YouTube
Last reviewed
May 4, 2026
Sources
15 citations
Review status
Source-backed
Revision
v1 · 3,695 words
Improve this article
Add missing citations, update stale details, or suggest a clearer explanation.
Suggest edit
CtrlK
Last reviewed
May 4, 2026
Sources
15 citations
Review status
Source-backed
Revision
v1 · 3,695 words
Add missing citations, update stale details, or suggest a clearer explanation.
YouTube is an American online video sharing and social media platform owned by Google LLC, a subsidiary of Alphabet Inc., and headquartered in San Bruno, California. Launched on February 14, 2005 by three former PayPal employees, YouTube is the second most visited website in the world after Google Search and the largest video platform on the internet by audience size, watch time, and creator base.
YouTube hosts user-uploaded videos, live streams, music, podcasts, and short-form vertical videos, supporting playback in roughly eighty languages and reaching more than two and a half billion logged-in monthly users. Viewers collectively watch over one billion hours of YouTube content every day, and the platform serves both as a global distribution channel for professional and amateur creators and as a primary entry point for video advertising on the open web.
YouTube is central to modern artificial intelligence in three distinct ways. First, its recommender system is one of the largest production deep learning systems in the world, ranking and serving billions of personalized video impressions each day. Second, the platform has become a major surface for generative AI features developed by Google and Google DeepMind, including the Lyria-powered Dream Track music tool, the Imagen-powered Dream Screen, and the Veo-powered video generation tools that ship inside YouTube Shorts. Third, the public corpus of YouTube videos has been used, with and without authorization, to train large multimodal models including OpenAI's Whisper speech recognition system, Google's Gemini family, and other frontier large language models.
| Field | Value |
|---|---|
| Founded | February 14, 2005 |
| Founders | Chad Hurley, Steve Chen, Jawed Karim |
| Headquarters | 901 Cherry Avenue, San Bruno, California, United States |
| CEO | Neal Mohan (since February 16, 2023) |
| Owner | Google LLC, subsidiary of Alphabet Inc. |
| Acquired | November 13, 2006, for US$1.65 billion in stock |
| Industry | Online video, advertising, music streaming, live TV |
| 2024 ad revenue | US$36.1 billion (Alphabet FY2024 reporting) |
| Monthly logged-in users | More than 2.5 billion (2024) |
| Watch time | More than 1 billion hours per day |
| Languages | Approximately 80 |
| Parent stock | NASDAQ: GOOGL, GOOG |
| Website | youtube.com |
YouTube originated in early 2005 with three former PayPal employees, Chad Hurley, Steve Chen, and Jawed Karim. The trio had become wealthy after eBay's 2002 acquisition of PayPal, where Hurley had worked as a designer and Chen and Karim as engineers. The popularly told origin story is that the founders were frustrated by how difficult it was to share short videos online, especially clips of a 2004 dinner party at Chen's apartment in San Francisco, although Karim has said this account is partly post hoc.
The domain youtube.com was activated by the founders on February 14, 2005, the date generally cited as the company's founding. The first uploaded video, titled "Me at the zoo," was a nineteen-second clip of Karim at the San Diego Zoo posted on April 23, 2005. A public beta opened in May 2005, and the formal launch followed in November 2005. Sequoia Capital led an early investment of US$3.5 million that month, and the platform's traffic grew explosively through 2006 as embedded video became a widespread feature of the social web.
On October 9, 2006, Google announced an agreement to acquire YouTube for US$1.65 billion in an all-stock transaction. The deal closed on November 13, 2006, less than two years after the company's founding. At the time, YouTube had roughly 65 employees and minimal revenue, and the price was the second largest acquisition in Google's history to that point. The transaction proved to be one of the most consequential in technology history: by 2024 YouTube generated more than US$36 billion in annual ad revenue alone, and analysts have repeatedly estimated the standalone business at hundreds of billions of dollars.
Following the acquisition, YouTube remained operationally independent within Google but progressively integrated with Google's advertising stack, account system, and infrastructure. The company moved into purpose-built offices in San Bruno, California, and Google began the long copyright settlement process that produced the Content ID system in 2007.
| Date | Event |
|---|---|
| February 14, 2005 | Domain youtube.com activated by Hurley, Chen, and Karim |
| April 23, 2005 | First video, "Me at the zoo," uploaded by Jawed Karim |
| November 2005 | Public launch following private beta |
| October 9, 2006 | Google announces US$1.65 billion all-stock acquisition |
| November 13, 2006 | Google acquisition closes |
| 2007 | YouTube Partner Program announced; Content ID launches |
| April 2010 | Live streaming first introduced via partnership broadcasts |
| October 2010 | Co-founder Chad Hurley steps down as CEO; Salar Kamangar named CEO |
| December 2011 | Channels redesign launched, shifting focus from videos to creators |
| February 5, 2014 | Susan Wojcicki named CEO, succeeding Kamangar |
| October 2015 | YouTube Red subscription launches (later YouTube Premium) |
| 2016 | Deep learning recommender system paper published at RecSys |
| February 2017 | YouTube TV announced; rollout begins April 5, 2017 |
| June 2018 | YouTube Music relaunched and YouTube Red rebranded to YouTube Premium |
| September 4, 2019 | US$170 million COPPA settlement with the FTC and New York |
| September 15, 2020 | YouTube Shorts launches in beta in India |
| July 13, 2021 | YouTube Shorts rolls out globally to more than 100 countries |
| September 2023 | Dream Screen and YouTube Create previewed at Made on YouTube |
| November 16, 2023 | Dream Track and Music AI Tools unveiled with Google DeepMind's Lyria |
| February 16, 2023 | Susan Wojcicki steps down; Neal Mohan named CEO |
| March 18, 2024 | Mandatory disclosure rules for realistic AI altered content take effect |
| April 2024 | New York Times reports OpenAI transcribed more than one million hours of YouTube to train GPT-4 |
| 2025 | Veo 3 Fast integration brings text-to-video generation with audio to YouTube Shorts |
YouTube operates a portfolio of consumer products and creator-facing services, all linked through a single Google account and supported by shared advertising and recommendation infrastructure.
| Product | Launch | Description |
|---|---|---|
| YouTube (main platform) | 2005 | Long-form and short-form video sharing with personalized recommendations, search, comments, and live streaming |
| YouTube Shorts | 2020 (India), 2021 (global) | Vertical short-form video format up to 60 seconds, competing with TikTok and Instagram Reels |
| YouTube Premium | 2015 (as Red), rebranded 2018 | Paid subscription that removes ads, allows downloads, and enables background play |
| YouTube Music | 2015, relaunched 2018, replaced Google Play Music December 2020 | Audio and music video streaming service bundled with Premium |
| YouTube TV | February 2017, rollout April 2017 | Live linear television streaming service in the United States with cloud DVR |
| YouTube Kids | February 2015 | Standalone app for children with curated content and parental controls |
| YouTube Studio | Mid 2010s | Web and mobile dashboard for creators covering analytics, monetization, and content management |
| YouTube for Artists | 2017 | Tools and analytics aimed at recording artists and labels |
YouTube's recommendation system is widely regarded as one of the largest production deep learning systems ever deployed. Hundreds of millions of users open the app to a personalized home feed and watch next queue, and recommendations are estimated to drive a majority of viewing time on the platform.
The system began as a collaborative filtering approach in the late 2000s, ranking videos by co-watch and co-click signals. A landmark 2016 paper presented at the ACM Conference on Recommender Systems, "Deep Neural Networks for YouTube Recommendations" by Paul Covington, Jay Adams, and Emre Sargin of Google, described the shift to a two-stage deep learning architecture. The first stage, a candidate generation network, narrows the corpus from millions of videos to a few hundred candidates per user using user history embeddings. The second stage, a ranking network, scores the candidates using richer features, including video metadata, user demographic signals, watch context, and a wide set of impression-time features. Both stages are trained on implicit feedback such as completed watches and explicit signals such as likes and "not interested" responses.
Since 2016 the production system has grown substantially in scale and complexity. Reported architectural advances include sequence models for user histories, multi-task ranking heads that jointly optimize for different forms of engagement and satisfaction, and large transformer-based models trained on user interaction sequences. YouTube's engineers have published or co-authored follow-up papers on multitask ranking, calibrated ranking, and user satisfaction modeling. The platform runs continuous A/B testing infrastructure that allows hundreds of concurrent experiments on small percentages of traffic, and ranking changes are deployed only after measured improvements on guarded engagement and satisfaction metrics.
The recommendation pipeline has also been a source of public controversy: critics have argued that engagement-optimized recommendations can amplify low-quality, sensational, or radicalizing content, a concern often summarized as the "rabbit hole" effect. YouTube has responded with a series of policy and ranking changes since 2019 that explicitly down-rank borderline content even when such videos generate strong engagement signals.
From 2023 onward YouTube has rolled out an expanding suite of generative AI features, almost all built in collaboration with Google DeepMind and Google Research. These features run inside the YouTube and YouTube Studio apps and are available primarily to creators in the United States with broader rollouts following.
| Feature | Launched | Description |
|---|---|---|
| Dream Screen | September 2023 (preview), expanded 2024 | AI-generated still and video backgrounds for YouTube Shorts using the Imagen 3 image model and later Veo for motion |
| Dream Track | November 16, 2023 | Generates 30-second AI music clips for Shorts in the voice and style of select participating artists, powered by Lyria |
| Music AI Tools | November 16, 2023 | Set of DeepMind tools that let producers generate instrumentation from a hum, transform timbre, and create backing tracks |
| Auto-dubbing | Expanded throughout 2024 | Automatic translation and dubbing of long-form videos into additional languages using neural translation and voice models |
| AI generated thumbnails and titles | 2024 | Suggested titles, descriptions, and thumbnail concepts surfaced inside YouTube Studio |
| Comments summarizer | 2024 | Generates topic summaries of comment threads to help creators triage feedback |
| Conversational AI for video Q&A | 2024 | A Gemini-powered companion that answers questions about a video while it plays |
| Inspiration tab in YouTube Studio | 2024 | Generative ideas for new videos based on a creator's audience and trending topics |
| Veo integration in Shorts | Announced 2024, expanded 2025 with Veo 3 Fast | Text-to-video generation with native audio inside the Shorts camera, producing short clips at 9:16 aspect ratio |
| SynthID watermarking | 2023 onward | DeepMind's invisible watermark embedded in AI-generated audio, image, and video content created on the platform |
A related policy change took effect on March 18, 2024, when YouTube began requiring creators to disclose realistic altered or synthetic content during upload. The disclosure label is shown to viewers in the video description, and additional prominent labels appear on the player for content involving sensitive topics such as elections, public health, or simulated real events. The policy applies to videos, Shorts, and live streams, and is intended to address the rising volume of AI generated and AI altered media on the platform.
At the 2025 Made on YouTube event, YouTube announced a custom version of Google DeepMind's Veo 3 video model called Veo 3 Fast, designed for low latency 480p generation inside YouTube Shorts and capable of producing video with audio for the first time. The integration also added Speech to Song, which transforms dialogue from eligible videos into instrumental backing tracks for new Shorts. All output is labeled with SynthID watermarks and visible AI content labels.
The public corpus of YouTube videos, which includes hundreds of millions of hours of speech, music, and footage, has become one of the most contested sources of training data for modern AI systems. Public videos are easily accessible, well captioned, and rich in real world audio and visual scenes, qualities that make them attractive for training large multimodal models.
In April 2024 The New York Times reported, in an investigation that drew on internal documents and interviews, that OpenAI had transcribed more than one million hours of YouTube video using its Whisper speech recognition model and used the resulting text to train GPT-4. The same reporting indicated that researchers at Google were aware of OpenAI's activity but did not publicly object, and that Google itself had used YouTube transcripts in training its own AI systems, including precursors to Gemini. YouTube's terms of service prohibit unauthorized scraping or use of its videos to train models, but enforcement has been limited and the legal status of the practice remains contested under fair use doctrine and a growing body of copyright litigation.
The disclosures put YouTube in an unusual position: the platform is simultaneously the target of large scale unauthorized scraping and, through Google's broader AI program, a beneficiary of training on its own corpus. Neal Mohan and other YouTube executives have publicly criticized OpenAI's reported use of YouTube data while acknowledging that Google does train on YouTube content under its own terms with creators. The episode has accelerated industry debate about training data licensing, watermarking, and the rights of creators whose work is used to train large language models and video generation systems.
YouTube enforces an extensive set of community guidelines covering hate speech, harassment, violent extremism, dangerous challenges, misinformation in regulated topics such as elections and public health, child safety, and copyright. Enforcement uses a combination of automated classifiers and human review across a global Trust and Safety organization. Channels accumulate strikes when they violate guidelines, and three strikes within a 90-day window result in channel termination.
Machine learning is central to moderation at YouTube's scale. Classifiers identify candidate violations across video, audio, transcript, thumbnail, and metadata signals, and the most clear cut cases are removed automatically while ambiguous cases are routed to human reviewers. The 2018 introduction of automated removal accelerated takedown times for many policy categories, although civil society groups have repeatedly criticized over-removal and inconsistent enforcement.
Copyright is handled primarily through Content ID, a fingerprinting system that has been in production since October 2007. Rights holders submit reference files for music, video, or both, and uploaded videos are matched against the reference library at upload and continuously thereafter. Rights holders can choose to block, monetize, or simply track matching uploads. By 2018 Google reported that it had invested more than US$100 million in Content ID, and the system has paid out billions of dollars to rights holders, particularly in the music industry. Disputed claims can be appealed and ultimately escalated to the formal DMCA process.
The YouTube Partner Program, first announced in May 2007, allows eligible creators to earn revenue from advertising shown on their videos and from paid features such as channel memberships and Super Chat. As of recent guidance, the basic eligibility threshold for the long form ads track requires 1,000 subscribers and either 4,000 valid public watch hours over the prior twelve months or 10 million Shorts views over the prior 90 days, plus an active linked AdSense account, two-step verification, and good standing on community guidelines.
Monetization channels available to qualifying creators include:
| Stream | Description |
|---|---|
| Advertising revenue | Pre-roll, mid-roll, and bumper ads served against eligible videos and Shorts; creators receive a share of the revenue |
| Channel memberships | Recurring monthly payments from viewers in exchange for badges, emoji, and members-only content |
| Super Chat and Super Stickers | Highlighted paid messages during live streams |
| Super Thanks | One time tips on regular videos |
| YouTube Premium revenue share | Share of subscription revenue based on Premium watch time of the channel's content |
| Merchandise shelf | Built-in storefront for branded merchandise |
| BrandConnect | Sponsorship marketplace connecting creators with brand campaigns |
| Shopping affiliate program | Affiliate revenue from products tagged in videos |
The creator economy on YouTube is now one of the largest in the world. The most-subscribed channel as of late 2024 was Jimmy Donaldson's MrBeast, which surpassed the Indian music label T-Series in June 2024 to take the top spot, with the children's channel Cocomelon ranking third. Forbes estimated MrBeast's 2024 earnings across YouTube, his Feastables snack brand, and other ventures at roughly US$85 million.
YouTube's growth has been accompanied by extensive scrutiny.
Algorithmic recommendations and the "rabbit hole." Beginning around 2018, journalists and researchers, including Zeynep Tufekci and Guillaume Chaslot, argued that YouTube's recommendation system had a tendency to push viewers toward progressively more extreme political, conspiratorial, or sensational content. YouTube responded with a series of recommendation policy changes targeting borderline content and authoritative news sourcing, and academic studies since then have produced mixed results on whether systemic radicalization remains a meaningful effect of the system's defaults.
Children's privacy and the COPPA settlement. On September 4, 2019, the Federal Trade Commission and the New York Attorney General announced a US$170 million settlement, then the largest COPPA penalty in history, with Google and YouTube over allegations that the platform had collected personal information from viewers of child directed channels without parental consent and used it for targeted advertising. As part of the settlement, YouTube began requiring channel owners to designate whether their content is made for kids and disabled personalized advertising, comments, notifications, and several other features on such content.
Elsagate and YouTube Kids content concerns. In 2017, journalists documented a wave of disturbing animated and live action videos that exploited children's search and recommendation patterns to surface inappropriate content, often featuring popular characters like Elsa, Spider-Man, and Peppa Pig. The episode prompted broader content reviews and changes to YouTube Kids curation.
Geopolitical content moderation. Since the 2022 invasion of Ukraine, YouTube has faced disputes with the Russian government over the removal of state-affiliated channels and has been intermittently throttled or blocked in Russia. The platform has made similar high profile decisions involving content related to the war in Gaza, Hong Kong protests, and electoral integrity in multiple democracies.
Generative AI and the music industry. The launch of Dream Track and Music AI Tools in late 2023 was met with a mix of cautious cooperation from major labels and sharp criticism from independent artists, songwriters, and rights organizations who argued that training generative music models on copyrighted recordings without explicit authorization is itself a form of infringement. Negotiations between Google and major labels over training data licensing have been reported throughout 2024 and 2025.
YouTube has had four chief executives since its founding, all reporting to Google's senior leadership and, since 2015, to Google chief executive Sundar Pichai.
| CEO | Tenure | Notes |
|---|---|---|
| Chad Hurley | 2005 to October 2010 | Co-founder; remained as advisor after stepping down |
| Salar Kamangar | October 2010 to February 2014 | Long-time Google product executive |
| Susan Wojcicki | February 5, 2014 to February 16, 2023 | Oversaw expansion to YouTube Music, YouTube TV, YouTube Premium, Shorts, and the COPPA settlement; passed away in August 2024 |
| Neal Mohan | February 16, 2023 to present | Joined Google through the 2007 DoubleClick acquisition; YouTube Chief Product Officer 2015 to 2023 before becoming CEO |
Mohan reports to Sundar Pichai, who is chief executive of both Google and Alphabet. YouTube remains a Google business unit rather than a separately listed Alphabet "Other Bet," and its results are reported within Google Services in Alphabet's quarterly disclosures.
YouTube runs on Google's planet-scale infrastructure, including Google's data center network, the Jupiter datacenter networking fabric, and the Espresso edge network. Video is encoded into a wide range of formats and bitrates, including the AV1 and VP9 codecs that Google has helped develop, and is delivered through Google's content delivery network and through the Google Global Cache appliances installed inside thousands of internet service provider networks worldwide. Live streaming, ad insertion, recommendations, and search all run on shared Google infrastructure and increasingly on the same Tensor Processing Unit accelerators that power Gemini and other Google AI workloads.