In-depth Reporting on WeChat Over the Past Decade

Tracking the Reconfiguration of Journalistic Depth under Platformization (2013–2025)

Ruomeng Liu
ruomeng.6@connect.hku.hk

Journalism and Media Studies Centre, The University of Hong Kong

June 2026

The puzzle

A decade of research documents how digital platforms disrupted Chinese journalism — business models, professional practice, the shrinking population of investigative reporters.

But almost all of it operates at the organizational or individual level.

The pathway from institutional change to textual change has been assumed, not measured.

We still know remarkably little about how the substance of the journalism itself — what actually appears on the page — has changed.

Why in-depth reporting?

In-depth reporting is the genre defined by going beyond surface facts: context, causation, complex social issues from multiple perspectives.

Why it matters in China

  • A primary channel for non-official analysis of social problems
  • Historically tied to the yulun jiandu (watchdog) tradition

Why it is vulnerable

  • The most resource-intensive form of journalism
  • Exposed to both platform economics (speed, virality) and political tightening

Yet scholarship subsumes it under investigative journalism — foregrounding exposés while neglecting the genre’s everyday output.

Research questions

We track the textual practice of the genre along three dimensions — focusing on output, not journalists’ stated beliefs.

RQ1 · Style — How has the rhetorical register changed? (pronouns, emotion, moral language)

RQ2 · Theme — How has thematic focus shifted over platformization?

RQ3 · Truth — Has the kind of “truth” the genre claims to reveal changed — exposing systemic dysfunction or bearing witness to individual experience?

Data

Corpus

  • 42,472 articles
  • 16 leading in-depth outlets
  • WeChat, 2013 → April 2025
  • Purposive sampling (GIJN + In-Depth Training Camp), cross-validated, screened on 4 criteria

Caveat: censored / deleted articles could not be scraped → findings are likely a conservative estimate.

Sample composition by year (top: articles; bottom: active outlets).

Methods: three lenses

Dimension Approach
Style Dictionary counting (jieba) — DUTIR emotion lexicon · C-MFD 2.0 moral foundations · pronouns
Theme BERTopic neural topic modeling + dynamic topic trajectories
Truth Sentence embeddings (semantic drift) + LLM classification of every “truth” sentence

A length confound runs through everything. Mean article length grew +52% (1,919 → 2,920 words). Density falls mechanically as articles lengthen — so we base inference on length-controlled estimates.

Finding 1 — Style

The vocabulary of condemnation softens; empathy grows.

Style: a selective recalibration

  • Second-person address ↓ 4.23 → 2.50 per 1,000 words (β = −0.18, p < .01)

  • Disgust language — criticism, contempt, moral condemnation — declines robustly even after length control (β = −0.22, p < .05)

  • Care (empathy) rises in per-word usage (β = +0.0015, p < .01); Sanctity keeps falling

  • The genre is not emotionally thinner — it shifts which registers it uses

Away from judgmental / hierarchical framing → toward empathetic language.

Finding 2 — Theme

From institutional accountability → toward everyday experience.

Theme: the center of gravity moves

29 topics, classified by trend + volatility.

  • Criminal Cases — steepest decline (accountability-organized subject)

  • Rising: Healthcare, Elderly Care, Labor, Fertility, Education, AI — individual & everyday experience

  • 8 rising topics show statistically significant field-wide convergence

Subjects organized around institutional scrutiny lose share; subjects of individual experience gain it.

Finding 3 — Truth

What the genre presents as “truth” has changed.

Truth: structural → individual

4,394 “truth” sentences → embeddings + LLM classification (κ = 0.68; cross-model validated).

  • Steady semantic drift, 2013 → 2025 (linear R² = .97, p < .001)

  • Among substantive claims, structural truth fell 78.9% → 63.9% (β = −0.025/yr, p = .002)

  • Individual truth rises; “truth” increasingly appears in formulaic / routinized usage

“Truth” is being evacuated of its accountability-oriented meaning.

Is this field-wide?

Convergence on the core trajectory

  • Sanctity declines in all 13 outlets; structural-truth decline holds under mixed-effects + leave-one-out → field-wide, not idiosyncratic

Divergence at the margins

  • Digital-native platforms ↑ first-person & disgust
  • Legacy outlets move the opposite way

→ Selective adaptation & reallocation, not uniform retreat.

Discussion

Three dimensions converge on one directional story:

Articles grew longer, the condemnatory register softened, thematic focus moved from institutions → everyday life, and fewer truth claims invoke structural causation.

The genre still performs “depth” — but what that depth refers to has changed.

We document what changed, not why. Consistent with several accounts — political tightening, platform economics, audience shift, demographics. The field-wide character makes any monocausal story unlikely.

Takeaway

Over a decade on WeChat, in-depth reporting grew longer, softened its judgmental register, and retreated from structural accountability as the basis of its truth claims.

The genre’s “depth” has been reconfigured — its center of gravity moved from institutional accountability toward individual experience.

Strategic adaptation that preserves the genre’s public-good function — or substantive depoliticization? The data can inform the question, but not yet resolve it.

Thank you

Ruomeng Liu

Journalism and Media Studies Centre · The University of Hong Kong

ruomeng.6@connect.hku.hk

Questions & discussion welcome.