Written by | Wang Pan, Wen Yehao

Edited by | Wu Xianzhi

The heyday of image-text information platforms seems to have quietly faded.

In the era dominated by image-text content, official accounts and information apps amassed a vast and unyielding user base. However, times have changed. Zhang Xiaolong's prophecy that "video expression will be the core content of the next decade" has now become a reality.

From the technology stack to product perspectives, information platforms that once served as user entry points have gradually become "trapped defenders of the portal era." Despite the industry's continuous efforts to update functions, fine-tune channels, and revise operations, these patches have never been able to reverse the tide of the times.

Fortunately, the variable that may truly determine the future direction of information platforms is gradually emerging—AI.

For the image-text information track, AI is not merely about simple algorithm iteration or product form renewal; it's about redefining the entire content consumption logic. Against this backdrop, information platforms have regained their footing.

However, this time, they must leverage a whole new set of capabilities to answer an age-old question: in an era where information is already overwhelming and the sense of information's meaning is gradually dissipating, how can they re-establish a reason for their existence?

What makes this "technological revolution" of information platforms different?

In fact, there is hardly any content track more accustomed to being repeatedly reshaped by technological cycles than information platforms.

Looking back at the development history of information platforms, from portals to apps, from manual editing to algorithm distribution, the information distribution model has undergone several generational shifts.

At the end of the last century, the "classified directory + portal" era gave birth to traditional portal websites like Tencent, Sina, and Sohu. Subsequently, search engines emerged, shifting the paradigm of information acquisition to the "search era." After that, under the Web 2.0 wave, social media and subscriptions rose, allowing users to obtain information directly from the source, bypassing portals. Until the 2010s, personalized recommendation algorithms emerged, and information streams became the mainstream, with platforms like Toutiao skyrocketing thanks to intelligent distribution.

It can be said that information platforms have always been at the epicenter of technological shuffles and are no strangers to them. In fact, "technological shuffle" is the most anticipated narrative theme in this track and the underlying variable that allows many platforms to rise against the trend.

For example, the evolution of information distribution models over time has been accompanied by the reshaping of commercial landscapes—new technologies give birth to new models, often "obliterating the previous generation of overlords" through more efficient content supply and commercial innovation.

However, this technological revolution is entirely different from the past.

In the past, every innovation was on the cusp of user growth and traffic dividends, so technological iterations often led to rapid expansion of the industry landscape and a dividend frenzy. To compete for the window period, platforms even invested heavily in subsidies and launched massive traffic wars.

Currently, the information industry has already entered an era of stock competition: incremental users have peaked, total usage time has been seized by strong competitors, and platforms are competing for limited user attention stocks. A QuestMobile report once pointed out that the information service industry has shifted from "traffic growth" to a new stage of "traffic operation."

In other words, today's information platforms no longer have the "easy wins" of the past but instead face the pressure of "rowing against the current, either advancing or retreating." Against this backdrop, even with AI driving it, it is almost impossible for them to replicate the "golden age" of savage growth from the past.

In other words, AI for information platforms is not a "miracle drug" that rejuvenates and soars the industry but more like a desperate fight for survival.

Therefore, this AI-driven transformation is far from the "minor repairs and patches" of previous years. It's not as simple as optimizing the recommendation model or creating a few fancy new features but involves a radical reshaping of the basic logic of information products—involving the triple reconstruction of core elements such as content production, distribution mechanisms, and user relationships.

Triple Reconstruction of Content, Distribution, and User Relationships

If in the past decade, the general structure of information platforms was human writing, machine promotion, and user viewing—with clear relationships and boundaries between the three. Then the true upheaval brought by AI is to dismantle and re-stitch this seemingly stable structure one by one, thereby better answering the three core questions of how content is produced, how it is distributed, and how it establishes a connection with users.

On the content production side, although AI has never replaced human content creators, it has deeply "intervened in formation."

In the past, manual creation was a stable production paradigm—authors wrote drafts, and editors adjusted and proofread them. But as AI gradually gains abilities such as generation, rewriting, abstraction, and image-text matching, content production gradually becomes a collaborative effort between humans and machines.

Taking "light content" as an example, the current industry practice is to generate a first draft based on specific viewpoints using a large model, which is then polished by humans, improving production efficiency and reducing error risks. For example, on platforms like Tencent News and The Paper, AI handles multiple auxiliary tasks such as title optimization, typo correction, image matching, and summary extraction.

The introduction of these capabilities has made it normal for humans and machines to collaborate on content production—machines provide density and efficiency, while humans retain perspective and judgment.

But the "reconstruction" of content is just the beginning. AI's deeper involvement occurs in the deconstruction of the distribution system.

Once upon a time, algorithm recommendations based on interest tags were considered the golden rule of content distribution—you click on it, you stay there, you read for how long, and the platform feeds you accordingly. Facts have proved that this mechanism has been unstoppable in the past decade, not only nurturing the golden age of image-text information but also enabling short video platforms to build deep interest pools.

At the same time, however, it has also buried the hidden danger that platforms are becoming increasingly "echo chambers"—users are exposed to homogeneous and confirmation bias information streams for a long time, gradually losing their ability to actively discover.

The intervention of AI has broken this traditional feeding structure: the path for users to "find content" has been greatly expanded—information acquisition has shifted from passive acceptance of recommendations in the past to two-way flow.

Even Douyin recently announced that its algorithm no longer relies solely on existing interests and historical behavior but introduces more dimensional information to present users with diverse content, technically alleviating the problem of information cocoons.

In the image-text information track, players like Tencent News are also exploring using AI to reshape the distribution system. For example, by using dynamic interest modeling to make recommendations escape static labels and perceive real-time changes in user preferences; content understanding has also evolved from shallow interest classification to deep portrayal based on value preferences.

This shift aims to break away from the algorithmic inertia of blindly catering to users and reopen the width and depth of the information world for them.

Additionally, information inherently has a companionship and tool attribute for users. Therefore, while AI strengthens the connection with users, it also maximizes the role of information platforms as "tool people."

Take NetEase News as an example; its AI comment function is quietly changing the way users interact. Imagine this scenario: after reading a report, click on an AI icon next to the comment box, and the system immediately generates three styles of comment drafts—rational, humorous, or sharp. Users can select and post with one click, making "posting a comment" no longer a burden but an expression experience triggered at any time.

Tencent News also brings AI to the comment section—AI first understands the article and each comment, then transforms into a high-EQ, energetic, and sunny "old friend" who responds to users with heartfelt words, adding a touch of human warmth to cold information.

At the "companionship" level, platforms such as Sina News and Tencent News are also embedding AI assistants into the daily interaction of information products. The former achieves full-process assistance through "Wisdom Xiaolang" for topic recommendations, intelligent summaries, instant Q&A, etc., while the latter provides functions such as word-selection questioning, AI summaries, and chatting while watching with "News Sister"—a framework for an "ubiquitous AI assistant" is taking shape.

Other functions are all aimed at being "fast, accurate, and usable"—amplifying the tool attribute of the platform itself. For example, in Tencent News live broadcasts, for common English original sound content in fields such as finance and technology, AI can generate real-time Chinese and English subtitles or translations, eliminating language barriers.

In fact, several years ago, the industry conceived the idea of returning news clients to their tool attributes. Within Tencent, an information assistant codenamed "DreamReader" was developed, integrating voice interaction and text reading functions, capable of automatically announcing the news briefing generated by "DreamWriter."

And in the era of AI, this idea is clearer. Information platforms do not have to compete fiercely for user duration like short videos but should strive to help users efficiently obtain valuable information—the real value may not lie in keeping users longer but in allowing them to leave with answers faster.

From "instant" to "precipitation," a rewrite of the value logic

In fact, information platforms have always had traffic but rarely truly owned "content assets." This is not because information platforms lack content; the real problem lies in the short life cycle of information content, which cannot form structural memory.

For example, image-text information platforms have massive news rotating like a kaleidoscope every day. If a piece of news is not fully consumed within 24 hours, it is almost destined to be "sunk" by the platform mechanism.

And a complex event spanning several months and involving multiple parties is often fragmented into dozens or even hundreds of isolated pushes, with timelines, causal relationships, key entities, and public opinion inflection points all scattered. Readers often only remember fragments and struggle to piece together a complete picture.

This production mode of nearly "one-time content" makes it difficult for information platforms to accumulate content assets—news is constantly being produced, and once it passes its timeliness, it is shelved, making it difficult to truly build content barriers.

The arrival of AI makes "memory" possible for the first time—in-depth integrated reports that originally required investigative journalists to repeatedly verify, organize, categorize, and write can now be realized to a certain extent by AI through cross-verification of multiple sources of information, becoming systematic and scalable.

Tencent News is an example. With the help of AI, after each hot news, it can generate a brief summary presenting key information and more accurately push related reports; below the news is the "Event Timeline" function—unfolding the timeline, character relationships, and core controversies, allowing users to sort out a complex news story in just a few minutes.

If users are still concerned about the subsequent development of the event, they only need to leave a keyword, and AI will customize an exclusive morning newspaper based on it: accurately screening from the rolling information stream, sorting out the most relevant information, and delivering it proactively every day. It is reported that the AI morning newspaper goes beyond news itself, capable of continuously tracking specific companies, industries, or even vertical topics, building a long-term and dynamic information entry for users.

On the other hand, in professional verticals, the semantic understanding and information integration capabilities of large models are enabling information platforms to organize originally fragmented information into a knowledge system that can be consulted and accumulated for the first time.

Taking Tencent News' "Finance Assistant" as an example, around a company, the system can automatically integrate its financial reports, executive statements, M&A dynamics, and regulatory responses, packaging them into a continuously updated database. Users no longer need to search everywhere; they can simply click on its assistant to get an overview of a company's performance trends or public opinion changes. Similarly, platforms like National Business Daily have also launched intelligent financial report analysis tools, moving financial information from "viewing" to "using."

When information is structured and "remembered" in this manner, it leaves an indelible mark of value on the platform. This ensures that it can continually influence both algorithmic recommendations and active user searches in the future.

AI serves as both a catalyst and a multiplier for unlocking content value. It efficiently consolidates fragmented information, imbuing it with enduring vitality—rendering the value of information content impervious to the ephemeral nature of timeliness. Through product applications, AI transforms high-quality data, guided by a premium content strategy, into long-term assets, consistently unlocking its multiplicative potential.

As the value of information content crystallizes, platforms can construct their own content assets and knowledge graphs, thereby redefining their value proposition. Image-text information transcends its role as a disposable consumer good, transforming into an asset that can be accumulated, reused, and continuously appreciated.

It's important to clarify that structuring does not imply the sacrifice of immediacy; rather, it adds an additional layer of value accumulation to information platforms.

In the long term, this represents a sustained value proposition and one of the few avenues for information platforms to regain prominence amidst the pressures posed by short videos.

Undoubtedly, this transformation has already commenced and is progressing at a rapid pace. These ever-evolving players are poised to achieve self-redemption within this new cycle, uncovering fresh possibilities for the information industry.