From short videos to long articles: When Douyin entrusts news to AI

Author | Xiao Sijia

Editor | Qiao Qian

Ten years after Douyin detonated the short - video industry, long - form writing, once regarded as inefficient and outdated, has regained attention.

At the end of 2025, Douyin launched a long - form text and image function, opening an entry for in - depth long - form writing creation to amateur creators and media organizations, and providing traffic support for high - quality long - form texts and images. Currently, users can only upload and publish articles through the Douyin web version.

As an entertainment app that emerged from the "short, flat, and fast" and "three - second attention span" culture, Douyin's move seems to indicate that it is trying to break away from the "traffic - first" logic and seek more long - term value accumulation. This transformation can also be seen in the launch of "Douyin's Selected Content" in September 2024.

While introducing more original and high - quality content, Douyin is also actively leveraging AI search technology to integrate information across the web and focus on consulting content. Users can click on "AI - selected Consultations" in the "Hot Topics" section at the top of the Douyin app to browse real - time hot news summaries and content overviews captured and generated by AI. Behind this design, Douyin clearly hopes to use AI to more efficiently carry and distribute more long - form content.

In the past two years, ByteDance has introduced AI technology into almost all content forms: Qishui Music involves AI in the music creation process, Hongguo Short Dramas uses AI to generate comic drama content, and Fanqie Novel incorporates AI into online writing. From news, music to novels and dramas, with ByteDance's continuous investment and promotion, AI is being systematically used in content production and distribution.

01 From "Short" to "Long"

In September 2025, Douyin conducted a small - scale internal test of the long - form writing function, supporting high - quality text and image creation activities with a minimum of 300 words and a maximum of 4000 words. After the function was officially launched, the word limit was further extended to 8000 words. On the home page, these long - form articles appear in the form of pictures plus scrolling text, accompanied by soft music. Users can click on "Read the Full Article" below the text to enter the complete content page.

Currently, many media organizations and individual creators have started to try this function, and the published content mostly focuses on in - depth stories, social news, and personal growth, which are relatively long - cycle and non - immediate narrative directions.

Douyin was not the only platform that wanted to expand its content boundaries last year. In July 2025, Xiaohongshu launched an innovative "long - form text to image" mode. The platform provides one - click AI typesetting, generates summaries and covers, and finally automatically splits the thousand - word content into right - swipable images for publication, cleverly adapting in - depth content to users' inherent swiping habits for text and image notes.

The successive emphasis on long - form text and images by Douyin and Xiaohongshu is essentially a competition for high - quality creators and users' stay time. The former chooses to expand horizontally by inserting long - form text and image content with a completely different style into the short - video stream; the latter digs deeper vertically on the existing text and image form by using AI tools to lower the content creation threshold. At a deeper level, this is also an inevitable choice for content platforms after entering the era of stock competition.

In fact, as an ancient content carrier, long - form writing always has its value.

The research report "The Attention Game" published by the integrated marketing agency Mavericks in 2024 dispelled a common misunderstanding: although short - form content is widely popular for its convenience, it is far inferior to long - form content in building trust. The report also pointed out that 34% of respondents believe that long - form writing is more credible and attractive than any other form. Despite being easy to browse, short - video or short - form text content ranks lowest in all evaluation dimensions. Only 10 - 12% of respondents approve of it in terms of factual accuracy, expression quality, and content relevance.

On the other hand, it is obvious that Douyin's move is also aimed at competing with the complete ecosystem built by WeChat. According to QuestMobile data, as of October 2025, Douyin had a monthly active user base of 907 million, while WeChat's monthly active user base exceeded 1.4 billion. There is still a gap in user volume. Currently, Douyin's monthly active user count ranks second in the entire network, only after WeChat.

As the ceiling of traffic growth gradually looms, Douyin may have realized that short - videos alone are no longer sufficient to support its next - stage expansion. It must look for new content forms outside the existing system to fill the long - missing piece of the puzzle.

In the social field dominated by WeChat, Douyin has obviously never given up on its layout. In March 2024, Douyin launched the "Co - raise an Elf" function: as long as two friends send messages to each other continuously for three days, they can light up the "Spark" logo; after six consecutive days of message - sending, they can invite each other to co - raise a "Little Fireman". As the number of interaction days increases, the spark value and the Little Fireman's level increase simultaneously; once there is no interaction for more than 24 hours, the spark and the Little Fireman will turn gray, and if the interaction is not restored within the specified time, they will even disappear completely.

This design quickly caught on among young users. To this day, raising a Little Fireman has become a trendy and ritual - filled social activity, giving rise to internet catchphrases like "You can argue, but the spark can't go out". By quantifying social relationships and tying them to the cost of loss, Douyin has significantly increased the frequency of interaction and emotional investment among users. This mechanism for enhancing social stickiness has indeed attracted a large number of young users, to some extent prompting them to shift the focus of their daily chats from WeChat to Douyin.

After successfully "retaining" social relationships, Douyin's next step is to try to accommodate deeper content expression. "Users who read news don't like watching videos; they prefer reading articles," explained an insider at ByteDance. To some extent, this move is equivalent to embedding Toutiao inside Douyin, offering more content forms to attract potential users to stay.

Another reality is that the scale of Toutiao has stopped growing. After long - term diversion by short - videos, the upper limit of pure news - based products has become increasingly clear. Instead of continuing to invest in a product with a visible ceiling, it is better to put long - form content into Douyin and redistribute users' attention with a larger traffic pool.

Just as WeChat has gradually built up its official accounts and video channels on the basis of social relationships, Douyin is also trying to expand its product boundaries with its powerful algorithm capabilities. However, different from WeChat, which naturally supports reading scenarios, in Douyin's highly entertainment - oriented ecosystem, it remains to be seen whether these knowledge - and depth - oriented long - form contents can truly capture users' minds.

02 Behind AI News and Information

In addition to relying on Douyin's traffic, ByteDance is also continuously using AI technology to empower traditional businesses.

Take Qishui Music as an example. In addition to using Douyin for traffic and exploring innovative commercialization models, the platform has also actively built an "AI + Music" creation ecosystem. In November 2025, Qishui Music launched the "Qishui AI Music Creation Laboratory", which helps creators simplify the processes of lyric writing, composition, recording, and arrangement through AI tools, effectively lowering the threshold for music creation and enabling batch content production. By January 2026, Qishui Music's monthly active users were close to 140 million, ranking fourth in the industry; its daily active users ranked third, only after Kugou Music and QQ Music, and even surpassing NetEase Cloud Music.

In the field of long - form content, ByteDance has also continued the "AI +" approach. Now, in Douyin's hot news section, AI can capture the latest news hotspots across the web around the clock, integrate video and text information, and generate intelligent summaries.

These news items usually come from mainstream and authoritative media. On the basis of efficient capturing, AI further sorts and summarizes them. In addition, users can click the "Continue to Ask" button at the bottom of the article to ask questions and seek interpretations from AI regarding the information content.

An insider at ByteDance believes that compared with traditional news articles, the advantage of AI - based consultations is that "users can either read the article or click to view related videos, making the content consumption path more complete." He also revealed to 36Kr that this kind of news and information "created" by AI will soon enter Douyin's news feed. In other words, these AI - generated news items will compete with original long - form content in the same traffic pool.

However, beyond the reading experience, a more fundamental but insufficiently discussed question is emerging: Can news and information be easily captured, disassembled, and redistributed? When AI starts to participate in content production, especially news content production, where should its boundaries lie?

In fact, in current practice, since AI cannot obtain first - hand on - site information, its way of handling news is only a "secondary integration" rather than real content production. However, this kind of integration does not rely on real - world fact - checking but on algorithmic relevance.

When this technology is applied in the news field, responsibility becomes blurred. Previously, information editing and fact - checking were the professional responsibilities of media organizations. But in the AI era, even if there are judgment errors, AI will not be held accountable, and platforms can often attribute it to technical issues or information references. However, the public's trust in news may be eroded more quickly in this process.

In other words, this content structure with high information density but low verification itself creates new cognitive risks. In February 2025, 14 mainstream media organizations, including Forbes, Condé Nast, and the Los Angeles Times, sued a company named Cohere, accusing it of copying articles from their websites in bulk without authorization for model training and generating news summaries, and that the process is prone to breeding and magnifying the risk of "hallucinations", which damages the reputation of the media or publishers.

Meanwhile, the value of information itself is constantly depreciating. In May 2025, the Graphite research team found that more than half of the content on the Internet is now generated by AI. AI can infinitely reorganize and distribute existing information, rapidly increasing information density, but it is difficult to improve information quality. Instead, it makes the communication environment noisy - people's attention is more consumed, and they need more judgment to avoid being flooded by false and junk information.

In the era of information explosion, people need to be guided and informed about which information is more worthy of attention and which noise can be ignored. Just as in an era of extreme material abundance, people can shop through any channel, but they still prefer to go to Costco to consume - it doesn't offer the most choices, but every product on the shelves has been carefully selected.