a browser.

Why fake stories win the race

False news isn’t just common; it spreads faster. MIT researchers looked at about 126,000 stories shared on Twitter from 2006 to 2017 and found that false stories were about 70% more likely to be retweeted than true ones (False news diffusion study). Fake stories are often more surprising and emotional, which makes them more appealing to our brains.

But there’s some good news. The main issue isn’t that people want to believe false stories; it’s that they share them without stopping to think (Reasoning and misinformation study). Just asking people to think about whether something is true before sharing it can cut down on the spread of false headlines (Accuracy prompt research). Pausing creates space to check the source before sharing.

Figure 1. Research shows the limits of visual detection

Note. Mean accuracy reported by each study. Sources: Nightingale & Farid (2022), 48.2% for real versus AI-synthesized faces; Diel et al. (2024), 55.5% pooled across studies (95% CI 48.9–62.1); Köbis et al. (2021); Groh et al. (2022). Chart adapted by AOFIRS.

Four questions that beat a deepfake

Think of checking information like climbing a ladder. The first step, looking for obvious mistakes, is the weakest. Each step you take beyond that makes your check stronger.

1. Does anything look off?

It’s still worth taking a quick look. MIT researchers recommend checking details like the face, skin, eyes, glasses, facial hair, moles, blinking, and lip movements (MIT deepfake detection guidance). Also consider whether shadows are consistent with the lighting. Inconsistent shadows can suggest manipulation, but multiple light sources, reflections, and perspective need to be considered before drawing a conclusion (Inconsistent shadow analysis). Keep in mind, though, that a video that looks perfect doesn’t prove it’s real. Fakes are improving all the time.

2. Does it make sense?

Context can reveal misleading captions, recycled footage, or implausible claims. Would this person actually say this, in this place, at this time? Does the weather, season, background, or language on signs match what’s claimed? What is the video trying to get you to do: panic, skip voting, or send money? If it urges you to act quickly, that’s a classic warning sign.

3. Where did it come from?

Take a screenshot and use a reverse image search like Google Lens or TinEye. You might find the same image was posted years ago about something else. Also, see if trusted news outlets are reporting the same story. For example, when an AI-generated image of an explosion near the Pentagon spread in May 2023, stock indices dipped briefly. No other photos, videos or witness accounts backed it up, and the Pentagon Force Protection Agency and the local fire department said publicly that no incident had taken place (Pentagon fake image reporting).

4. Can I confirm it another way?

This is the most reliable step. In 2024, the engineering firm Arup confirmed that an employee in Hong Kong had joined a video call with people who appeared to be the company’s chief financial officer and colleagues. All of them were deepfake re-creations, and the employee sent HK$200 million (about US$25.6 million) to scammers (Arup deepfake fraud reporting). A call back on a number already known to be genuine could have exposed the scam. The same advice applies if you get a panicked message from a “relative” asking for money: hang up and call them directly. Some families even agree on a code word for emergencies.

Remember: you can withhold trust or action without declaring something fake. If the evidence is insufficient, call it unverified.

What about AI labels and Content Credentials?

Some cameras, editing apps and AI image generators now attach provenance information, often called Content Credentials, based on an open technical standard from the Coalition for Content Provenance and Authenticity (C2PA provenance standard). If a file carries them, a free inspection tool can show what made the file and how it was edited. Treat this as one more clue for question 3, not a verdict. The U.S. National Institute of Standards and Technology notes several limits (NIST synthetic content report). A digital signature shows who vouched for the metadata, not that the metadata is true. Metadata is often stripped when files are shared. Watermark robustness varies; some can be removed, degraded, or spoofed. And a missing label proves nothing on its own. The same report says automated “AI detectors” often fail on new generators and on compressed or resized files. A detector score should never replace the four questions above.

The ads that do not look like ads

Not all misleading content is completely false. Sometimes, it’s just trying to sell you something without making it obvious. “Native advertising” is paid content that looks like regular articles. In one study, only 7% of readers realized a native article was actually an ad (Native advertising recognition study). Stanford researchers found that 82% of middle-school students couldn’t spot a clearly labeled “sponsored content” item as advertising (Stanford civic reasoning research). And when European regulators looked at 576 social media influencers, 97% had posted commercial content, but only one in five consistently told their followers it was an ad (EU influencer disclosure investigation).

Figure 2. Advertising recognition and disclosure findings

Research shows that readers and social media users can overlook commercial content when advertising is presented in a format that resembles ordinary content.

Note. Sources: Wojdynski & Evans (2016); McGrew et al. (2017); European Commission (2024). Chart adapted by AOFIRS.

Here is how to catch it:

• Hunt for the label. Look above the headline, beside the author’s name, and at the very bottom for words like “sponsored,” “partner content,” “paid post,” “#ad,” or “paid partnership.”
• Follow the links. If every recommendation links to a shop or comes with a discount code, someone is probably earning a commission.
• Pay attention to what’s missing. Real reviews and real journalism usually mention drawbacks and include different viewpoints. Ads rarely do.
• Copy a sentence from the article and paste it into a search engine with quotation marks. Repeated wording may reflect a shared press release, syndication, or copying. It does not provide independent corroboration.

In the United States, the Federal Trade Commission says ads can’t be disguised as something else, and influencers must clearly say when they’re paid or get free products (FTC native advertising policy; FTC influencer disclosure guidance). The FTC has also banned fake reviews and testimonials (FTC fake review rule). But these rules only work if people follow them, so it’s still important to check for yourself.

When the crowd is not really a crowd

Sometimes the trick is not a single fake story but a fake impression: that “everyone” thinks something, or that “scientists disagree” when they largely do not. Researchers studied a covert campaign run by South Korea’s intelligence service during the 2012 presidential election. They found that the fake accounts gave themselves away through traces of coordination among them, not through the features of any single account (Coordinated disinformation study). Decades ago, a tobacco company memo put the strategy bluntly: “Doubt is our product” (Tobacco industry source document). Later, internal documents showed how the sugar industry funded research that steered attention away from sugar’s health risks (Sugar industry document study).

Be cautious of groups with neutral, public-sounding names if it’s hard to find out who funds them. Try searching the group’s name along with the word “funding”; this often shows who is really behind it. For example, when Stanford researchers asked college students to check a policy website run by an industry-linked group, only 6% figured out who was actually behind it (Stanford civic reasoning research).

Check your own blind spots

The hardest bias to spot is your own. We all look for evidence that supports what we already believe (Confirmation bias research). We’re also more likely to believe things just because we’ve seen them before (Repeated exposure study). Search engines can make this worse by showing results based on our location, language, and search history. The AOFIRS course suggests searching the same topic in another language or changing the country setting in a news search to see a different perspective (AOFIRS course source note).

Table 1. Three habits that keep your own bias in check

The good news: these skills can be learned

You don’t need a degree to learn these skills. When college students took a short online course in “lateral reading” (leaving a website to see what others say about it), the number who checked sources this way jumped from 3 out of 87 to 67 out of 87 (Lateral reading course study). After a 15-minute online game in which players tried making fake news, they rated fake posts as less reliable, while their ratings of real news stayed about the same (Misinformation game study). Short videos that explain common tricks also helped, even when shown as YouTube ads (Prebunking video study).

Your two-minute checklist

1. Pause. Notice if you feel angry, scared, or thrilled. That is your cue to slow down.
2. Ask who is behind it. Search for the source on other sites rather than trusting its own “About” page.
3. Find the original. Where and when did it first appear? Check the date, and use the Wayback Machine if a page has changed.
4. Look for other coverage. Look for independent reporting or primary confirmation. Early reports may be limited, so missing coverage alone does not prove a fake.
5. Ask who benefits. Is someone selling, lobbying, or campaigning?
6. Confirm another way before acting on any request for money, passwords, or urgent action.

More people are worried about fake content online. In 2026, 62% of people across 48 markets said they were concerned about telling what is real from what is fake online (Reuters digital news report). But worrying isn’t enough. Asking the right questions is what really helps. So, the next time something surprising pops up in your feed, you won’t have to guess; you’ll know what to ask.

For the full evidence, statistics, case studies, and bias guide, see the companion AOFIRS research report, visual fact-checking research report.

If you verify media professionally and need frame-by-frame analysis, metadata checks and forensic tools, continue with the investigator-level guide, professional deepfake verification guide.

Conclusion: Verify the Source Before You React

You do not need to identify every visual artifact to make a safer decision. Pause, examine the context, trace the original source, check commercial interests, and seek independent confirmation. For urgent requests involving money, passwords, or personal information, use a contact channel you already know is genuine.

Visual clues, labels, and detector scores can support an investigation, but none settles authenticity on its own. If you cannot verify a claim, keep it labeled as unverified and avoid amplifying it. A useful fact check explains the evidence, its limits, and what remains unknown.

Copyright Statement

Copyright © 2026 Association of Internet Research Specialists (AOFIRS). All rights reserved. All rights in this article, including its text, tables, and adapted charts, belong to AOFIRS. No part of it may be reproduced, distributed, or transmitted in any form or by any means without the prior written permission of AOFIRS, except for brief quotations with clear attribution as permitted by copyright law. Third-party works cited remain the property of their respective owners.

References

1. Association of Internet Research Specialists. (n.d.). Lesson 31: Search in news, articles & blogs [Unpublished course material, Chapters 1–17]. AOFIRS.
2. Bond, S. (2023, May 22). Fake viral images of an explosion at the Pentagon were probably created by AI. NPR. Pentagon fake image reporting
3. Breakstone, J., Smith, M., Connors, P., Ortega, T., Kerr, D., & Wineburg, S. (2021). Lateral reading: College students learn to critically evaluate internet sources in an online course. Harvard Kennedy School (HKS) Misinformation Review, 2(1). Lateral reading course study
4. Brown & Williamson Tobacco Corporation. (1969). Smoking and health proposal [Internal memorandum, Bates No. 690010951–690010959]. Truth Tobacco Industry Documents, University of California, San Francisco. Tobacco industry source document
5. Coalition for Content Provenance and Authenticity. (n.d.). About C2PA. Retrieved November 4, 2026, from C2PA provenance standard
6. Diel, A., Lalgi, T., Schröter, I. C., MacDorman, K. F., Teufel, M., & Bäuerle, A. (2024). Human performance in detecting deepfakes: A systematic review and meta-analysis of 56 papers. Computers in Human Behavior Reports, 16, Article 100538. Human deepfake detection meta-analysis
7. European Commission. (2024, February 14). Investigation of the Commission and consumer authorities finds that online influencers rarely disclose commercial content [Press release IP/24/708]. EU influencer disclosure investigation
8. Federal Trade Commission. (2016). Enforcement policy statement on deceptively formatted advertisements. Federal Register, 81(74), 22596–22601. (Original statement announced December 22, 2015.) FTC native advertising policy
9. Federal Trade Commission. (2019). Disclosures 101 for social media influencers. FTC influencer disclosure guidance
10. Federal Trade Commission. (2024, August 14). Federal Trade Commission announces final rule banning fake reviews and testimonials [Press release]. FTC fake review rule
11. Groh, M., Epstein, Z., Firestone, C., & Picard, R. (2022). Deepfake detection by human crowds, machines, and machine-informed crowds. Proceedings of the National Academy of Sciences, 119(1), Article e2110013119. Human and machine detection study
12. Kearns, C. E., Schmidt, L. A., & Glantz, S. A. (2016). Sugar industry and coronary heart disease research: A historical analysis of internal industry documents. JAMA Internal Medicine, 176(11), 1680–1685. Sugar industry document study
13. Kee, E., O’Brien, J. F., & Farid, H. (2013). Exposing photo manipulation with inconsistent shadows. ACM Transactions on Graphics, 32(3), 1–12. Inconsistent shadow analysis
14. Keller, F. B., Schoch, D., Stier, S., & Yang, J. (2020). Political astroturfing on Twitter: How to coordinate a disinformation campaign. Political Communication, 37(2), 256–280. Coordinated disinformation study
15. Köbis, N. C., Doležalová, B., & Soraperra, I. (2021). Fooled twice: People cannot detect deepfakes but think they can. iScience, 24(11), Article 103364. Deepfake overconfidence study
16. Magramo, K. (2024, May 16). British engineering giant Arup revealed as $25 million deepfake scam victim. CNN Business. Arup deepfake fraud reporting
17. McGrew, S., Ortega, T., Breakstone, J., & Wineburg, S. (2017). The challenge that’s bigger than fake news: Civic reasoning in a social media environment. American Educator, 41(3), 4–9, 39. Stanford civic reasoning research
18. MIT Media Lab. (n.d.). Detect DeepFakes: How to counteract misinformation created by AI. Retrieved November 4, 2026, from MIT deepfake detection guidance
19. National Institute of Standards and Technology. (2024). Reducing risks posed by synthetic content: An overview of technical approaches to digital content transparency (NIST AI 100-4). U.S. Department of Commerce. NIST synthetic content report
20. Nickerson, R. S. (1998). Confirmation bias: A ubiquitous phenomenon in many guises. Review of General Psychology, 2(2), 175–220. Confirmation bias research
21. Nightingale, S. J., & Farid, H. (2022). AI-synthesized faces are indistinguishable from real faces and more trustworthy. Proceedings of the National Academy of Sciences, 119(8), Article e2120481119. Synthetic face perception study
22. Pennycook, G., & Rand, D. G. (2019). Lazy, not biased: Susceptibility to partisan fake news is better explained by lack of reasoning than by motivated reasoning. Cognition, 188, 39–50. Reasoning and misinformation study
23. Pennycook, G., & Rand, D. G. (2022). Accuracy prompts are a replicable and generalizable approach for reducing the spread of misinformation. Nature Communications, 13, Article 2333. Accuracy prompt research
24. Pennycook, G., Cannon, T. D., & Rand, D. G. (2018). Prior exposure increases perceived accuracy of fake news. Journal of Experimental Psychology: General, 147(12), 1865–1880. Repeated exposure study
25. Reuters Institute for the Study of Journalism. (2026, June 16). Overview and key findings of the 2026 Digital News Report. University of Oxford. Reuters digital news report
26. Roozenbeek, J., & van der Linden, S. (2019). Fake news game confers psychological resistance against online misinformation. Palgrave Communications, 5, Article 65. Misinformation game study
27. Roozenbeek, J., van der Linden, S., Goldberg, B., Rathje, S., & Lewandowsky, S. (2022). Psychological inoculation improves resilience against misinformation on social media. Science Advances, 8(34), Article eabo6254. Prebunking video study
28. Vosoughi, S., Roy, D., & Aral, S. (2018). The spread of true and false news online. Science, 359(6380), 1146–1151. False news diffusion study
29. Wojdynski, B. W., & Evans, N. J. (2016). Going native: Effects of disclosure position and language on the recognition and evaluation of online native advertising. Journal of Advertising, 45(2), 157–168. Native advertising recognition study