Experts expect social media to flood with about 500,000 videos and voice deepfakes in 2023. This is a big increase from previous years. In 2021, there were 14,678 deepfake videos online, which was double the number in 2018.
By December 2020, researchers found more than 85,000 fake videos. The number kept growing every six months. In 2019, 96% of these deepfakes were pornographic. They mostly targeted celebrities.
The DeepNude app got shut down because it was really popular with 545,000 visits and 95,464 active users. In another case, a UK energy company lost $243,000 because of a deepfake voice scam.
Deepfake fraud has been on the rise in North America. In the U.S., instances of deepfakes used for fraud increased from 0.2% to 2.6% between 2022 and Q1 2023. In Canada, the percentage rose from 0.1% to 4.6%.
According to a worldwide survey, 71% of people don’t know what deepfakes are. However, 57% think they can recognize one.
According to Pew Research Center, 77% of Americans want to limit misleading deepfake content. In a UK study, only 21.6% could identify a deepfake correctly.
Introduction
Unveiling the latest Deepfake Statistics, our comprehensive analysis delves into the current trends, growth, and popularity of deepfakes as of November 2023.
The “2023 State of Deepfakes” report by homesecurityheroes provides a comprehensive analysis of the current state of deepfake technology, based on a study of 95,820 deepfake videos, 85 dedicated channels across various online platforms, and over 100 websites connected to the deepfake ecosystem. The report states that the number of deepfake videos online in 2023 is 95,820, marking a 550% increase from 2019. [1]
According to DeepMedia, a firm specializing in deepfake detection, about 500,000 video and voice deepfakes are expected to be shared globally on social media in 2023. [2]
The number of deepfake videos on the internet has doubled since 2018, reaching 14,678 in 2019, as told by Deeptrace, a cybersecurity company’s report titled “The State of Deepfakes: An Overview,” published in September 2019, [3]
A report named ‘The State of Deepfakes 2020’ by the company Sensity noted that over 85,000 harmful deepfake videos were detected up to December 2020, and the number has been doubling every six months.
Initially gaining popularity in 2017, deepfakes were predominantly used to generate fake pornography of celebrities. Even in 2019, 96% of all publicly posted deepfakes were pornographic, mainly targeting individuals in the entertainment industry. [4]
The report “The State of Deepfakes: An Overview” was published by Deeptrace in September 2019 [3]. Deeptrace, founded in 2018, is a company dedicated to researching deepfakes, their evolving capabilities, and associated threats. The quantitative data collection for the report occurred between June 1st, 2019, and July 31st, 2019. Additionally, examples and case studies were gathered until August 31st, 2019. The report provides several key statistics on the prevalence and impact of deepfakes:
Total Number of Deepfake Videos Online: As of the report’s publication, there were 14,678 deepfake videos online, marking an almost 100% increase from the previous count of 7,964 in December 2018.
Content Type: 96% of these deepfake videos were pornographic, while only 4% were non-pornographic, highlighting the predominant use of deepfakes in creating explicit content.
Views on Deepfake Pornography Websites: The top four deepfake pornography websites alone had garnered over 134 million video views, despite being a relatively new phenomenon.
Demographic Breakdown: 100% of the subjects in deepfake pornography were women. In contrast, deepfake videos on YouTube featured 61% male subjects and 39% female subjects, indicating a gender bias in the type of content created.
Deepfake Pornography Websites: The report found that non-Western subjects featured in almost a third of videos on deepfake pornography websites. Notably, South Korean K-pop singers made up a significant portion, constituting 25% of the subjects targeted. The breakdown of nationalities in deepfake pornography was as follows:
USA: 41%
South Korea: 25%
UK: 12%
Canada: 6%
India: 13%
Taiwan: 2%
Other: 3%
YouTube Channels: In contrast, deepfake videos on YouTube predominantly featured Western subjects. The nationality breakdown for YouTube channels was:
USA: 73%
UK: 14%
Canada: 4%
South Korea: 2%
Other: 5%
Professions Targeted: Almost all (99%) subjects in deepfake pornography were actresses and musicians in the entertainment industry. However, YouTube deepfake videos showcased various professions, including politicians and business owners.
DeepNude App Popularity: The DeepNude app, which generated nude images from clothed photos, received 545,162 visits and 95,464 active users in June before being taken offline. It was later sold for $30,000 and continued to be distributed online.
Use in Fraud: Synthetic voice audio has been used in fraud attempts. In one case, a UK energy firm was deceived into wiring $243,000 to a fraudulent account by a voice clone of the CEO’s German parent company. The loss was covered by fraud insurance.
According to the report in 2023, Deepfake pornography constitutes 98% of all deepfake videos online. Of these, 99% target women, with South Korean singers and actresses being the most targeted group (53%). Additionally, 94% of those featured in deepfake pornography work in the entertainment industry. [1]
Seven out of the top ten pornography websites host deepfake content. The top ten dedicated deepfake pornography websites have amassed a total of 303,640,207 video views, with monthly traffic of 34,836,914. [1]
There has been a 464% increase in the production of deepfake pornographic videos from 2022 to 2023. [1]
Websites dedicated to deepfake pornography hold about a 90% market share, compared to 10% held by mainstream pornography websites. [1]
One in every three deepfake tools allows the creation of adult content, with 92.3% of these platforms offering free access to create deepfake adult content. [1]
The proportion of deepfakes in North America more than doubled from 0.2% to 2.6% in the U.S. between 2022 and Q1 2023. This significant increase in deepfake activity is also observed in Canada, where the proportion jumped from 0.1% to 4.6% during the same period. At the same time, printed forgeries, which constituted 4% to 5% of all fraud in 2022, dropped to 0% in the last quarter. [Source: Business Wire]
A 300% increase in deepfake-based fraud attempts was reported in 2020. [Source: Digital Guardian]
A significant 31 times increase in deepfake fraud from 2022 to 2023 is reported by the Onfido article. [6]
Deepfake Usage Demographics
Region
Awareness of Deepfakes
Ability to Distinguish Deepfakes
Global
29%
57% believe they can spot
Mexico
40%
82% believe they can spot
UK
32%
–
Spain
–
75% do not know
Germany
–
57% cannot tell
The proportion of deepfakes in North America more than doubled from 0.2% to 2.6% in the U.S. between 2022 and Q1 2023. This significant increase in deepfake activity is also observed in Canada, where the proportion jumped from 0.1% to 4.6% during the same period. [Source: Business Wire]
A survey conducted by iProov included 16,000 respondents across eight countries (the U.S., Canada, Mexico, Germany, Italy, Spain, the UK, and Australia) in 2022, several key findings were reported:
Global Awareness of Deepfakes: 71% of respondents globally stated that they do not know what a deepfake is, indicating a significant lack of awareness. Just under a third of global consumers reported being aware of deepfakes. The awareness level differs by country, with Mexico (40%) and the UK (32%) having the highest percentage of respondents who say they know what a deepfake is. Conversely, Spain and Germany have the lowest awareness, with 75% of respondents in both countries admitting they do not know what a deepfake is.
Increase in Awareness Over Time: The percentage of people who know what a deepfake is has more than doubled since the previous survey conducted by iProov in 2019. In 2019, only 13% of respondents said they knew what a deepfake was, compared to 29% in 2022. This increase indicates a growing awareness of the deepfake threat, but the fact that only 29% are aware in 2022 still points to a significant knowledge gap.
Ability to Distinguish Deepfakes: Around 57% of global respondents believe they could spot a deepfake, which is up from 37% in 2019. However, 43% admit they would not be able to tell the difference between a real video and a deepfake. Respondents from Mexico are the most confident, with 82% believing they could identify a deepfake, whereas respondents from Germany are the least confident, with 57% saying they would not be able to tell the difference.
Statistics on Technological Advancements in Deepfakes
Category
Percentage
Deepfake Detection Accuracy
99%
Deepfake Videos Made with DeepFaceLab
95%
Deepfake Creation Model Increase
84%
User-Friendly Deepfake Tools
42%
The rise of Generative Adversarial Networks (GANs) since 2014 and the increasing availability of user-friendly deepfake tools have significantly advanced deepfake technology. Approximately 42 user-friendly deepfake tools exist, attracting about 10 million monthly searches. [1]
As reported by CSO Online, an 84% increase in deepfake creation models was observed from 2019 to 2020.
Over 95% of all deepfake videos are made using DeepFaceLab. [Source: Gemini Advisory]
The Skynet Today article notes that some deepfake detection models can achieve upwards of 99% accuracy. [9]
Americans thinking it’s hard to recognize altered content
61%
Americans believing it’s easy to recognize altered content
53%
UK participants identifying deepfake without warning
32.9%
UK participants identifying deepfake with warning
21.6%
UK participants showing heightened skepticism
45%
77% of Americans are in favor of restricting misleading, altered videos and images, commonly referred to as “deepfakes,” comes from a Pew Research Center survey. This survey was conducted between February 19 and March 4, 2019. [Source: Pew Research Center]
The same report suggests that 66% of Americans sometimes encounter misleading altered videos and images, with 15% encountering them often. [Source: Pew Research Center]
The report also states that 61% of U.S. adults think it’s too much to ask the average American to recognize altered videos and images, while 53% of Americans believe it’s easy for them to recognize altered content. The report indicates that 77% of both Republicans and Democrats support restrictions on altered videos and images. [Source: Pew Research Center]
In a research study titled “Deepfake detection with and without content warnings” with 1093 participants from the United Kingdom published on 27th Nov. 2023 [11], it was found that:
When individuals were not warned about deepfakes, only 32.9% could detect anything out of the ordinary in videos containing a deepfake. This rate was not significantly different from the control group, where 34.1% detected something unusual in authentic videos.
On the other hand, when warned that at least one video in a set is a deepfake, only 21.6% of participants correctly identified the deepfake as the sole inauthentic video. Nearly half mistakenly identified at least one genuine video as a deepfake.
A large share of participants (45%) identified more than one video as a deepfake, indicating heightened skepticism.
Final Words:
As we’ve seen, the prevalence and impact of deepfakes are rapidly evolving, posing significant challenges in various sectors. It’s crucial to stay informed and vigilant, as these statistics highlight the urgent need for awareness and solutions in the ever-growing deepfake landscape.
