PaperlayerView exact revision on arXiv

Source-linked AI summary

AI-Generated Content in Cross-Domain Applications: Research Trends, Challenges and Propositions

Jianxin Li, Liang Qu, Taotao Cai, Zhixue Zhao, Nur Al Hasan Haldar, Aneesh Krishna, Xiangjie Kong, Flavio Romero Macau, Tanmoy Chakraborty, Aniket Deroy, Binshan Lin, Karen Blackmore, Nasimul Noman, Jingxian Cheng, Ningning Cui, Jianliang Xu

arXiv:2509.11151v1cs.AI

TL;DR

Research on AIGC’s cross-domain progress and challenges remains limited despite its widespread applications. This vision paper synthesizes technical foundations and societal impacts across domains; its conclusion identifies unresolved challenges and future research directions.

  • Problem

    Few studies systematically review AIGC’s latest progress and emerging challenges across domains, while detectors often assume human-written, single-modality inputs.

    Source: §3 · p. 3, §11 · p. 18

  • Method

    The vision paper synthesizes insights on AIGC’s technical foundations, content spread and use, and societal impacts across domains.

    Source: §11 · p. 18

  • Results

    The paper identifies technical challenges and research propositions spanning detection, societal impacts, and AIGC applications across domains.

    Source: §11 · p. 18

  • Takeaways & Limitations

    The synthesis identifies adaptive attacks, multimodal verification, and inconsistent disclosure as outstanding research challenges.

    Source: §3.2 · p. 4, §5.2 · p. 9

    Detection faces adaptive style-conversion attacks and complex multimodal verification; hallucinations require techniques distinct from those for deliberate disinformation.

    Source: §3.2 · p. 4

Abstract

from arXiv · show

Artificial Intelligence Generated Content (AIGC) has rapidly emerged with the capability to generate different forms of content, including text, images, videos, and other modalities, which can achieve a quality similar to content created by humans. As a result, AIGC is now widely applied across various domains such as digital marketing, education, and public health, and has shown promising results by enhancing content creation efficiency and improving information delivery. However, there are few studies that explore the latest progress and emerging challenges of AIGC across different domains. To bridge this gap, this paper brings together 16 scholars from multiple disciplines to provide a cross-domain perspective on the trends and challenges of AIGC. Specifically, the contributions of this paper are threefold: (1) It first provides a broader overview of AIGC, spanning the training techniques of Generative AI, detection methods, and both the spread and use of AI-generated content across digital platforms. (2) It then introduces the societal impacts of AIGC across diverse domains, along with a review of existing methods employed in these contexts. (3) Finally, it discusses the key technical challenges and presents research propositions to guide future work. Through these contributions, this vision paper seeks to offer readers a cross-domain perspective on AIGC, providing insights into its current research trends, ongoing challenges, and future directions.

1. Introduction

AIGC generates human-like content across modalities and is being adopted in many sectors. The introduction situates this growth in generative AI and outlines the article’s cross-domain roadmap.

1. Introduction

  • AIGC generates text, images, audio, and video with fluency and coherence approaching human ability.The introduction frames it as a transformative trend in artificial intelligence.Source: §1 · p. 1
  • Generative AI has developed from probabilistic and neural approaches through GANs and diffusion models to transformer-based architectures spanning modalities.The passage traces early models and later architectural advances.Source: §1 · pp. 1–2
  • LLMs use pre-training followed by fine-tuning or prompt conditioning, and increasing model scale is associated with capabilities such as in-context learning and summarization.The models are trained to predict the next token in a sequence.Source: §1 · p. 2
  • LLMs are used across marketing, healthcare, education, legal services, and software development, while raising concerns about trust, disinformation, intellectual property, and societal impact.Prompt-based learning supports downstream tasks with little labeled data.Source: §1 · p. 2
  • The article covers LLM training and use, AI-generated content and detection, platform spread, societal impacts, data sovereignty, and security risks.Sections 5 through 9 address public trust, marketing, public health, organizational behavior, and education.Source: §1 · p. 2

2. Overview of Generative AI Framework

The overview presents generative AI as a modular pipeline: models are pretrained, optionally fine-tuned, and then adapted to tasks through prompts. It also surveys training strategies and prompt-learning approaches.

2.1. Framework Overview of Generative AI

  • Generative AI training and use comprise pre-training, optional task-specific fine-tuning, and prompt-guided downstream use without changing model parameters.The framework supports zero-shot and few-shot adaptation.Source: §2.1 · p. 2

2.2. Generative AI Training Strategies

  • Pre-training builds general language capabilities from large corpora; fine-tuning can adapt models to tasks or human preferences, including through RLHF.The passage notes that post-hoc alignment matters because large-scale training does not prevent outputs misaligned with user intent.Source: §2.2 · p. 2

2.3. Generative AI Pre-training Strategies

  • Pre-training uses large-scale text and objectives such as next-token or masked-token prediction, with performance tied to data scale, diversity, quality, and architecture.Some architectures combine objectives to support versatility and robustness.Source: §2.3 · p. 2
  • Transformer pre-training is supported by optimization, memory, curriculum, and distributed-computing techniques for efficient large-model training.The resulting general-purpose models support downstream tasks through fine-tuning or prompting.Source: §2.3 · p. 3

2.4. Prompt-learning for Generative AI

  • Prompt-learning ranges from instruction design to in-context examples and chain-of-thought prompts that add intermediate reasoning steps.The passage describes chain-of-thought prompting as improving performance on complex reasoning tasks.Source: §2.4 · p. 3

3. AI-generated Content and Detection

AI-generated disinformation challenges detectors designed for human-written, often text-only inputs. The section reviews fragmented detection efforts, describes adversarial and multimodal difficulties, and proposes more robust, interpretable approaches.

3. AI-generated Content and Detection

  • False information spreads up to six times faster than truthful news, while detectors face AI-generated content across text, images, and video.Traditional detectors often focus on text and assume human-written inputs, leaving detection questions about realistic articles, paired media, and timely intervention.Source: §3 · p. 3

3.1. Overview of Existing Studies

  • Earlier fake-news methods addressed linguistic patterns, propagation, and credibility within specific scopes, but no single detector integrates text, image, and video for AI-generated content and adaptive attacks.Recent specialized approaches address separate parts of the problem, leaving a fragmented landscape.Source: §3.1 · pp. 3–4

3.2. Research Challenges

  • Style-conversion attacks can mimic trusted outlets to fool style-based detectors, while multimodal verification must assess each modality and cross-modal semantic consistency.The section also distinguishes deliberate disinformation from unintentional hallucinations, which require different detection techniques.Source: §3.2 · p. 4

3.3. Research Propositions

  • Proposed directions include using LLMs as advisors or tool-using agents and structuring their reasoning for auditable, explainable verification.The section describes agents that retrieve external evidence and interpretable systems that decompose fact-checking or aggregate logical answers.Source: §3.3 · pp. 4–5

3.4. Concluding statement

  • The conclusion calls for hybrid, interpretable, multimodal detectors that resist stylistic manipulation and check cross-modal consistency.It presents these as integrated systems combining LLM capabilities with specialized style-agnostic content analysis.Source: §3.4 · p. 5

4. Spread and Use of AI-generated Information

AIGC uses existing digital distribution systems to spread at greater scale and speed, while its movement across platforms complicates detection, provenance, and intervention. The paper proposes research on measuring and modeling spread and strengthening provenance, detection, and governance across sectors.

4.1. Overview of Existing Studies

  • AIGC travels through social feeds, search, recommendation, and syndication systems; low production costs and easy repurposing support faster, broader dissemination.Empirical studies report rapid viral reach, while marketing and journalism integrate generated material into high-volume distribution pipelines.Source: §4.1 · p. 5

4.2. Research Challenges

  • Algorithmic amplification and cross-platform migration accelerate reach, while fluent synthetic fabrications can engage audiences before moderation or fact-checking.The challenges also include provenance loss, detector brittleness, transformed deepfakes, watermark trade-offs, hallucination cascades, unequal language coverage, and scarce end-to-end evaluations.Source: §4.2 · pp. 5–6
  • Equitable safeguards remain difficult: detectors can misclassify non-native writing, low-resource settings are underserved, and organizations lack validated cross-platform auditing protocols.The cited challenges also include scarce evaluations linking provenance, detection, intervention timing, and audience impact.Source: §4.2 · p. 6

4.3. Research Propositions

  • The propositions call for measuring spread velocity, modeling platform-specific patterns, preserving provenance through transformations, integrating multimodal detection into dissemination, and mapping sectoral spread signatures.The proposed analyses and safeguards are intended to inform detection and intervention timing, tailored strategies, and sector-specific policies and metrics.Source: §4.3 · p. 6

4.4. Concluding statement

  • The conclusion pairs technical interventions, including provenance-preserving formats and cross-platform detection, with governance tailored to sector-specific spread dynamics.The same infrastructures that enable personalized marketing, timely news summaries, and creative production can also amplify misinformation, provenance loss, and manipulative content.Source: §4.4 · pp. 6–7

5. Impact of AI-generated content to public trust

AIGC challenges established cues of authorship and credibility, with trust effects varying by context and domain. The reviewed work describes both trust gains and risks, including disclosure dilemmas in news, persistent political misinformation beliefs, and concerns about health and educational uses.

5. Impact of AI-generated content to public trust

  • AIGC can destabilize provenance and authorship cues, while accurate, transparent, relevant information may augment trust and inaccuracies or obscured provenance may degrade it.The paper frames trust as consequential across information providers and the public, and calls for provenance, bias mitigation, and context-sensitive disclosure in system design.Source: §5 · p. 7

5.1. Overview of Existing Studies

  • Trust responses vary with domain sensitivity, AI familiarity, task, and disclosure; entertainment may invite acceptance while news, politics, and healthcare may prompt scrutiny.The literature describes both improved efficiency, accessibility, and personalization and risks to credibility when provenance, accuracy, or intent is unclear.Source: §5.1 · p. 7

5.1.1. AI-Generated Content in News Media

  • Readers often rate AI-authored news comparable to human-authored news in clarity, grammar, and informational density, yet human-like output can be hard to identify without labeling.Disclosure may lower perceived credibility even for accurate content, while nondisclosure risks ethical and reputational harm; synthetic media exposure can erode trust in genuine journalism.Source: §5.1.1 · pp. 7–8

5.1.2. Political and Social Impacts

  • Studies report that targeted AI political narratives can exploit voter profiling, while beliefs adopted from AI-generated misinformation may persist despite later corrections.Awareness of AI fakery can also give political actors plausible deniability to dismiss authentic evidence as fabricated.Source: §5.1.2 · p. 8

5.1.3. Health Information and AI-Generated Content

  • Health communication trust depends on transparency, explainability, and consistency, although disclosure may reduce perceived credibility even when content is accurate.Explainable answers citing medical guidance score higher in credibility, while errors can damage trust and cause spillover skepticism.Source: §5.1.3 · p. 8

5.1.4. Education and Academic Integrity

  • Students may trust AI tutors for verifiable tasks but question nuanced feedback; academic integrity and detector fairness are additional concerns.The review describes clear rubrics, educator oversight, and communication about AI use, and reports that framing AI as an assistant with teachers retaining final authority may strengthen trust.Source: §5.1.4 · pp. 8–9

5.2. Research Challenges

The paper identifies unresolved trust challenges spanning disclosure, bias, authenticity, evaluation, and governance. These concerns include how to sustain confidence while deploying AIGC efficiently and fairly.

Transparency and Disclosure

  • Disclosure practices are inconsistent, and choosing how much and what kind of AI-use information to provide without undermining consumer confidence remains an open challenge.Some organizations label AI-assisted content while others obscure its origin.Source: §5.2 · p. 9

Bias and Fairness

  • Training-data biases create risks of discriminatory or culturally insensitive outputs, and identifying, measuring, and mitigating those biases remains unresolved.Source: §5.2 · p. 9

Authenticity versus Efficiency Trade-off

  • Scalable personalized production may come at the cost of perceived authenticity when automation is excessive, leaving a balance to investigate.Source: §5.2 · p. 9

Evaluation Metrics for Trust

  • Accuracy and BLEU scores do not capture sociopsychological aspects of trust, motivating multidimensional trust evaluation frameworks.Source: §5.2 · p. 9

Regulatory and Ethical Alignment

  • Emerging policies such as the EU AI Act and FTC guidelines leave limited clarity about operationalization, including embedding compliance without stifling innovation.Source: §5.2 · p. 9

5.3. Research Propositions

The paper proposes trust-centered system design, hybrid human–AI communication, broader trust measurement, cross-cultural research, and governance standards. It concludes that AIGC adoption depends not only on technical performance but also on addressing transparency, fairness, and authenticity.

5.3. Research Propositions

  • Proposed research includes trust-centered generation, human oversight in sensitive settings, interdisciplinary trust measurement, cross-cultural studies, and standards, audits, and certification.The proposed measurement methods combine behavioral experiments, longitudinal surveys, and computational indices for cognitive and affective trust dimensions.Source: §5.3 · p. 9

5.4. Concluding Statement

  • The conclusion argues that adoption will depend on addressing transparency, fairness, and authenticity alongside technical performance to protect public trust.It frames AIGC as a trust-sensitive technology and calls for integration that strengthens rather than erodes the social contract between people and digital systems.Source: §5.4 · p. 9

6. Impact of AI-generated Content to Digital Marketing

AI-generated content is being used in digital marketing to create diverse, tailored text and visual assets, with reported efficiency and performance gains. The section highlights unresolved challenges in multimodal and personalized generation, and proposes broader evaluation and culturally aware methods.

6.1. Overview of Existing Studies

  • Text and visual marketing systems generate ad copy, product titles, SEO content, image backgrounds, banners, try-on images, and promotional videos.Examples include preference-refined shopping titles, prompt-driven backgrounds, and text-to-image advertising frameworks reporting real-world CTR improvements.Source: §6.1 · p. 10, Figure 3 · p. 10, Figure 4 · p. 10

6.2. Research Challenges

  • Unified multimodal marketing generation remains largely unexplored because it requires aligned, controllable, high-quality outputs across modalities.Marketing videos must combine consistent images, text, and audio; scalable end-to-end frameworks remain at an early stage.Source: §6.2 · p. 10
  • Current personalization often remains one-size-fits-all, rarely using fine-grained interaction histories or cultural backgrounds to match users’ preferred styles.The section says this can limit alignment with preferences and lead to suboptimal performance.Source: §6.2 · pp. 10–11

6.3. Research Propositions

  • Existing marketing benchmarks such as CAMERA [132] and ADPARAPHRASE V2.0 [133] cover text, leaving broader content types and use cases underrepresented.The authors call for open, standardized datasets that support fair, consistent, and reproducible comparisons.Source: §6.3 · p. 11
  • Culture-aware generation should account for audience context while optimizing engagement, since elements effective in one market may carry negative connotations elsewhere.The proposed direction seeks to maintain metrics such as CTR while making content culturally appropriate.Source: §6.3 · p. 11

6.4. Concluding statement

  • The section concludes that text- and image-based marketing generation has advanced, while multimodal generation, user-specific preferences, open benchmarks, and cultural awareness remain underexplored.The conclusion also reports improved content-creation efficiency and results through data-driven methods.Source: §6.4 · p. 11

7. Impact of AI-generated Content to Public Health

AI-generated content may help provide accessible health information and preliminary guidance, but its use raises challenges around accuracy, equity, accessibility, and transparency. The section proposes research directions addressing these concerns and emphasizes trustworthy, user-friendly content.

7. Impact of AI-generated Content to Public Health

  • AI systems can summarize health information and offer preliminary guidance, potentially helping people facing high costs or limited access to qualified clinicians.The paper cautions that hallucinated or inconsistent advice can be harmful and calls for expert validation and cross-checking before large-scale deployment.Source: §7 · p. 11

7.1. Overview of Existing Studies

  • Prior work examines clinician use and AI-content detection, while public-health studies also raise questions about LLM authorship and chatbot outbreak guidance.Detection studies use TF-IDF features and random forests; reported clinician trust and utility are mixed.Source: §7.1 · p. 12

7.2. Research Challenges

  • Fluent but incorrect or unsupported advice challenges evidence fidelity, while training-data skews can leave rural, low-resource-language, and marginalized communities underserved.The paper notes that advice may assume healthcare infrastructure unavailable to some populations.Source: §7.2 · p. 12
  • Medical terminology and recommendations may be inaccessible or impractical when they exceed users’ health literacy or local resources.The paper gives examples including technical terminology, unavailable blood-pressure machines, and specialist referrals in regions without specialists.Source: §7.2 · p. 12
  • Medical advice also needs transparent sources, scope, limitations, and data protections, with disclaimers balanced against usefulness to avoid overreliance or disengagement.The section frames this balance as a role-clarity challenge.Source: §7.2 · p. 12

7.3. Research Propositions

  • Proposed directions include validated-source retrieval, provenance indicators, and medical-professional review to strengthen factual reliability and reduce hallucinations.The authors also propose transparent communication of sources, scope, limitations, and privacy protections while balancing disclaimers and trust.Source: §7.3 · pp. 12–13
  • The section proposes balanced data, fairness constraints, and socio-contextual inputs, alongside layperson-friendly language and resource-sensitive recommendations.These directions target more equitable, culturally relevant, and practically applicable medical information across diverse populations.Source: §7.3 · p. 13

7.4. Concluding statement

  • The conclusion says public-health AIGC’s potential depends on information that is factual, unbiased, transparent, accessible, and aligned with health literacy.It describes the field as rapidly evolving alongside new techniques, studies, and regulatory frameworks.Source: §7.4 · p. 13

8. Impact of AI-generated Content to Organizational Behaviour

AI is increasingly part of marketing work, shifting collaboration toward integrated human–AI workflows and raising questions about trust and organizational change. The section reviews these issues, formulates research propositions, and argues that realizing potential benefits requires cultural and structural alignment.

8.1. Overview of Existing Studies

  • Research describes human–AI marketing as a shift toward collaborative workflows, with interoperability, trust, mutual learning, and leadership shaping adoption.Studies also report persistent negative consumer perceptions despite disclosure, mixed career-perception effects, and psychological barriers to adoption.Source: §8 · p. 13, §8.1 · pp. 13–14

8.2. Research Challenges

  • Effective marketer–AI teams require clear task allocation and mutual learning, yet skill gaps, unclear roles, and resistance can hinder cross-functional collaboration.The section calls for interventions that strengthen AI fluency, trust, and productive human oversight.Source: §8.2 · p. 14
  • Consumer trust remains difficult to sustain despite disclosure, with psychological barriers and message appeal shaping receptivity to AI-mediated advertising.The authors identify communication strategies, ethical frameworks, and cross-cultural and longitudinal study as research needs.Source: §8.2 · p. 14
  • The effects of AI adoption on organizational design and employee career perceptions remain mixed, and transitions can threaten engagement and strategic alignment.The section calls for research on leadership, governance, and change management.Source: §8.2 · p. 14

8.3. Research Propositions

  • The propositions predict benefits from hybrid human–AI content creation, transparent rationales, hybrid marketer skills, and AI-literacy training.They include hypotheses about engagement, conversion, marketer acceptance, and confidence in producing high-quality, on-brand assets.Source: §8.3 · p. 14
  • Trust propositions predict stronger consumer trust from training-data provenance disclosure, human endorsement cues, and personalized disclosure formats.These are proposed hypotheses, not reported empirical findings.Source: §8.3 · p. 14
  • Organizational propositions predict higher innovation under dual human–AI leadership and a moderating role for clear AI decision boundaries in employee engagement.The conclusion frames AI-generated marketing as a strategic shift requiring cultural transformation, communication, leadership, and structural alignment.Source: §8.3 · p. 14, §8.4 · p. 14

9. Impact of AI-generated Content to Education

Marketing education is adapting to generative AI’s use as a tutor, teammate, and tool, while confronting gaps in AI literacy, assessment, and ethical preparation. The section proposes curriculum-wide integration, educator development, further research, and responsible-use frameworks.

9.1. Overview of Existing Studies

  • Studies frame generative AI as a tutor, teammate, and tool, supporting guidance, brainstorming, and tasks such as analysis, editing, and campaign creation.Tutoring guidance calls for instructor monitoring and student AI literacy; teamwork benefits from students preparing prompts and critiquing outputs.Source: §9.1 · p. 15

9.2. Research Challenges

  • AI-literacy gaps, academic-integrity concerns, possible erosion of foundational skills, and educators’ need for AI-pedagogy training are central challenges.Assessment practices also need rethinking, beyond checking whether student work is authentic.Source: §9.2 · pp. 15–16
  • Teaching AI responsibly requires students to address privacy, bias, and misinformation, critically evaluate outputs, and meet legal, regulatory, and ethical requirements.The section calls for responsible-use frameworks developed with academia, industry, and regulators and embedded in marketing programs.Source: §9.2 · p. 16, §9.3 · p. 16

9.3. Research Propositions

  • Marketing curricula should integrate AI-driven experiences throughout coursework and develop prompt engineering, AI-assisted analysis, artificial-influencer management, and AI-aided content creation.AI should not be treated as a standalone topic or course.Source: §9.3 · p. 16
  • Curriculum redesign requires pedagogical adaptation, dedicated resources, faculty upskilling, and targeted short courses or flexible online training for current marketing professionals.The section also calls for new conceptual frameworks to guide integration.Source: §9.3 · p. 16
  • Research should examine AI’s effects on tutoring, group dynamics, personal learning, and experiential learning, alongside adoption factors and barriers.The section also calls for continuous monitoring from both academic and student perspectives.Source: §9.3 · p. 16

9.4. Concluding statement

  • AI offers pedagogical and administrative opportunities, but raises challenges for teaching, assessment, privacy, and ethical and legal standards.Educators and institutions are urged to safeguard integrity and inclusivity while navigating these challenges.Source: §9.4 · p. 16

10. Data Sovereignty and Security Risks in AI-generated Content

AIGC’s reliance on broad data flows raises sovereignty, privacy, and security risks in online marketing. The section reviews mitigation approaches and proposes integrating privacy, transparency, and robustness into future systems.

10. Data Sovereignty and Security Risks in AI-generated Content

  • Online marketing’s use of customer data with generative models increases the relevance of breaches, prompt injection, model inversion, and personal-information leakage.AIGC’s data flows can also create conflicts between cross-region use and regulatory frameworks including GDPR and CCPA.Source: §10 · pp. 16–17

10.1. Overview of Existing Studies

  • Current responses include privacy-preserving technologies, adversarial defenses, and sovereignty enforcement, but each faces deployment trade-offs.Examples include quality loss, latency, multimodal scaling challenges, and tensions with decentralized workflows.Source: §10.1 · p. 17

10.2. Research Challenges

  • Data localization can conflict with global data integration for high-performance models, while federated learning remains challenged by regional statistical heterogeneity.The section identifies this as a sovereignty–functionality trade-off.Source: §10.2 · p. 17
  • Public LLMs and text-to-image models can memorize training data, enabling near-exact reconstruction through inversion attacks; differential privacy can reduce output quality.The section identifies memorization and reconstruction as a persistent challenge.Source: §10.2 · p. 17

10.3. Research Propositions

  • Future privacy-aware training could penalize sensitive-data memorization; selective feature suppression is proposed for personalized health campaigns.The section also proposes combining blockchain with generative AI for data validation, transparency, and verification.Source: §10.3 · p. 17
  • Robustness proposals include information bottlenecks and latent-space watermarks, with lightweight and standardized cross-modal defenses identified as future needs.The stated aim is to bridge generative capabilities and security demands.Source: §10.3 · pp. 17–18

10.4. Concluding statement

  • AIGC deployment must reconcile sovereignty compliance and attack resilience with model functionality, rather than treating either as a standalone safeguard.The conclusion calls for both as core architectural principles.Source: §10.4 · p. 18

11. Conclusion

The paper surveys AIGC’s technical foundations and societal impacts across domains, then identifies unresolved challenges and directions for future research.

11. Conclusion

  • The vision paper addresses the limited cross-domain review of AIGC’s latest progress and emerging challenges by bringing together insights from multiple disciplines.It aims to offer a cross-domain perspective on research trends, challenges, and future directions.Source: §11 · p. 18
  • Its scope spans training, detection, and the spread and use of AI-generated content, alongside societal impacts across diverse domains.The paper also examines data sovereignty and generative-AI security risks.Source: §11 · p. 18
  • The paper identifies multimodal generation, bias and fairness, trust, ethics, and context-aware approaches among its challenges and future research directions.Source: §11 · p. 18
Loading 2509.11151v1…