No AI Adoption
News or media firms with no official AI adoption announcement during the sample period.
- Baseline group for DID
- Candidate matched controls
- Useful for placebo event dates
AI Adoption Types in News Organizations and Stock Market Reactions
Comparing AI-assisted journalism, AI-written journalism, and non-adoption through event study and multi-valued DID designs.
3
AI Adoption Types
No adoption, AI-assisted adoption, and AI-writing adoption as a multi-valued treatment.
5
Event Windows
CAR[0,+1], [-1,+1], [-3,+3], [-5,+5], and [+2,+10].
3
Core Hypotheses
Adoption type and stock market reaction are tested directly; credibility, machine heuristic, and signaling effect remain background theory.
4
Theory Blocks
Source credibility, machine heuristic, and signaling effect.
1. AI Adoption Types
Main treatment variable: AI_Type = 0, 1, 2
AI-Assisted Adoption
AI supports search, summarization, translation, data organization, fact-checking, editing, or drafting.
- AI as a tool
- Human-AI complementarity
- Efficiency and productivity signal
AI-Writing Adoption
AI is presented as a direct producer or presenter of news content, such as automated articles or AI anchors.
- AI as a communicator
- Change in perceived authorship
- Trust and accountability concerns
2. Background Theory
Source credibility, machine heuristic, and signaling effect explain why AI roles matter
AI Adoption TypeNo AI, AI-assisted, or AI-written journalism.
Source AttributionReaders infer whether the message comes from human editorial judgment or an automated system.
Credibility EvaluationExpertise, trustworthiness, transparency, and accountability shape perceived news quality.
Audience SignalGoogle Trends-based public interest becomes the observable audience-attention proxy.
Market ReactionInvestors translate credibility and audience response into AR, CAR, volume, and volatility.
Source Credibility
Expertise and Trustworthiness
News credibility depends on whether the source is viewed as competent, accountable, and trustworthy. AI-writing adoption can weaken source credibility if readers perceive the article source as less transparent or less responsible than a human journalist.
Machine Heuristic
Automation as a Mental Shortcut
When audiences see a machine as the content producer, they may use a machine heuristic: AI can be perceived as efficient and data-driven, but also cold, less contextual, and less morally accountable.
Signaling Effect
Investor Interpretation
AI adoption announcements act as market signals about future efficiency, cost structure, innovation capability, and reputational risk. AI-assisted adoption mainly sends an efficiency and complementarity signal, while AI-writing adoption sends a mixed signal: cost reduction potential may be offset by credibility risk and possible reader backlash.
Audience Attention
Public-Interest Proxy
Google Trends spikes are treated as public-interest signals, not sentiment. The measure captures whether AI adoption announcements attract unusual audience attention around the event window.
Human-AI Complementarity
Role Boundary Matters
The market reaction should be stronger when AI is framed as supporting journalists and weaker when AI is framed as replacing editorial judgment or news authorship.
3. Hypothesis Tracker
Expected coefficient signs and comparisons
H1Overall effect
Adoption types produce different reactions
Stock market reactions differ across no AI adoption, AI-assisted adoption, and AI-writing adoption.
H2Beta Assist > 0
AI-assisted adoption is positive
AI-assisted adoption generates a more positive market reaction than no AI adoption.
H3Base < Beta Writing < Beta Assist
AI writing is weaker than AI assistance
AI-writing adoption generates a more positive reaction than no AI adoption, but a weaker reaction than AI-assisted adoption.
4. Quasi-Experimental Design
Real firm announcements instead of scenario experiments
| Analysis | Unit | Main Comparison | Main Output |
|---|---|---|---|
| Event Study | AI adoption announcements | AI-assisted adoption vs. AI-writing adoption | AR, CAR[0,+1], CAR[-1,+1], CAR[-3,+3] |
| Multi-valued DID | Firm-day panel | AI-assisted, AI-writing, and no AI firms | Assist x Post, Writing x Post |
| Matched DID | Matched firm-day panel | Comparable treated and control firms | PSM |
| Audience Evidence | Event-platform panel | Reader reactions linked to CAR | Google Trends interest proxy |
DID Specification
ARit = beta1(AI_Assisti x Postit) + beta2(AI_Writingi x Postit) + beta3AI_Assisti + beta4AI_Writingi + beta5Postit + gamma Xit + mui + lambdat + epsilonit
AI_Assisti and AI_Writingi are compared against no AI adoption. Postit captures the period after the announcement, and the main-effect terms separate baseline adoption-type differences from post-announcement treatment effects. Firm and date fixed effects absorb stable firm characteristics and common market shocks.
5. Model-Free Evidence
Raw patterns to show before regression results
Expected CAR Pattern
Replace the conceptual values with estimated means and medians after data collection.
This is a dashboard sketch, not an empirical estimate.
Evidence to Report
| Table M1 | Mean and median CAR by AI adoption type |
| Figure M1 | Average AR and CAR around the event date |
| Figure M2 | Google Trends interest index around the AI adoption event date |
| Table M2 | Mean GoogleTrendsInterest by AI adoption type and event window |
6. Data and Variables
Event, financial, text, and audience data joined into a panel
1. AI EventsPress releases, investor relations material, annual reports, and first major news coverage.
2. Event ScreeningExclude earnings, M&A, CEO turnover, lawsuits, and other confounding events.
3. Type CodingNo AI, AI-assisted adoption, and AI-writing adoption.
4. Market DataDaily stock returns, market returns, trading volume, and volatility.
5. Audience MeasuresGoogle Trends as a public-interest proxy.
6. Theory MeasuresSource credibility, perceived accountability, machine heuristic, and signaling effect.
Dependent Variables
AR, CAR[0,+1], CAR[-1,+1], CAR[-3,+3], abnormal trading volume, and abnormal volatility.
Key Mechanism Variables
SourceCredibility, MachineHeuristic, SignalingEffect, GoogleTrendsInterest, AI_Assist, and AI_Writing.
7. Endogeneity and Robustness
Threats to identification and planned responses
| Threat | Problem | Response |
|---|---|---|
| Selection bias | AI adopters may differ from non-adopters before treatment. | PSM-based matched DID. |
| Confounding events | Other major firm events may occur near the AI announcement. | News screening in [-3,+3], exclusion rules, and event dummies. |
| Theory measurement | Source credibility and machine heuristic are latent constructs. | Use multiple text indicators and platform-level validation. |
| Reverse causality | Distressed firms may adopt AI as a response to poor performance. | Prior return controls, pre-trends, lead terms, and distressed-firm exclusions. |
| Parallel trends | DID requires comparable pre-treatment trajectories. | Event-time DID and pre-treatment coefficient tests. |
| Measurement error | AI type and theory-mechanism coding may be noisy. | Multiple coders, Cohen's kappa, dictionaries, and supervised coding. |