AI Generated Broadcast Content

From PRIMO.ai
Jump to navigation Jump to search

Youtube search... ...Google search


AI generated broadcast content is audio, video, and text for radio, television, streaming, and podcasts that is partly or wholly produced by artificial intelligence. The term covers a wide spectrum, from assistive tools that transcribe, caption, tag, and clip programming to generative AI systems that write scripts, synthesize voices, animate presenters, and assemble entire shows. What separates the current wave from older broadcast automation, such as playout scheduling and pre-recorded "voice tracking," is that the machine now produces the content itself rather than simply sequencing material that people made. In practice most deployments sit somewhere in the middle, with large language models drafting, speech synthesis voicing, and human editors approving at least part of the output.

The field moved quickly from novelty to infrastructure. China's state news agency Xinhua introduced an AI news anchor in 2018, and 2023 brought a cluster of launches that showed how far the technology had come: Futuri announced RadioGPT to generate local radio talk segments, Spotify introduced an AI DJ, IBM began generating spoken commentary for Wimbledon highlight videos, and the startup Channel 1 previewed a newscast read by photorealistic avatars. In 2024 NBCUniversal used an AI re-creation of sportscaster Al Michaels's voice to narrate millions of personalized Olympic recaps, and at the 2026 Winter Olympics in Milan and Cortina, AI systems generated thousands of replays and highlight clips for rights holders. Translation followed the same arc, with YouTube opening automatic AI dubbing in 27 languages to all creators in February 2026.

The appeal is economic and practical. Synthetic voices and automated scripting can fill overnight radio shifts, localize programming for small markets, and produce content variants in volumes that would be impractical for human staff alone, while captioning and translation make broadcasts accessible to deaf and hard-of-hearing viewers and to speakers of other languages. The risks are just as concrete, because broadcasting is a high-trust medium built on recognizable human voices, and audiences consistently say they are uncomfortable with news made mostly by machines; artificial presenters rank among the least accepted uses of AI in journalism. Controversies over an undisclosed AI radio host in Sydney, a short-lived AI-run station in Kraków, and flawed AI dubs on a major streaming service show how quickly experiments damage trust when transparency, quality control, or consent is missing.


AI can now script, voice, and assemble broadcast content at a scale no newsroom could staff, but a broadcast is only worth hearing if audiences can trust who, or what, is speaking to them.


The use of AI generated broadcast content has several advantages in journalism. One advantage is the ability to provide real-time translations or captions for live broadcasts. For example: AI can be used to:

  • analyze the audio or video feed in real-time and generate captions or translations on the fly, making the content more accessible to viewers who are deaf or hard of hearing or who speak a different language. This can help improve the inclusivity and reach of broadcast content
  • generate highlights of sports events, analyze players' performance, and provide real-time commentary or insights during a live broadcast
  • analyze weather data and generate forecasts & alerts

How AI Broadcast Systems Work

Most AI broadcast products are pipelines rather than single models. A typical system decides what to cover, writes a script, renders it as speech or video, assembles it into a program, and delivers it, with human review inserted at one or more points. Each stage relies on a different family of techniques, which is why a problem at one stage (a fabricated fact in the script, a mispronounced name in the voice, a mismatched lip movement in the video) can undermine an otherwise polished product. Breaking the process into stages also clarifies where accountability lies, because the most consequential editorial decisions usually happen before any synthetic voice is heard.

Finding and Selecting Stories

The first stage is selection. Futuri's TopicPulse, which feeds its AI radio products, monitors social media and news sources (the company has cited more than 250,000 of them) to surface stories trending in a particular market. Sports and weather systems instead ingest structured data, such as play-by-play statistics or numerical forecast output, which makes them more reliable because the facts arrive in machine-readable form. Live programming is increasingly processed with speech recognition and computer vision so that spoken words and on-screen events can be indexed, searched, and clipped, and classification and recommendation models then decide which items reach which audiences.

Writing the Script

Scripts are usually produced by large language models prompted with station style guides, story summaries, and formatting rules. RadioGPT, for example, combined trend data with OpenAI's GPT-3 and later GPT-4 models to write talk breaks for one, two, or three AI hosts, and its successor product integrates several language models. The central weakness at this stage is hallucination: language models can state false details fluently, which is tolerable in banter about a song but serious in news. Better-designed systems therefore constrain generation to verified material; the IOC's "Olympic GPT" assistant for Milano Cortina 2026 was built to draw only on verified Olympic data rather than the open internet, an approach closely related to retrieval-augmented generation. For anything presented as news, many organizations also require an editor to approve the script before it is voiced.

Giving the Script a Voice and a Face

Modern text-to-speech systems are neural networks trained on large amounts of recorded speech, and voice-cloning services can imitate a specific person from a comparatively short sample. Broadcasters use both stock voices and clones of real presenters: NBC's AI version of Al Michaels was trained on his past NBC broadcast audio, and in 2023 Portland's Live 95.5 introduced "AI Ashley," an AI version of its host Ashley Elzinga. Video presenters add facial animation, lip synchronization, and avatar rendering, built either from scans of real actors (Channel 1's approach) or from fully synthetic faces produced with image and diffusion models. Listening tests suggest the audio side is already convincing: in a 2024 study presented by Futuri, participants hearing paired samples identified the AI voice as human and the human voice as AI 60 percent of the time, although AI video avatars were judged close to, but not yet ready for, prime time.

Assembling, Personalizing, and Delivering Programs

The final stages turn components into programs. Automated editing systems cut highlights from live feeds, insert graphics, and generate replays; at Milano Cortina 2026, Olympic Broadcasting Services (OBS) produced AI replays within seconds and offered an automated clipping service that let broadcasters specify the sport, athlete, duration, and advertising breaks they wanted. Personalization multiplies output, as when Peacock's 2024 Olympic recap tool assembled roughly seven million distinct versions of a daily ten-minute highlight package. Archive systems apply automatic tagging and semantic search so that decades of footage can be retrieved by description, and monitoring tools use anomaly detection to flag technical faults in outgoing signals. The economic logic is that once a pipeline exists, each additional episode, language, or personalized variant costs very little.

Where Humans Stay in the Loop

How much human oversight remains varies enormously, and it is one of the most important differences between products. NBC said a team of editors would review its AI-generated Olympic recaps before release, and Channel 4 described its AI presenter experiment as governed by editorial guidelines that require disclosure whenever AI is used. At the other extreme, the podcast startup Inception Point AI has said it releases about 3,000 episodes a week with a staff of eight, and its chief executive acknowledged that the team does not listen to most episodes before publication. This distinction now has legal weight in Europe, where the AI Act exempts AI-generated public-interest text from labeling only when it has undergone human review or editorial control and someone holds editorial responsibility for it (see Law and Policy below). Human-in-the-loop review does not guarantee accuracy, but it creates a point of accountability that fully automated pipelines lack.

Applications

Radio and Audio Streaming

Radio was the first broadcast medium to test fully synthetic hosts at scale, partly because talk breaks between songs are short, formulaic, and expensive to staff around the clock. Futuri launched RadioGPT in February 2023 with Alpha Media in the United States and Rogers Sports & Media in Canada as beta partners, relaunched it as Futuri AudioAI in late 2023, and in 2024 added CoHostAI, which lets a human host converse with an AI personality on air, and CallerAI, which lets listeners phone in and talk to an AI. Broadcasters interviewed by Radio World reported using AI hosts mainly for overnight shifts and fill-ins rather than marquee programs. The most common uses remain behind the scenes, including show preparation, scheduling, transcription, and turning broadcasts into podcast episodes and social posts.

Three episodes shaped public reaction. In October 2024, OFF Radio Kraków, a channel of Polish public broadcaster Radio Kraków, relaunched with three AI-generated presenters and aired a simulated interview with the late Nobel laureate Wisława Szymborska; after a former presenter's petition gathered more than 23,000 signatures, the station ended a planned three-month experiment after about a week. In April 2025, reporting revealed that "Thy," the host of a four-hour weekday show on ARN's Sydney station CADA, had been generated with ElevenLabs technology modeled on a company employee and had broadcast for roughly six months without disclosure, which was legal in Australia at the time. In late 2025 iHeartMedia, the largest US radio operator with more than 850 stations, adopted a "Guaranteed Human" pledge that its on-air hosts and the music it plays will not be synthetic, while continuing to use AI tools behind the scenes for scheduling, research, and editing.

Personalized audio streams are the commercial counterpart to AI radio. Spotify's DJ, launched in 2023, combines the company's recommendation engine with commentary generated using OpenAI technology and spoken in a synthetic voice modeled on Spotify's Xavier "X" Jernigan, and in May 2025 it began taking spoken requests by mood, genre, artist, or activity from Premium users in more than 60 markets. Spotify reported that DJ engagement had nearly doubled over the preceding year, a sign that listeners may accept synthetic voices more readily when the voice is a disclosed product feature rather than a presenter passing as human. The feature is described in more detail in the Spotify DJ section below.

Podcasts and On-Demand Audio

Generative audio has also changed podcasting. Google's NotebookLM introduced "Audio Overviews" in 2024, turning documents into conversational summaries read by two synthetic hosts, and publishers have used similar tools to convert articles into audio. At industrial scale, Inception Point AI, led by former Wondery executive Jeanine Wright, reported more than 5,000 shows and about 3,000 new episodes a week at roughly one dollar per episode, covering niches such as local weather and pollen reports, with hosts that identify themselves as AI at the start of episodes. Critics describe this model as "AI slop" that floods directories and dilutes discoverability and revenue for human creators, while the company argues that cheap production makes otherwise unviable niche content possible. The debate has prompted industry calls for a standard way to disclose AI-generated content in podcast feeds.

News and Synthetic Presenters

In news, AI is used most widely for transcription, translation, summarization, and archive search, and least for the person on camera. Synthetic anchors attract outsized attention because they are the most visible form of automation. Channel 1, a Los Angeles startup, released a showcase newscast in December 2023 with photorealistic anchors built from scans of real people and outlined plans for personalized newscasts drawn from partner outlets, freelance reporters, and AI-written stories based on public records. It also proposed generating imagery of events that cameras had not captured, which critics warned would blur the line between illustration and evidence, even though the company promised to label AI-generated imagery.

In October 2025, Channel 4's investigative series Dispatches revealed at the end of the documentary "Will AI Take My Job?" that its presenter, "Aisha Gaban," had been entirely AI-generated, with face, voice, and movements created without any real-world filming; the broadcaster called it a British television first. Channel 4 said it would not make a habit of the practice and that its news depends on fact-checked, impartial journalism, which it called "something AI is not capable of doing." Earlier projects made stronger claims: the 2023 website NewsGPT promoted itself as a 24-hour channel with "no reporters and no bias," an assertion examined in the NewsGPT section below. Survey data help explain broadcasters' caution, since only 19 percent of respondents in a six-country Reuters Institute study said they were comfortable with news organizations creating an artificial presenter or author.

Sports

Sports broadcasting is one of the most mature uses because games generate rich, structured data and far more footage than any human team can cover. Since 2023, Wimbledon and IBM have used generative AI built on IBM's watsonx platform to add spoken commentary and captions to highlight videos, with the stated long-term aim of covering matches that lack human commentators, such as junior, senior, and wheelchair events. NBCUniversal's 2024 Peacock recaps paired automated highlight selection with Al Michaels's cloned voice, and editors reviewed the output before release.

At Milano Cortina 2026, OBS planned more than 6,000 hours of content, about six times the duration of the competition itself, while cutting its on-site infrastructure by about a quarter. Its partner Alibaba Cloud supplied AI replay systems that reconstruct key moments in three dimensions in 15 to 20 seconds, which the company said produced 4,198 video highlights during the Games, and the IOC introduced the first Olympic assistants built on large language models, powered by Alibaba's Qwen models. AI also generated English translations of selected radio exchanges between alpine skiers and their coaches, which were broadcast live for the first time. Together these deployments show AI becoming part of the core production chain rather than an add-on for digital extras.

Weather

Weather coverage depends on numerical forecasts, and machine learning has begun to change those forecasts directly. On 25 February 2025 the European Centre for Medium-Range Weather Forecasts (ECMWF) made its Artificial Intelligence Forecasting System (AIFS) operational alongside its physics-based model, describing it as the first fully operational open machine-learning weather model and reporting a reduction of roughly 1,000 times in the energy needed per forecast; an ensemble version followed on 1 July 2025. Because national weather services build their products on such model output, AI increasingly shapes a forecast before any presenter explains it. Generative tools add a second layer by scripting localized forecasts and alerts for radio, apps, and automated weather podcasts, but a forecast still has to be communicated with its uncertainty intact, a task that short generated scripts can easily oversimplify.

Accessibility, Translation, and Dubbing

Accessibility was among the first broadcast uses of AI and remains among the least controversial. Automatic speech recognition produces live captions for viewers who are deaf or hard of hearing, and machine translation combined with speech synthesis produces subtitles and dubbed audio tracks. Audiences broadly approve: in the Reuters Institute's 2025 survey, translation was one of the most accepted uses of AI in news, with 53 percent comfortable. Quality still depends heavily on vocabulary, accents, and background noise, so captions of names, places, and technical terms need correction workflows when accuracy matters.

Dubbing has moved from pilot to platform scale. YouTube, which began testing AI dubbing with a small group of creators, extended automatic dubbing to all creators in 27 languages in February 2026, added more expressive synthetic speech, introduced a setting that lets viewers choose dubbed or original audio, and began piloting lip synchronization that adjusts speakers' mouth movements to match the translated track. Entertainment has proved less forgiving. Prime Video announced an AI-aided dubbing pilot in March 2025, but when "AI beta" English and Latin American Spanish dubs appeared on anime such as Banana Fish and No Game No Life: Zero in late November 2025, criticism of their flat delivery and protests from voice actors led Amazon to remove several English tracks within days. The episode illustrated a recurring pattern: synthetic voices are often acceptable for information, but performance-driven content exposes their emotional limits and raises labor questions for professional actors.

Streaming, Recaps, and Generative Entertainment

Streaming platforms use AI throughout the viewing experience, from recommendation and search to automatically generated recaps. Recaps show both the promise and the risk of summarization: in December 2025 Prime Video withdrew an AI-generated season recap of its series Fallout after viewers pointed out factual errors, an embarrassment for a feature meant to save fans time. More ambitious experiments generate the programs themselves. Fable's Showrunner platform, backed by Amazon's Alexa Fund, opened to the public in 2025 and lets users create animated episodes from short prompts using the company's SHOW-2 model, extending the research described in the AI Showrunner section below. Whether such generative television becomes a durable medium or remains a novelty depends on unresolved questions about intellectual property, revenue sharing with rights holders, and whether audiences want to watch content that anyone can generate.

Trust, Ethics, and Regulation

What Audiences Accept

Survey research shows a persistent "comfort gap" between human and machine news production. The Reuters Institute's Generative AI and News Report 2025, based on YouGov surveys of about 2,000 people in each of Argentina, Denmark, France, Japan, the United Kingdom, and the United States, found that 12 percent were comfortable with news made entirely by AI, rising to 21 percent with a "human in the loop," 43 percent for news led by a human with some AI assistance, and 62 percent for news made entirely by humans. Respondents were most accepting of back-end tasks such as spelling and grammar correction (55 percent) and translation (53 percent), and least accepting of front-facing uses such as creating a realistic image where no photo exists (26 percent) or an artificial presenter or author (19 percent). Only 19 percent said they saw AI labels on news daily, which suggests that disclosure practices lag behind actual use.

Audience habits are also shifting toward AI intermediaries. According to the Reuters Institute's Digital News Report 2025, 7 percent of online news users used AI assistants to get news, rising to 15 percent of those under 25. Radio research points the same way: Futuri reported in 2024 that 20 percent of radio listeners believed they already listened to a station that used AI, whether or not it did, and iHeartMedia cited research showing that about 70 percent of people use AI tools while about 90 percent prefer media produced by humans. These findings suggest that people readily use AI themselves while still placing a premium on recognizable human voices in the media they trust.

Accuracy and Misrepresentation

Accuracy problems arise not only with synthetic presenters but whenever AI systems summarize or repackage journalism. In the largest study of its kind, published in October 2025, the European Broadcasting Union (EBU) and the BBC coordinated 22 public service media organizations in 18 countries, working in 14 languages, to evaluate about 3,000 news answers from ChatGPT, Microsoft Copilot, Google Gemini, and Perplexity. Professional journalists found at least one significant issue in 45 percent of answers, serious sourcing problems in 31 percent, and major accuracy problems, including invented details and outdated information, in 20 percent; Gemini performed worst, with significant issues in 76 percent of its responses. The researchers released a News Integrity in AI Assistants Toolkit that defines a good AI news answer and catalogs failure modes, and separate BBC audience research found that people often blamed news organizations as well as AI developers when an assistant misrepresented a story. For broadcasters, the lesson is that any AI-generated summary, recap, or bulletin needs evaluation against its source material before it reaches air.

Deepfakes, Impersonation, and the Liar's Dividend

The same tools that let a station clone a willing host can impersonate unwilling ones. In February 2023, the research firm Graphika reported what it described as the first state-aligned influence operation to use AI-generated video of fictitious people: a pro-China network known as Spamouflage circulated clips of "Wolf News" anchors that were almost certainly made with avatars from the commercial video service Synthesia. In January 2024, AI-generated robocalls imitating President Joe Biden's voice urged New Hampshire voters to stay home from the primary, and on 8 February 2024 the US Federal Communications Commission unanimously ruled that AI-generated voices count as "artificial" under the Telephone Consumer Protection Act, bringing such calls under existing consent rules. Broadcasters have become targets as well, as when fake reports carrying Euronews branding used AI-generated voiceovers imitating its journalists to spread anti-Ukraine narratives.

Researchers also warn of a "liar's dividend," in which the mere possibility of a deepfake lets people dismiss authentic recordings as fabricated; the human rights technologist Sam Gregory has argued that generative AI makes it easier both to fake reality and to deny it. Detection alone is not a complete answer. UC Berkeley's Hany Farid has reported that people are only slightly better than chance at identifying AI-generated content, which shifts the burden toward provenance systems and media forensics performed by specialists.

Provenance and Labeling

Rather than trying to detect every fake, provenance systems record where authentic content came from. The Coalition for Content Provenance and Authenticity (C2PA), whose roots include Project Origin, a 2020 effort by the BBC, CBC/Radio-Canada, The New York Times, and Microsoft, defines "Content Credentials," cryptographically signed metadata that describes how a file was captured, edited, and published. Broadcasters are among the leading adopters: the BBC and CBC/Radio-Canada were the first publishers on the IPTC's verified news publisher list to publish C2PA-signed content, France Télévisions signs daily news content, and a 2025 IBC Accelerator project championed by the BBC, ITN, the EBU, and others built an open-source tool for stamping video at the point of publication. Hardware is catching up as well. Sony's PXW-Z300, launched in July 2025, was billed as the first camcorder to embed C2PA signatures in video files, and version 2.3 of the specification, launched in early 2026, added a method for signing live and broadcast streams segment by segment.

Provenance has limits that matter in practice. Metadata can be stripped when files are re-encoded or screen-recorded, which is why the industry pairs signed manifests with watermarks and fingerprints that can help recover lost credentials; the Z300 initially signed only MP4 files rather than the MXF format widely used in broadcast production; and a valid credential proves where a file came from, not that its content is true. Platforms add their own labels, such as YouTube's disclosure requirements for realistic synthetic content and its "captured with a camera" label, but those labels do not travel with content across services. Provenance therefore works best as one layer of a verification process that still includes newsroom checks such as open-source intelligence techniques.

Law and Policy

The European Union has adopted the most comprehensive rules. Article 50 of the EU AI Act, which applies from 2 August 2026, requires deployers to disclose deepfake image, audio, and video content and to disclose AI-generated text published to inform the public on matters of public interest, unless that text has undergone human review or editorial control and a person or organization holds editorial responsibility for it. For evidently artistic, creative, satirical, or fictional programs, the deepfake disclosure can be made in a way that does not spoil the work, while providers of generative systems must mark their outputs in a machine-readable, detectable way, with a grace period to 2 December 2026 for systems already on the market. To support compliance, the European Commission published a voluntary Code of Practice on Transparency of AI-Generated Content on 10 June 2026, which the Commission and the AI Board later confirmed as an adequate way to demonstrate compliance. These rules sit within the EU's broader approach to AI governance and risk-based regulation.

The United States has no comprehensive equivalent. Beyond the 2024 robocall ruling, the FCC proposed in July 2024 to require on-air and political-file disclosure of AI-generated content in television and radio political ads; Commissioner Brendan Carr argued that the agency lacked authority, the rule was not adopted before the 2024 election, and Carr became chairman in 2025. In Congress, a revised NO FAKES Act was introduced on 20 May 2026 to create a federal, licensable right in each person's voice and visual likeness, with exemptions for news reporting, commentary, criticism, and parody, and the Senate Judiciary Committee advanced it unanimously on 18 June 2026. State laws, such as Tennessee's ELVIS Act protecting voice and likeness, add a patchwork of further obligations under US law. In Australia, by contrast, there was no legal requirement to disclose an AI host when the CADA case surfaced, which prompted calls from industry observers for disclosure rules.

Labor, Consent, and Compensation

Employment and consent sit at the center of most public objections, a recurring theme in debates about AI ethics. The OFF Radio Kraków backlash focused on former presenters who said their contracts had been ended to make way for AI, although station management attributed the decision to very low ratings. The National Association of Voice Actors condemned Prime Video's AI anime dubs, and some performers said they would refuse future dubbing work for the service. Consent can also be murky inside a company: CADA's "Thy" borrowed the voice and likeness of an employee from the finance department, and critics argued that presenting a synthetic Asian woman as an on-air host amounted to manufactured diversity. By contrast, NBC developed its AI Al Michaels in collaboration with the broadcaster himself, an example of the consent-based model that proposals such as the NO FAKES Act would put on firmer legal footing.

Limitations and Open Questions

Several technical limits remain. Language models still fabricate details and misattribute sources, as the EBU-BBC study showed, and the risk grows when outputs are generated faster than anyone can review them. Synthetic speech is convincing in short segments but can falter on unfamiliar names, emotional range, and comic timing, which is one reason the Prime Video anime dubs drew such strong criticism. Personalization adds a subtler problem: when every listener hears a different version of a show, it becomes harder to correct errors publicly or to hold a broadcaster accountable for what was said.

Economic and cultural questions are equally open. Cheap synthetic output can flood podcast directories and streaming catalogs, making discovery harder for human creators and spreading advertising revenue across vast numbers of low-cost episodes. There is comparatively little independent research on long-term audience effects, such as whether disclosed AI hosts erode trust over time or whether listeners simply grow used to them. Regulation is also fragmented, so a synthetic segment that must be labeled in the European Union may carry no label when the same content airs elsewhere, and provenance systems only help when publishers, platforms, and devices all support them.

Recent Developments

The period from 2025 to 2026 marked a shift from isolated experiments to routine deployment, accompanied by the first binding transparency rules and a visible counter-movement emphasizing human presenters. Several of the most consequential events involved quality failures rather than technical breakthroughs, which pushed broadcasters toward human review, disclosure, and provenance. The table below summarizes key developments discussed in this article.

Date Development Significance
February 2025 ECMWF makes its AIFS machine-learning forecast model operational First fully operational open ML weather model; an ensemble version followed in July 2025
March 2025 Prime Video announces an AI-aided dubbing pilot Streaming services begin testing synthetic dubbing on licensed titles
April 2025 CADA's AI host "Thy" revealed Undisclosed AI host had broadcast for about six months in Sydney
May 2025 Spotify DJ adds voice requests in more than 60 markets Interactive synthetic host becomes a mainstream consumer feature
July 2025 Fable opens Showrunner with Amazon Alexa Fund backing; Sony launches the PXW-Z300 User-generated AI television and C2PA-signed video capture reach the market
October 2025 EBU-BBC News Integrity in AI Assistants study; Channel 4's AI presenter; Reuters Institute Generative AI and News Report Evidence on assistant errors and audience comfort; public demonstration of a convincing synthetic presenter
November and December 2025 iHeartMedia's "Guaranteed Human" pledge; Prime Video removes AI anime dubs and an AI Fallout recap Backlash and quality failures push the industry toward disclosure and human review
February 2026 YouTube auto dubbing for all creators in 27 languages; AI replays and LLM assistants at Milano Cortina; C2PA 2.3 adds live stream signing Translation, sports production, and provenance operate at platform scale
May and June 2026 NO FAKES Act reintroduced and advanced by the Senate Judiciary Committee; EU Code of Practice on AI-generated content published Voice and likeness rights and labeling rules move toward enforcement
August 2026 EU AI Act Article 50 transparency obligations apply Deepfake and AI-generated public-interest text disclosure becomes a legal duty in the EU

Further Resources

Additional Viewing

Load video
YouTube

In this Berkeley Talks lecture, published by UC Berkeley in April 2026, digital forensics pioneer Hany Farid traces how generative AI moved from producing static fake files to interactive deepfakes that can hold live conversations, and explains why human viewers perform only slightly better than chance at spotting synthetic media. He argues for focusing on the platforms and advertising systems that profit from harmful fakes rather than relying on detection alone. The lecture provides the technical background for the deepfakes and provenance sections of this article. Watch on YouTube

Load video
YouTube

In this TED talk, WITNESS executive director Sam Gregory explains how hyper-realistic synthetic audio and video make it easier both to fake reality and to dismiss genuine evidence as fake, and outlines three steps for protecting the public's ability to tell human-made from synthetic content. Although it is shorter than a lecture, it gives a clear, nontechnical account of the "liar's dividend" and of why disclosure and provenance matter for anyone who broadcasts or verifies news. Watch on YouTube

Load video
YouTube

The Reuters Institute's opening episode for its Digital News Report 2026, hosted by director Mitali Mukherjee with lead author Jim Egan and research director Richard Fletcher, reviews global trends in how people get news, including platform dominance, declining trust, news influencers, and a segment on the use of AI chatbots for news. It complements the audience research summarized in the What Audiences Accept section and shows how AI intermediaries fit into wider shifts in news consumption. Watch on YouTube

RadioGPT

YouTube ... Quora ...Google search ...Google News ...Bing News

Media company Futuri launched RadioGPT, billed as the world's first radio platform powered by artificial intelligence, in February 2023. RadioGPT uses Futuri's TopicPulse technology to scan online news sources and social media and identify topics and trends in local markets, then writes on-air scripts with OpenAI's GPT models and voices them with synthetic or cloned voices. Stations could choose from more than 100 voices, give each a personality, and run single, duo, or trio hosted shows for individual dayparts or an entire station. Less than nine months after launch, Futuri relaunched the product as Futuri AudioAI, which integrates several large language models and extends the system beyond radio to podcasts and other media.

  • The platform creates scripts for radio broadcasts, which are delivered on air by AI-generated personalities, using OpenAI's GPT-3 and later GPT-4 models, the same model family that powers ChatGPT.
  • Futuri's technology can automate other processes as well, including creating website blogs, social media posts, short-form videos and converting on-air content into podcasts.
  • In April 2024 Futuri added CoHostAI, which lets human hosts converse with AI personalities during live broadcasts, and CallerAI, which lets listeners call in and talk directly with an AI.

Reception was mixed from the start. Some broadcasters adopted AI voices for overnight, weekend, and fill-in shifts, while radio professionals warned that the technology could eliminate many on-air jobs, and at least one Futuri customer publicly said it would drop the company's products in protest. Futuri's own 2024 audience research found that listeners frequently could not tell AI voices from human ones, a result that strengthens the case for disclosure rather than settling it. Later controversies over AI hosts in Poland and Australia, and iHeartMedia's "Guaranteed Human" response, are discussed in the Radio and Audio Streaming section above.

NewsGPT

NewsGPT launched in March 2023 as an online news channel that uses artificial intelligence to generate news stories with a Large Language Model (LLM) and present them without human reporters. Its chief executive, Alan Levy, promoted it as news with "no reporters and no bias," claiming that the absence of human editors would remove human bias and error and that AI could produce stories in seconds. Those claims deserve skepticism. Language models reflect the choices and biases embedded in their sources and training data, they can fabricate details, and without editors there is no one to catch mistakes before publication; independent evaluations such as the 2025 EBU-BBC study found significant problems in nearly half of AI assistants' answers about the news.

Speed and the absence of a human byline are not the same as accuracy or impartiality, which in broadcast journalism depend on verification, sourcing, and editorial accountability. NewsGPT is better understood as an early demonstration of automated aggregation than as proof that AI removes bias. Its claims nonetheless anticipated the questions that later services such as Channel 1 and AI-generated podcast networks would face about who checks the facts when no journalist is involved.

Spotify DJ

Spotify introduced DJ in February 2023 as an AI guide that combines the company's personalization technology with generative AI built on OpenAI technology, which Spotify said its music editors use to provide facts about the music, artists, and genres being played. Spotify has emphasized that editorial expertise shapes the commentary, describing generative tools as a way for editors to scale their knowledge rather than replace it. The spoken commentary is delivered in a synthetic voice modeled on Xavier "X" Jernigan, Spotify's head of cultural partnerships, making DJ one of the first mass-market products built around a cloned broadcast-style voice.

Available through the music feed in the Spotify mobile app, DJ personalizes listening through a curated selection that includes spoken-word commentary with light-hearted banter and context about songs and artists the user has previously played, working much like a radio DJ who tailors a show to each listener (Spotify expands its AI-powered DJ feature globally | Paul Sawers - TechCrunch). In May 2025 Spotify added voice requests, letting Premium subscribers in more than 60 markets hold the DJ button and ask for music by genre, mood, artist, or activity, and in September 2025 it extended requests to Android Auto. Spotify said DJ listener engagement had nearly doubled in the year before voice requests launched, which makes the feature one of the most widely used examples of a synthetic host that listeners know is synthetic.

Television (TV)

Youtube search... ...Google search

Television has been reshaped by video streaming services that use machine learning to make entertainment more personalized, accessible, and automated. Beyond customizing each viewer's home screen through recommendation systems, machine learning informs other steps in the workflow, from the coordination of logistics to how certain shows are recorded and produced, and it has helped increase both the number of streaming subscribers and the amount of television they watch (Machine Learning Makes TV More Watchable Than Ever). Broadcasters now also use AI for captioning, dubbing, highlight clipping, and archive search, as described in the applications sections above. On the receiving end, television sets increasingly include AI picture processors that identify the type of content being shown and adjust upscaling, noise reduction, and sound accordingly, a form of image enhancement illustrated by the manufacturer videos below.


Can Samsung's AI upscaling really make TV images better? Artificial intelligence and machine learning aim to bridge the resolution gap between 4K and 8K TVs and the video actually available today.


Load video
YouTube
Philips 2020 OLED: 'AI' video processor, 8K OLED prototype, more
Philips' picture quality guru, Danny Tack, demonstrates the latest 'AI' based video processor as well as future developments such as 8K OLED, intelligent burn-in protection, and "next level source perfection". More information here: http://www.flatpanelshd.com/focus.php?subaction=showfull&id=1580113886

Load video
YouTube
Sony - New BRAVIA XR TV Announcement
The world's first cognitive intelligence TV*, BRAVIA XR enables you to completely immerse in an ultra-realistic visual and sound experience at home. Introducing BRAVIA XR Features: http://youtu.be/Z02epMU4Avs

Load video
YouTube
Introducing Samsung’s Machine Learning Super Resolution (MLSR): 8K Picture Quality on TV through AI
Samsung’s Machine Learning Super Resolution (MLSR) utilizes AI technology to compare low and high-quality versions of the same content to learn the technological differences between the two and form a vast database. It analyzes millions of pieces of video content and finds a correlation. Based on its analysis, it can select the optimum filters that support brightness, the level of blackness, spread and other errors from all inputs, and transform low-definition content to close to 8K high definition. The input content is recognized as ‘real-time’ based on a frame and is enhanced by scenes, which makes it possible to upgrade image and sound quality immediately, regardless of whether the video source is live streaming or OTT (Over The Top). Other upscaling techniques require human input to compare low-resolution and high-resolution scenes and find ways to reconstruct them. However, Samsung’s AI Technology enables it to self-study millions of images on its own using MLSR, allowing much-improved accuracy compared to conventional technologies.

Load video
YouTube
LG ThinQ AI Demonstration
Thomas TV-Guy

Load video
YouTube
Android TV: How to shape the Future of TV with Operator Tier, Pierre Donath (3SS)
3SS - 3 Screen Solutions. Pierre Donath's speech at Google's Android TV workshop at Connected TV World Summit 2019! Our Product: http://3ss.tv/3ready/

Load video
YouTube
VG SMART TV with Artificial Intelligence
VG India

Load video
YouTube
Panasonic JZ2000 OLED TV First Look
... Elsewhere in the TV, the headline upgrades are a new HXC Pro AI Processor, cutting-edge HDR support, HDMI 2.1 functionality (plus a new Game Mode Extreme), and better speakers. The new processor promises to use AI to "analyze and detect what content is being viewed, whether that be sports, movies, news, etc, and automatically optimize both picture and sound quality". That's pretty much what all the AI processors from the likes of Samsung, Philips and so on do, but if it brings improved image quality to a model that was already right at the top of our list of the best TVs, then so much the better.

Load video
YouTube
Smart Mirror Touchscreen (with Face ID) using Raspberry Pi 4 | Full Tutorial
Eben Kouao. You can find the compiled face detection image on our website: http://smartbuilds.io/

AI Showrunner | Simulation

Fable Studio, the San Francisco company led by Edward Saatchi that previously won Emmy awards for interactive storytelling, introduced its "AI Showrunner" research in July 2023 as part of a project called The Simulation. The accompanying paper, by Philipp Maas and colleagues, describes SHOW-1, a system that combines large language models, custom diffusion models, and a multi-agent simulation so that story progression and character behavior emerge from AI characters living in a simulated world rather than from a single prompt. To demonstrate it, Fable generated episodes in the style of South Park, stating that they were made for research only and would not be released as a product, although critics noted that the demonstration arrived in the middle of the 2023 Hollywood writers' strike. The approach treats a TV episode as the output of an agent-driven world, with AI handling writing, animation, voicing, and editing, and it connects AI television to broader work on AI storytelling.

In 2024 Fable announced Showrunner as a streaming platform, and in July 2025 it opened the service to the public in alpha with an undisclosed investment from Amazon's Alexa Fund. Powered by a newer model, SHOW-2, Showrunner lets users generate animated scenes or full episodes from a short text prompt, either within existing story worlds such as the satirical comedy Exit Valley or from scratch, and add characters based on their own photos. Fable has said it plans to share revenue with creators whose shows become popular and has discussed licensing arrangements with studios for fan-made episodes, and it has tested an "AI TV" channel that streams generated shows continuously. The company planned a credit-based subscription costing roughly $10 to $40 per month. Key features of the platform include:

  • It generates animated scenes or full episodes from a short text prompt, handling script, voices, animation, and editing.
  • It lets users remix existing shows, create derivative episodes, or insert themselves as characters.
  • It runs on Fable's proprietary SHOW-2 model, which builds on the SHOW-1 research and focuses on animation to reduce computing costs compared with photorealistic video.
  • It raises open questions about training data, copyright, and revenue sharing; Saatchi has said the system was trained on publicly available data.

Load video
YouTube
Load video
YouTube

Multimedia Broadcasting

Artificial intelligence (AI) plays a significant role in multimedia broadcasting, changing how content is produced, distributed, and consumed. The main uses fall into the categories below, ranging from back-end production and quality control to audience-facing personalization, and most broadcasters combine several of them in a single workflow. Many of these functions are assistive rather than generative, which is why they have drawn far less controversy than synthetic presenters.


  • Automated Captioning and Subtitling:
    • AI-powered speech recognition can automatically generate accurate captions and subtitles for live broadcasts, improving accessibility for hearing-impaired audiences.
  • Voiceovers and Dubbing:
    • AI-driven voice synthesis can generate natural-sounding voiceovers in multiple languages, making it easier to create content for a global audience.
  • Content Moderation:
    • AI can be used to automatically detect and filter out inappropriate content, ensuring that broadcasts meet regulatory standards and are suitable for all audiences.
  • Real-time Analytics:
    • AI algorithms can analyze viewer reactions and sentiments in real-time, helping broadcasters adjust content and marketing strategies on the fly to maximize engagement.
  • Content Indexing and Search:
    • AI can automatically tag and index multimedia content, making it easier for broadcasters to search and retrieve archived material quickly.
  • Automated Production:
    • AI-powered robotic cameras and directors can automate the filming and editing process, reducing production costs and improving efficiency.
  • Ad Insertion and Targeting:
    • AI can personalize ad content and insertion times based on viewer preferences, increasing ad effectiveness and revenue for broadcasters.
  • Quality Control:
    • AI can monitor broadcast signals in real-time, automatically detecting and alerting operators to technical issues, ensuring high-quality delivery.
  • Content Recommendation:
    • AI-driven recommendation systems suggest related content to viewers, keeping them engaged and increasing content consumption.
  • Real-time Translation:
    • AI can provide real-time translation for interviews or live events, making broadcasts accessible to a global audience.
  • Audience Engagement:
    • Chatbots and virtual Assistants powered by AI can engage with viewers in real-time, answering questions and providing additional information.
  • Predictive Analytics:
    • AI can analyze historical data to predict viewer preferences and trends, aiding in content planning and scheduling.
  • Content Provenance and Labeling:
    • Signed Content Credentials based on the C2PA standard record how video and images were captured and edited, helping audiences and platforms distinguish authentic broadcast material from synthetic or manipulated media.
    • Disclosure labels for AI-generated presenters, voices, and imagery are becoming a legal requirement in some jurisdictions, including the European Union from August 2026.

References

Retrieved from "https://primo.ai/w/index.php?title=AI_Generated_Broadcast_Content&oldid=53400"