(AI) Content is King?

Still largely unknown to the average user outside China, Jinri Toutiao (Chinese for “today’s headlines”) is as unicorn as unicorns can get. Launched just 6 years ago by Zhang Yiming, a Chinese software engineer and serial entrepreneur, Toutiao is most recently valued at $75B, boasting 120M daily active users who spend an average of 74 minutes on the site (2x those of Snapchat and Instagram) and projected revenue between $4.5-7.6B in 2018 [1, 2, 3]. It is also one of the few independent multi-billion dollar startups that have resisted investment from China’s 3 Internet behemoths, Alibaba, Tencent and Baidu [4]. The numbers are surely impressive – but what exactly does Toutiao do, what is its “secret sauce”, and is the “secret sauce” scalable beyond China?


At first glance, Toutiao appears to be a media company whose chief product is a news aggregator that serves personalized content to its users. However, the company views itself first and foremost as an artificial intelligence company that is applying its technology to tackle the entire media consumption lifecycle, from content creation, curation, recommendation to interaction. [5] Liu Zhen, Toutiao’s Senior Vice President reaffirmed in an interview that the way Toutiao approach their content platform that is different from traditional media is to “use artificial intelligence (AI) to connect people and information”. [6]


Toutiao’s effective use of AI, and more specifically various machine learning methods including deep learning, has been the “secret sauce” and growth engine behind its meteoric growth. Compared to Toutiao, traditional news aggregators fall behind due to two major reasons: i) heavy reliance on slow human editors who are constrained by the speed at which they can digest and curate new content, and ii) inability to provide continuous, personalized content to simultaneously hundreds of millions of users in absence of a scalable, algorithmic engine. Thanks to its AI technology, Toutiao can serve more relevant content at a much lower cost (vs. comparable hiring of human editors to achieve similar level of personalization), giving it an edge in an industry where users’ attention, and ultimately value, is captured by content quality, relevance and quantity. [5] Toutiao’s focus on AI also strengthens its own competitive advantage over other news aggregators over time due to the data network effect. As users spend more time reading on the app, the app becomes smarter as it learns more about the users’ behaviors and interests, fine-tunes its recommendation algorithm and in return serves better content. Users then become more engaged and feed more data back into the system, creating a virtuous self-improving cycle for personalization effectiveness. [7]


While Toutiao primarily focused on the recommendation engine in the early days, more recent focus has been on leveraging AI for actual content creation and curation / policing, especially in light of China’s political climate. [8] At the core, Toutiao’s business is a relatively straightforward application of recommendation engines to content distribution – the actual technology and ad-driven business model are not new. In the long run, Toutiao can only stay relevant if it can crack the problem of how to generate good, relevant content in a scalable, distributed manner.


For content creation, Toutiao debuted its bot reporter Xiaoming during the 2016 Summer Olympics in Rio de Janeiro, pushing out 450 stories between 500-1000 words summarizing the game results within minutes after the final whistle. Xiaoming bot has been generally a success – its stories produce similar read rates (# of reads divided by # of impressions) compared to those written by human writers.  For curation, Toutiao’s algorithm is also getting trained on spotting “fake news”, initially based on labeled data from user reports, and now also on fake news datasets generated by its own separate algorithm trained to write fake news. [5, 9]

(Xiaomingbot during Summer Olympics)

It remains to be seen whether reliance on technology leadership alone is enough in an industry as content-driven as the media. To complement their current AI content creation effort, Toutiao should also look into partnering with high-quality content creators on a variety of exclusive content, similar to the way Netflix has leveraged some of Hollywood’s most creative minds for their Netflix shows. In the long run, Toutiao might need to also build a platform allowing more creative voices and content to be sourced from its own user base, expanding the current stream of content inventory.  However, crowd-sourced content creation will have to be dealt with caution especially in the context of China’s news censorship. [8]


Moving forward, even bigger questions remain: How can Toutiao further develops its value proposition regarding personalization while also not creating a content chamber for its own readers? As the Chinese media demand saturates, can Toutiao’s AI technology that has been developed specifically for China scale and replicate similar successes in the rest of the world?

(790 words)


[1] The Economist, “Toutiao, a Chinese news app that’s making headlines,” The Economist, Nov 18, 2017, [https://www.economist.com/business/2017/11/18/toutiao-a-chinese-news-app-thats-making-headlines], accessed November 2018.

[2] Yunan Zhang, “Softbank in talks to invest in Bytedance,” The Information, Sept 27, 2018, [https://www.theinformation.com/articles/softbank-in-talks-to-invest-in-bytedance], accessed November 2018.

[3] Yue Wang, “Bytedance said to be raising funds at $70-75 Billion Valuation,” Forbes, Aug 15, 2018, [https://www.forbes.com/sites/ywang/2018/08/15/bytedance-said-to-be-raising-funds-at-70-75-billion-valuation/#ca94f02824cf], accessed November 2018.

[4] Juro Osawa, Yunan Zhang, “How a little-known startup became a tech behemoth,” The Information, Mar 12, 2018, [https://www.theinformation.com/articles/how-a-little-known-startup-became-a-tech-behemoth], accessed November 2018.

[5] Anu Hariharan, “The hidden forces behind Toutiao,” YCombinator Blog, Oct 12, 2017, [https://blog.ycombinator.com/the-hidden-forces-behind-toutiao-chinas-content-king/], accessed November 2018.

[6] Juro Osawa, “How a news startup caught China’s Tencent by surprise,” The Information, Jul 19, 2017, [https://www.theinformation.com/articles/how-a-news-startup-caught-chinas-tencent-by-surprise], accessed November 2018.

[7] Eva Xiao, “China’s most addictive news app Toutiao eyes world domination with AI feeds,” Tech In Asia, Dec 5, 2017, [https://www.techinasia.com/bytedance-overseas-expansion-strategy-break-down], accessed November 2018.

[8] Raymond Zhong, “A saucy app knows China’s taste in news. The censors are worried,” The New York Times, Jan 1, 2018, [https://www.nytimes.com/2018/01/02/business/china-toutiao-censorship.html], accessed November 2018.

[9] Kaifu Lee, AI Superpowers: China, Silicon Valley, and the new world order (New York, NY: Houghton Mifflin Harcourt, 2018), p. 108-109.


Smarter e-shopping with The North Face and Watson


GE Digital Windfarm – How Machine Learning is Helping GE Renewables Drive Process Improvement

5 thoughts on “(AI) Content is King?

  1. Hi there,

    I enjoy reading this piece, it is well written, and here are some points to consider:

    1) Toutiao’s AI-powered newsfeed creates the value for the company because it allowed higher advertisement fee charge. For example, businesses need to pay 4 times more to place their ads on Toutiao than on Baidu.com. Toutiao has the pricing power because it promises a higher click conversion rate as it is tailored to each user of its app. Every click, the amount of time spent on each news article, whether the user is using an iphone or a xiaomi phone, are inputs that help create the user portraits that helps more accurate advertisement targeting.

    2) I think Toutiao’s strategy of globalization is not to replicate the same model as they have in China and copy it to the rest of the world, they have been active in acquiring international companies like musicly, which are more acceptable and accessible to the U.S. users

  2. This was fascinating to me for two reasons:
    1. I really struggle in keeping up with news and have tried a bunch of different things this year (ex. TheSkimm, NYTimes app, etc) but nothing has really “stuck.”
    2. I’m interested in the future of content creation, news, and journalism in general, particularly since only 16% of US readers are willing to pay for news. (http://www.niemanlab.org/2018/11/so-some-people-will-pay-for-a-subscription-to-a-news-site-how-about-two-three/)

    Rather than answering your questions I find myself pondering two of my own:
    1. How well-written are the articles being produced by the reporter bot? Is it actually indistinguishable from a human writer, or is it clear that it’s being written by AI? In the English language today I don’t think I’ve seen a comparable product (or at least nothing beyond straight-up retweeting/paraphrasing from other news engines), but having robot “reporters” that can mimic the writing of real humans is very compelling. (And would make for a good sci-fi story…)
    2. I like the idea of them potentially partnering with thought-leader content creators, but that would mean essentially poaching them from whatever publications they currently write for (unless they’re self-employed). If they end up writing for Toutiao, I wonder if that would compromise their integrity in any way?

    Overall, I think it would be difficult to replicate this business outside of China given how different and complicated journalism is in every country. But I am curious to see what Toutiao does next!

  3. Stacy above notes that click ad rates are higher for Toutiao articles than for Baidu search results. But I wonder how successful continued monetization of Toutiao will be. The company may face a reckoning with its own content providers. These content producers and news organizations face a devil’s bargain with Toutiao. The more they allow the company to use their content, the less relevant their own homepages and the more personalized and sticky Toutiao will be. Sooner rather than later, they must recognize that serving content to Toutiao is only speeding their own demise. You draw an analogy to Netflix, and as we’ve seen, Disney has decided to pull its own content from Netflix to create its own distribution service and take control of its own destiny. Toutiao must successfully navigate this narrow path to appease its content producers while buying enough time to create its own content.

  4. Tying together the news content recommendation and the tailored advertisement, both by using AI is brilliant and the fact that they were able to resist the funding of Baidu, Tencent and Alibaba is an amazing feat on its own. All three giants must have tried to copy the AI solution on their own but must have failed and it shows that Toutiao indeed has something unique in terms of technology.

    Expanding their business into creating news contents by having the AI to write article is amazing too. It may not be able to write an analytical piece, but AI is better suited to write short, quick updates based on facts of the event. It would deeply impact journalism as most journalists will not be able to compete against AI in breaking the news.

    For global exansion, Toutiao could leverage on AI to teach itself the different languages. AI is truly an exciting field and it is really beginning to impact many industries, the way we work, and even our daily lives.

  5. This is really an interesting article. The question is how valuable this tool could be in another market, and which social implications it would have. I believe that the traditional press generates insights not only by paraphrasing knowledge which already exists, but by pulling facts and conclusions together. Going forward, this responsibility will be with online media, as newspapers continue to disappear. So I am wondering which consequences it would have if there are no journalists anymore who could make conclusions, but a machine learning tool which just put facts together in a new way. Would this still be seen as “free press”? It nearly seems as if we limit our opportunities to explore new relations in the world. This is actually a question that could be raised for all AI tools — they are fed with existing knowledge but lack creativity. Could this limit our ability to innovate in the long term ?

Leave a comment